Data processing system

Abstract

The present application relates to a medical information integrated system which integrally collects, accumulates, multiple health-related information of each patient's medical record, biological data, genetic information, environmental factors, and comprehensively analyzes these data using generative AI, thereby highly accurately predicting the health risk of each individual, automatically generating and proposing the best preventive medical plan at the stage before the onset of symptoms. In addition, it also has the functions of real-time remote monitoring or alarm notification, automatic scheduling of remote diagnosis, while achieving the reduction of the burden of medical practitioners and the improvement of the QOL of patients, and the elimination of medical disparities.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to solve is that medical practitioners face enormous time and human costs in collecting and analyzing diverse and massive amounts of health-related data, including individual patient medical records, real-time biological data, genetic information, and environmental factors, to design and propose appropriate diagnostic or preventative medical plans. Furthermore, human judgment alone is insufficient for early and accurate prediction of each patient's risk, and for individually determining the optimal timing or content of intervention. This situation makes it difficult to consistently provide high-quality and personalized medical services to every patient, especially in cases of chronic diseases, the elderly, or those residing in remote areas with limited medical resources. This results in increased consultation frequency and the burden on medical practitioners, widening healthcare disparities, increased medical costs, and a decline in patients' quality of life. In addition, existing diagnostic support tools or simple remote monitoring systems cannot fully realize advanced functions such as learning long-term health patterns, predicting future risks, or automatically generating individually optimized preventative medical plans, making it difficult to grasp changes in patients' health status early and intervene medically at the appropriate time. Summary of the Invention

[0005] The means for solving the problem involved in this invention comprises the following structure: a patient medical record acquisition unit, a biological data acquisition unit, a genetic information acquisition unit, an environmental factor acquisition unit, a long-term memory management unit, a generative AI processing unit, a preventive medical plan proposal unit, and a remote monitoring and collaboration unit. The medical record acquisition unit acquires each patient's diagnostic history, treatment history, and past medical history from a database. The biological data acquisition unit acquires real-time biological information such as the patient's heart rate, blood pressure, blood sugar level, body temperature, and respiratory rate from wearable devices or medical machines. The genetic information acquisition unit acquires the patient's genetic markers or disease risk gene information. The environmental factor acquisition unit acquires environmental factor data such as the patient's lifestyle habits, smoking history, drinking history, exercise habits, living environment, and occupational environment. The long-term memory management unit accumulates and manages each patient's data from the past in a time-series manner. The generative AI processing unit comprehensively analyzes all current medical records, biological data, genetic information, and environmental factors. Through multivariate analysis and pattern recognition, it predicts each patient's health risk and detects increased risk before symptoms appear. Based on the analysis results of the generative AI processing unit, the preventive medical plan proposal unit generates optimal preventive medical plans for nutrition, exercise, sleep, lifestyle improvement, and medical intervention according to the patient's genetic predisposition, lifestyle, environmental exposure, and past health patterns. These plans are then proposed to medical professionals and patients. The remote monitoring and collaboration unit acquires biological or diagnostic data from patients remotely in real time and automatically notifies medical professionals when abnormal values ​​or increased risk occur. Remote diagnosis can be implemented when necessary. Thus, all these processes are performed by generative AI. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8 This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10 It represents an emotion graph that maps multiple emotions.

[0016] Figure Labels Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) The present invention is implemented as a comprehensive medical information system consisting of a server and terminals. The server side includes a medical record acquisition unit, a biological data acquisition unit, a genetic information acquisition unit, an environmental factor acquisition unit, a long-term memory management unit, a generative AI processing unit, a preventive medical plan proposal unit, and a remote monitoring and collaboration unit. The terminal side includes patient terminals and medical practitioner terminals. Patient terminals are equivalent to smartphones, tablets, wearable devices, home medical machines, etc., while medical practitioner terminals are equivalent to personal computers, tablets, dedicated diagnostic terminals, etc. Specifically, this comprehensive medical information system adopts a distributed computing architecture. The system consists of multiple computing node clusters. The server side has a parallel computing cluster for high-speed execution of large-scale language models or deep learning models. This parallel computing cluster has a hardware structure that interconnects multiple graphics processing units. The system asynchronously receives high-frequency and high-capacity time-series biological data sent from the terminal side. The system temporarily caches the received data in an in-memory database. The system processes the cached data using a hybrid batch processing and stream processing approach. The system applies an end-to-end encryption protocol in the communication between the terminal and the server. The system executes an algorithm for dynamically compressing the communication payload. This system minimizes network bandwidth consumption. It reduces communication latency. It achieves real-time data synchronization. It distributes rendered analysis results to healthcare professionals' terminals with low latency. It deploys lightweight inference models as edge computing for patient terminals. It enables initial analysis at the edge. It distributes the computational load on the server side. It increases the overall processing throughput of the system. This system functions not only as a simple information collection system but also as a data-driven advanced computing platform. It provides real-time support for decision-making in healthcare settings. This system is particularly effective in telemedicine and preventative healthcare.

[0040] First, the Medical Record Acquisition Department operates on a server, collaborating with electronic medical record systems or medical institution databases to automatically acquire each patient's diagnostic history, treatment history, past medical history, medication history, allergy information, hospitalization history, surgical history, examination results, image diagnostic data, vaccination history, and rehabilitation records. For example, this includes information such as a patient's past hospitalization for heart disease, prescription medications for diabetes, past surgical history or allergy records, diagnostic results from MRI or CT images, detailed numerical data from blood or urine tests, vaccination dates or adverse reaction records, and rehabilitation progress records. This information is managed uniformly using the patient ID as the key, enabling the integration and time-series management of data from various institutions, even when a patient has received treatment at multiple medical institutions. Specifically, this Medical Record Acquisition Department establishes connections to external medical information systems via a secure application programming interface. It converts medical records described in different data formats into a standardized data schema. The department executes a natural language processing model based on the Transformer architecture. Finally, it extracts medical entities from unstructured text data. This medical record acquisition unit converts the extracted entities into embedding vectors in a multi-dimensional vector space. For example, this unit receives the patient's chief complaint or the doctor's diagnostic opinion text as an input tensor. This input tensor is a sequence of token IDs for words. This natural language processing model calculates the contextual meaning of each token using a self-attention mechanism. The output of this natural language processing model represents a probability distribution of the disease name or drug name. This unit extracts the entity if the peak value of the output probability distribution exceeds a predetermined threshold. This unit associates the extracted entities with the patient's time-series profile. This unit applies a convolutional neural network to the image diagnostic data. This unit extracts feature maps from the images. This unit stores the feature maps as numerical vectors. This unit constructs heterogeneous mixed data into a unified graph structure. This unit defines patients, diseases, or drugs as nodes. This unit defines their relationships as edges. This unit prepares the input data for subsequent inference processing using the graph neural network. This unit automatically imputes missing values ​​in the data using a statistical imputation algorithm. This medical record acquisition department improves data quality. It eliminates manual data entry by healthcare professionals. It prevents human error in data entry. It dramatically increases the processing speed of data integration. This medical record acquisition department is used as the data sharing foundation in wide-area medical collaboration or community integrated care systems.

[0041] The biometric data acquisition unit acquires biometric data in real time from wearable devices or home medical machines (blood pressure monitors, electrocardiographs, blood glucose meters, thermometers, pulse oximeters, smartwatches, activity meters, sleep monitors, respiratory sensors, scales, body composition analyzers, etc.) on the terminal side and sends it to the server. For example, it automatically collects data such as blood pressure measurements taken at home, heart rate fluctuations during sleep, daytime steps or calorie expenditure, nighttime respiratory rate, peak heart rate during exercise, body temperature changes, blood oxygen saturation, changes in weight or body fat percentage, sleep depth or number of awakenings, and any respiratory arrest events. This data is transmitted from the terminal to the server via encrypted communication and accumulated as time-series data on the server side. For example, it envisions a patient measuring body temperature and blood pressure every morning upon waking, recording heart rate and sleep patterns at night using a wearable device, and measuring weight, body fat percentage, and muscle mass using a body composition analyzer on weekends. Specifically, this biometric data acquisition unit receives streaming data asynchronously transmitted from multiple sensors. This biological data acquisition unit synchronizes received data based on timestamps. It maps various biological data with different sampling rates onto a common time series axis. The unit performs data interpolation algorithms, using spline interpolation or linear interpolation to estimate missing intervals. It applies noise removal filters, employing Kalman filters or moving average filters to reduce high-frequency noise originating from sensors. The unit constructs a multi-dimensional time series tensor from the preprocessed time series data, for example, combining data such as heart rate, blood pressure, and body temperature. It generates a two-dimensional array with both time and sensor dimensions. This time series tensor is then input into a subsequent anomaly detection model. The unit also features a structure for executing lightweight machine learning models at the edge device. Initial data filtering is performed at the edge. The unit reduces the transmission frequency for data within the normal range. This biometric data acquisition unit sends high-resolution raw data to the server only when abnormal signs are detected. This biometric data acquisition unit significantly reduces the load on the communication network through dynamic bandwidth control. This biometric data acquisition unit optimizes storage consumption on the server side. This biometric data acquisition unit suppresses power consumption in battery-powered wearable devices. This biometric data acquisition unit enables long-term continuous monitoring. This biometric data acquisition unit demonstrates extremely high technical effectiveness in home monitoring of patients with chronic diseases or in care services for the elderly.

[0042] The Genetic Information Acquisition Department operates on a server, receiving and managing genetic marker information or disease risk gene information obtained by patients from genetic testing services or medical institutions on a patient-by-patient basis. Examples include specific SNP information related to heart disease risk, genetic factors for diabetes or hypertension, genetic polymorphisms of drug-metabolizing enzymes, risk genes for breast or colorectal cancer, risk genes for Alzheimer's disease, genetic predisposition to obesity or lipid disorders, and risk genes for drug side effects. This information, based on patient consent, is used for analysis by medical practitioners or AI. For example, it can utilize data uploaded by patients from external genetic testing services or the results of automated, collaborative genetic combination tests performed at medical institutions. Specifically, this Genetic Information Acquisition Department encodes base sequence data or single nucleotide polymorphism data into structured high-dimensional vectors. It represents the presence or absence of each genetic marker or the type of variation as binary or categorical values. This high-dimensional vector is then input into a dimensionality reduction algorithm. Principal component analysis or an autoencoder is used to extract latent feature vectors from the genetic information. This genetic information acquisition department integrates extracted latent feature vectors into the patient's profile data. This department has a module for calculating the correlation between genetic information and phenotypic data. This department evaluates the statistical significance between specific gene variants and past disease history. This department uses homomorphic encryption technology to process genetic data in an encrypted state. This department does not decrypt genetic information even during computational processing on the server. This department achieves ultimate privacy protection. This department regularly retrieves the latest genomic research findings from external databases. This department dynamically updates the association scores between genetic markers and disease risk. This department maintains the patient's genetic risk assessment based on the latest medical insights. This department is not limited to the static storage of genetic information. This department utilizes genetic information as a dynamic computational resource. This department functions as a foundational technology for personalized medicine or customized treatment. This department provides indispensable data in advanced medical reasoning, such as predicting the efficacy of specific anticancer drugs or pre-determining side effect risks.

[0043] The Environmental Factors Acquisition Department uses input interfaces on the terminal side or external database collaboration functions on the server side to acquire data on patients' lifestyle habits (smoking history, drinking history, exercise habits, diet, sleep habits, stress levels, medication adherence), living environment (air pollution levels, noise, temperature, humidity, insulation performance of the residence, access to nearby medical institutions), and occupational environment (work type, stress levels, work environment, presence of night shifts, long working hours, workplace safety and hygiene conditions). For example, it's equivalent to patients inputting their daily cigarette count or alcohol consumption in a smartphone application, automatically acquiring step count or exercise intensity data from a fitness application, obtaining local temperature or PM2.5 concentration from an external meteorological database, and collaborating on workplace stress assessment results. Furthermore, applications such as patients recording their daily diet with photos and generative AI estimating nutrients through image analysis are also possible. Specifically, this Environmental Factors Acquisition Department collects unstructured and semi-structured data from a variety of sources. It applies a convolutional neural network-based object detection model to the dietary image data. The Environmental Factors Acquisition Department determines the regions of food ingredients from the input RGB image tensor. This environmental factor acquisition unit outputs category labels and bounding box coordinates for each food item. It estimates the volume of each food item using the determined pixel area and distance information obtained from a depth sensor. The estimated volume is then compared to a food database. The unit calculates the calorie intake or macronutrient content as a numerical vector. This numerical vector is accumulated as daily dietary data. Spatial interpolation algorithms are applied to environmental data obtained from an external meteorological API. The unit estimates the concentration of particulate matter or temperature and humidity at the patient's precise residential location. This unit constructs a time-series multivariate matrix from these environmental factor data. This matrix is ​​then input into a subsequent causal inference model. The unit extracts hidden correlations between environmental factors and changes in biological data. For example, it calculates the delayed correlation between a specific drop in air pressure and a rise in the patient's blood pressure. The unit quantifies the impact of subtle environmental changes that are difficult for human intuition to grasp. This unit significantly reduces the input burden on patients through automated data collection. This environmental factors acquisition department eliminates self-reporting bias. It achieves objective and highly accurate acquisition of environmental data. This department is extremely useful in identifying environmental triggers in asthma patients or in the prevention and management of weather-related diseases.

[0044] The Long-Term Memory Management Department operates on a server, comprehensively accumulating and managing each patient's medical records, biological data, genetic information, and environmental factors in a time-series manner. For example, it records data such as blood pressure fluctuations over the past five years, seasonal heart rate variations, blood glucose changes before and after treatment interventions, weight changes before and after lifestyle improvement programs, recovery progress before and after surgery, sleep patterns before and after hospitalization, changes in health indicators accompanying life events such as job changes or relocations, medication adherence, and adverse reactions after vaccinations. All data is recorded in a time-series format. This allows for detailed tracking of long-term changes in each patient's health status, or changes in health indicators accompanying specific events (surgery, hospitalization, relocation, job change, changes in family structure, disasters, etc.). Specifically, this Long-Term Memory Management Department employs a hybrid storage architecture combining a distributed graph database and a time-series database. It stores frequently generated biological data in the time-series database and relational data such as medical records and life events in the graph database. The department dynamically synchronizes indexes between the two databases, enabling high-speed retrieval of complex time-series patterns. This long-term memory management system executes data compression algorithms. It extracts features from older data using near-lossy compression to save storage space. It applies lossless compression to recent data to maintain high resolution. It features data stratification capabilities that automatically move storage levels based on data importance or access frequency. It prevents performance degradation due to long-term data accumulation. It enables querying of decades of life log data with millisecond latency. It extracts biological response patterns from similar past events triggered by a patient's life events. These extracted patterns serve as initial weights for subsequent predictive models. This long-term memory management system is more than just a data repository. It functions as a data pipeline for automatically generating high-quality learning datasets for continuous learning in artificial intelligence models. It serves as the foundation for managing personal health records throughout a lifetime. It is best suited for accurately tracking slow changes in health status or the progression of chronic diseases with age.

[0045] The generative AI processing unit runs on a server, comprehensively analyzing all the aforementioned data. Generative AI combines multiple AI technologies, including natural language processing, image analysis, time series data analysis, pattern recognition, causal reasoning, and generative models. For example, natural language processing AI is used to extract disease risk from a patient's medical records; image recognition AI is used to analyze electrocardiogram waveforms or image data; and time series analysis AI is used to detect abnormal patterns in time series data such as blood pressure or heart rate. Furthermore, generative AI performs multivariate analysis on a patient's genetic factors, lifestyle habits, and environmental factors to predict future disease incidence or symptom exacerbation risks. For example, in the case of a specific combination of genetic markers and a history of hypertension, insufficient exercise, and smoking habits, AI automatically calculates that the risk of heart disease is increased several times higher than usual. Additionally, generative AI learns each patient's long-term health patterns, extracting individual-specific patterns such as "a tendency for blood pressure to rise every winter," "a tendency for a sharp rise in blood sugar levels after taking specific medications," "a tendency for stress levels to worsen and sleep quality to decrease after changing jobs," "a tendency for respiratory function to decrease during hay fever season," and "a tendency for blood pressure to worsen during periods of low medication adherence." These analytical results are automatically generated as easily understandable explanatory texts or reports for medical practitioners or patients using AI-based natural language generation capabilities. For example, specific suggestions are automatically generated such as, "Since your blood pressure tends to rise in winter, increase your salt intake and exercise starting in November," or "Since your sleep quality decreased during your last three job changes, increase your sleep monitoring after changing jobs." Specifically, this generative AI processing unit employs a multimodal neural network architecture. This generative AI processing unit projects text data, time-series numerical data, and image data into a common latent feature space. For example, this generative AI processing unit simultaneously receives, for instance, a one-dimensional time-series array representing the patient's blood pressure fluctuations over the past year, a binary vector representing genetic markers, and a two-dimensional tensor of the most recent chest X-ray image as input. This generative AI processing unit applies a Long Short-Term Memory network or a one-dimensional convolutional layer to the time-series data. This generative AI processing unit applies a residual network to the image data. This generative AI processing unit applies a Transformer encoder to the text data. This generative AI processing unit combines the feature vectors extracted from each modality in a combining layer. This generative AI processing unit inputs the combined high-dimensional vector into a fully connected layer. It outputs a continuous score between 0 and 1, representing the probability of developing a specific disease in the future. The unit then determines whether this score exceeds a pre-set warning threshold. If the threshold is exceeded, the unit uses a gradient-weighted class activation mapping technique to identify highly involved input features. Finally, the unit embeds these identified features and the disease probability score into a prompt from a large-scale language model. This enables the large-scale language model to generate personalized warning text for patients.This generative AI processing unit acquires natural language string data as output from a large-scale language model. This string data is then sent to the subsequent user interface unit. The generative AI processing unit uses the cross-entropy error function in model learning. It optimizes the weights of each network using backpropagation. This generative AI processing unit mathematically elucidates the complex nonlinear interactions of multiple factors that are difficult for human doctors to detect. This generative AI processing unit significantly improves diagnostic accuracy. This generative AI processing unit plays a significant role in personalized risk assessment or early intervention decision support in preventive medicine.

[0046] The preventive healthcare plan proposal department runs on a server and automatically generates optimized preventive healthcare plans for each patient based on the analysis results of the generative AI processing department. For example, for patients at high risk of heart disease, a plan combining "aerobic exercise three times a week," "practicing a low-sodium diet," "regular electrocardiogram checkups," and "attending stress management programs" is proposed; for patients at high risk of diabetes, recommendations include "calorie restriction diet," "post-meal blood glucose measurement," "strengthening exercise habits," and "ensuring sufficient sleep." Furthermore, the generative AI flexibly adjusts the plan content according to the patient's living or working environment. For example, for night shift workers, it automatically generates "diet, exercise, and sleep plans suitable for shift work"; for the elderly, it automatically generates "muscle strength training programs to prevent falls"; and for local residents, it automatically generates individually optimized content such as "regular health checkups utilizing remote consultations." In addition, it also considers the patient's preferences, lifestyle, family structure, economic situation, and religious restrictions, such as "proposing alternative ingredients when certain foods need to be avoided for religious reasons," "proposing low-cost exercise plans considering economic burden," and "proposing health programs that can be implemented with family," enabling detailed customization. These plans are automatically distributed to patient or healthcare practitioner terminals and updated in real time based on patient understanding and implementation status. For example, if a patient fails to implement a plan, AI performs causal analysis and automatically generates easily implementable alternatives or suggestions to maintain motivation. Specifically, this preventive healthcare plan proposal department possesses a hybrid inference engine combining reinforcement learning algorithms and constrained optimization solvers. This preventive healthcare plan proposal department explores action sequences that minimize the distance between the patient's current health state vector and the target health state vector. This preventive healthcare plan proposal department receives the patient's disease risk score, living environment parameters, and categorical variables representing economic constraints as input. This preventive healthcare plan proposal department calculates the expected rewards of various intervention plans in a Markov decision process-based simulation environment. This preventive healthcare plan proposal department defines the degree of improvement in health indicators as a positive reward and the difficulty of plan implementation or economic cost as a negative penalty as a reward function. This preventive healthcare plan proposal department generates structured JSON data as output, containing a list of specific action tasks and recommended execution frequency and priority scores for each task. This preventive healthcare plan proposal department inputs this structured data into a large-scale language model. The Preventive Medicine Plan Proposal Department translates the proposal text into natural language that aligns with the patient's health literacy level. The department continuously monitors historical task execution data from the patient's terminal. If the task incomplete rate exceeds a certain threshold, the department determines the plan is too difficult. The department dynamically updates the parameters of the reinforcement learning model. The department recalculates alternative tasks with a higher probability of execution.This preventative healthcare program proposal department dramatically improves patient adherence through this adaptive feedback loop. It fundamentally revolutionizes existing guidance methods based on one-size-fits-all guidelines. It achieves dynamic health management fully synchronized with an individual's lifestyle. This preventative healthcare program proposal department functions as a concrete solution for the severe prevention of lifestyle-related diseases or for reducing medical costs.

[0047] The remote monitoring and collaboration unit operates on a server, receiving biometric data in real time from patients' home medical devices or wearable devices and sequentially sending it to the generative AI processing unit. The generative AI analyzes the received data in real time and automatically alerts healthcare professionals via their terminals when anomalies or increased risk are detected. For example, it immediately notifies healthcare professionals of sudden increases in heart rate, dangerously high blood sugar levels, or respiratory arrest during sleep, and schedules remote consultations or emergency responses as needed. Furthermore, in cases where patients have difficulty moving or reside in remote locations, the AI ​​automatically determines the feasibility of a remote consultation and notifies both the patient and healthcare professional of the scheduled appointment. This allows for flexible utilization of remote consultations in various situations, such as when patients are elderly and unable to travel to hospitals, during infectious disease outbreaks requiring restricted movement, or when access to medical facilities is difficult during disasters. Specifically, this remote monitoring and collaboration unit employs an event-driven stream processing architecture. It asynchronously processes thousands of biometric data packets per second via a message queue system. This remote monitoring and collaboration unit applies a sliding window anomaly detection algorithm to the received time-series data stream. It receives continuous numerical arrays of ECG waveform tensors or blood oxygen saturation values ​​from the most recent minutes as input data. The unit uses an autoencoder-based neural network to reconstruct the input data. It calculates the reconstruction error between the input data and the reconstructed data. This reconstruction error is output as an anomaly score. If the anomaly score exceeds a dynamically set threshold, the unit determines that a critical event has occurred. Based on this determination, the unit initiates the subsequent alarm generation module. In the alarm generation module, the unit comprehensively evaluates the patient's location, the severity of vital signs, and the availability of nearby medical facilities. The unit determines the optimal response. It immediately executes the automatic ambulance request protocol or pushes a notification to the attending physician. It executes a decision tree algorithm to determine the necessity of remote examination. The unit receives the patient's mobility score or infection risk index as input. This remote monitoring and collaboration department outputs a binary flag indicating the eligibility for remote consultation. When this flag is true, the department automatically generates a session URL for the video conferencing system. The department automatically registers the appointment in the calendars of relevant personnel. This department minimizes delays in medical intervention. This department dramatically improves the prognosis of acute illnesses. This department plays a crucial role in eliminating healthcare access gaps in sparsely populated areas and preventing healthcare collapse during pandemics.

[0048] Furthermore, the generative AI automatically adjusts the explanatory text or plan content based on the patient's comprehension level or implementation status. For example, it uses easily understandable expressions for the elderly and provides professional explanations for medical professionals, achieving optimal information delivery based on user attributes. For instance, if the patient fails to understand the plan content, it automatically generates explanations using illustrations or videos, and can also generate multilingual explanatory texts for foreign language speakers. Specifically, this generative AI applies conditional text generation technology in the decoder part of the natural language generation model. This generative AI receives structured data of medical analysis results and the target user's attribute vector as input. This attribute vector contains numerical parameters representing age group, level of professional knowledge, native language, or visual cognitive ability. This generative AI embeds these input parameters as contextual prompts. This generative AI applies vocabulary difficulty adjustments or sentence length limits based on the user's reading comprehension score. This generative AI generates customized text strings as output. In addition to text generation, this generative AI also executes a text-to-image conversion model in parallel. This generative AI dynamically renders pixel data of infographics or animations that visually supplement the meaning of the generated text. This generative AI collects log data on user scrolling speed, dwell time, and click actions on the terminal. It then executes a classification model to estimate the user's comprehension level from this log data. When comprehension is deemed low, the AI ​​further refines the granularity of the explanation. It activates a conversational chatbot interface to directly extract the user's questions. This adaptive information prompting mechanism eliminates information asymmetry in healthcare. It technically guarantees informed consent from patients. This generative AI completely removes communication barriers in international hospitals with multinational patients or in community healthcare settings facing an aging population.

[0049] Furthermore, generative AI anonymizes, aggregates, and analyzes group-level risk propensity or preventative effects from multiple patient data. It can also be used to propose health-enhancing policies or optimize healthcare resource allocation to medical institutions or administrative agencies across a region. For example, in cases of increased heart disease risk in a specific area, AI can propose health-inspiring programs or group physical examinations for local residents. Additionally, during infectious disease outbreaks, aggregating patient fever or cough symptom data can detect early signs of infection spread and notify administrative agencies. Specifically, this generative AI applies a federated learning architecture or secure multi-party computation technology. It comprehensively analyzes data from multiple facilities without infringing on individual privacy. It performs local gradient calculations on individual patient terminals or edge servers of medical institutions. The calculated gradient information is aggregated only to a central server. It updates the weights of the global model. As input data, it receives anonymized time-series tensors of symptom frequency or abnormal vital sign detections associated with specific geographic grid codes. It executes a spatiotemporal graph convolutional network. This generative AI models both spatial adjacency and temporal progression. As output, it generates a numerical matrix of heatmap data representing the future probability of disease occurrence in each region or a predicted number of hospital beds. This predicted matrix is ​​then fed into a subsequent optimization algorithm. It calculates the optimal allocation plan for limited medical personnel or medical supplies. It outputs probabilistic scores indicating signs of infectious disease clusters several days to weeks in advance. When this score exceeds a warning level, it automatically sends an API request to the administrative dashboard system. This generative AI provides data-driven support for public health decision-making. It eliminates the time lag in epidemiological investigations. It functions as a city-level healthcare operating system within smart cities. It provides a robust technological foundation for medical triage during pandemic responses or disasters.

[0050] Thus, the implementation of this invention, with the roles of server and terminal shared, uses generative AI as its core to comprehensively and precisely achieve a highly practical and scalable integrated medical information system, from individual optimization for each patient to health management at the group level. For example, it automatically collects data unconsciously acquired by patients in their daily lives (steps, sleep, diet, stress, temperature, noise, air quality, medication status, etc.), and the generative AI comprehensively analyzes this data to predict future health risks, individually propose optimal preventative measures, immediately notify medical professionals in case of abnormalities, and automatically dispatch remote consultations or emergency responses when needed. This fully automated process is a significant feature of this invention. Furthermore, it comprehensively considers the diverse factors that existing medical systems have not fully considered, such as the patient's living environment or social background, economic status, family structure, hobbies, and religious restrictions, enabling detailed individual optimization—another important feature of this invention. Specifically, this system is deployed on cloud infrastructure as a containerized group of modules based on a microservice architecture. The system has a structure where each functional module can be independently and horizontally scaled. This system dynamically activates autoscaling to handle surges in data traffic. It maximizes the availability and fault tolerance of the entire system. This system is fundamentally different from existing electronic health record systems that merely record or display data. It constructs a highly computational pipeline for real-time processing of multidimensional, heterogeneous, and mixed data. It extracts complex correlations beyond the limits of human cognition through deep learning models. It provides feedback on the extracted insights as immediately actionable plans. It functions as a closed-loop control system. It fully automates the entire process from data collection and analysis to intervention and outcome re-evaluation. It dramatically reduces the workload of healthcare professionals. It equalizes the quality of healthcare services. It powerfully advances patient adherence and behavioral change through the power of algorithms. It completely shifts the paradigm of preventative medicine from post-mortem response to pre-mortem prediction. It functions as the de facto standard for next-generation digital healthcare platforms. It also has broad application potential as the foundation for insurance companies to build dynamic pricing models or for pharmaceutical companies to collect real-world data.

[0051] (System Composition) The system involved in this embodiment includes a medical record acquisition department, a biological data acquisition department, a genetic information acquisition department, an environmental factor acquisition department, a long-term memory management department, a generative AI processing department, a preventive medical plan proposal department, a remote monitoring and collaboration department, a user interface department, and a security management department.

[0052] The Medical Records Acquisition Department collaborates with electronic medical record systems or medical institution databases to automatically acquire a wide range of medical information for each patient, including diagnostic history, treatment history, past medical history, medication history, allergy information, hospitalization history, surgical history, examination results, image diagnostic data, vaccination history, and rehabilitation records. For example, it comprehensively collects hospitalization history for heart disease, prescription history for diabetes medication, records of past surgeries or allergic reactions, diagnostic results from MRI or CT images, detailed numerical data from blood or urine tests, vaccination dates or adverse reaction records, and rehabilitation progress records, managing them uniformly using the patient ID as the key. Furthermore, even for patients treated at multiple medical institutions, it can integrate data from each institution and manage it chronologically. Specifically, this Medical Records Acquisition Department has a data integration engine that collaborates with distributed database systems. It parses incompatible data formats output from electronic medical record systems from different vendors in real time. The extracted data is converted into structured data in JSON format, conforming to internationally standardized medical information exchange specifications. This medical record acquisition unit employs rule-based syntactic analysis and machine learning-based semantic analysis in its transformation process. For example, it receives physician opinions described in free text as input. It uses a named entity recognition model to determine tags representing disease names, locations, or severity. The unit automatically maps these tags to the code structure of the standard disease master data. It stores the mapped codes in a unified patient profile database. For image diagnostic data, it analyzes the header information in DICOM format. It extracts image pixel data as a 3D tensor. It normalizes this 3D tensor to match the input format of subsequent image recognition models. It performs data cleaning algorithms to detect duplicates or inconsistencies in the data. For example, it integrates prescriptions for the same medication from different hospitals through directory merging. It completely eliminates data silos. It constructs a comprehensive and accurate longitudinal dataset of patients. It fundamentally supports the inference accuracy of machine learning models. This medical record acquisition department functions as a hub system of the regional medical information network.

[0053] The biometric data acquisition unit acquires biometric data in real time from wearable devices worn by the patient, home medical machines, smartphone applications, etc., and sends it to the server. For example, it automatically collects a wide variety of biometric data, such as blood pressure and electrocardiogram data measured by a blood pressure monitor or electrocardiograph, blood glucose levels measured by a blood glucose meter, body temperature measured by a thermometer, blood oxygen saturation measured by a pulse oximeter, steps or calories burned measured by a smartwatch or activity meter, sleep depth or number of awakenings measured by a sleep monitor, respiratory rate or respiratory arrest events measured by a respiratory sensor, and weight, body fat percentage, and muscle mass measured by a scale or body composition analyzer. For example, it envisions a patient measuring body temperature and blood pressure every morning upon waking, recording heart rate and sleep status at night using a wearable device, and measuring weight, body fat percentage, and muscle mass using a body composition analyzer over the weekend. This data is sent from the terminal to the server via encrypted communication and accumulated as time-series data on the server side. Specifically, this biometric data acquisition unit collaborates with an edge gateway module equipped with an IoT device management protocol. This biometric data acquisition unit receives raw data streams in binary format from various sensor devices connected via Bluetooth or Wi-Fi. It decodes the raw data stream and converts it into an array of floating-point numbers representing physical quantities. The unit assigns a timestamp and device ID to the converted data. It queues the data in a write buffer of a time-series database. A retransmission control mechanism is installed to ensure reliable data transmission. When the network is disconnected, the unit temporarily stores the data in the local memory on the edge device side. Differential data is automatically synchronized when communication is restored. For high-sampling-rate data such as electrocardiogram waveforms, the unit performs a Fast Fourier Transform at the edge. It extracts a feature vector in the frequency domain and sends only this feature vector to the server. This edge computing technology dramatically reduces communication bandwidth consumption. The unit distributes the data processing load on the server side. This biometric data acquisition unit enables real-time monitoring of vital signs. It is highly effective in the early detection of arrhythmias and in screening for sleep apnea syndrome.

[0054] The Genetic Information Acquisition Department receives and manages genetic marker information or disease risk gene information obtained by patients from genetic testing services or medical institutions on a patient-by-patient basis. Examples include specific SNP information related to heart disease risk, genetic factors for diabetes or hypertension, genetic polymorphisms of drug-metabolizing enzymes, risk genes for breast or colorectal cancer, risk genes for Alzheimer's disease, genetic predisposition to obesity or lipid disorders, and risk genes for drug side effects. Data obtained from external genetic testing services can be uploaded by patients, or results from combined genetic tests performed automatically at medical institutions can be used. Furthermore, family history or disease predispositions among relatives are also collected to improve the accuracy of genetic risk assessment. Specifically, this Genetic Information Acquisition Department has a dedicated bioinformatics pipeline for processing large volumes of genomic sequence data output from next-generation sequencers. This department extracts specific single nucleotide polymorphisms or insertion / deletion variants from input VCF format files. The department compares the extracted variant data with a reference genomic sequence. The department then compares the pathological significance of each variant with annotation databases and assigns a score. This genetic information acquisition unit encodes the score into a multidimensional genetic risk vector. It associates this vector with the patient's master record. A graph data structure is used to process family history data. The unit constructs a network of nodes and edges from the patient's pedigree. Each node is assigned a disease history attribute. A network propagation algorithm is applied to this graph data. The unit calculates the potential genetic risk probability among asymptomatic relatives. This calculated probability value is provided as a priori probability for subsequent disease prediction models. The unit functions not as a static repository of genetic information but as a dynamic risk calculation engine. It accelerates the integrated analysis of genomic and clinical data. It provides an indispensable technological foundation for supporting optimal drug selection or dosage decisions based on pharmacogenomics.

[0055] The Environmental Factors Acquisition Department gathers data on patients' lifestyle habits, living environment, and occupational environment from multiple perspectives. For example, it acquires lifestyle data such as smoking history, drinking history, exercise habits, diet, sleep habits, stress levels, and medication adherence; living environment data such as local air pollution levels, noise levels, temperature, humidity, building insulation, and visits to nearby medical institutions; and occupational environment data such as work type, stress levels, work environment, presence of night shifts, long working hours, and workplace safety and hygiene. This is equivalent to patients inputting their daily cigarette count or alcohol consumption into a smartphone app, automatically obtaining step count or exercise intensity data from a fitness app, and acquiring local temperature or PM2.5 concentration from an external meteorological database, as well as workplace stress assessment results. Furthermore, applications such as patients recording their daily diet with photos and generative AI estimating nutrients through image analysis are also possible. Specifically, this Environmental Factors Acquisition Department employs a data lake architecture that integrates heterogeneous data asynchronously collected from diverse data sources. The department receives a time-series coordinate array of location information obtained from the smartphone's GPS sensor as input. This environmental factor acquisition unit analyzes the coordinate array using a spatial clustering algorithm. It automatically identifies polygonal regions representing the patient's home, workplace, or frequently visited locations. Using the centroid coordinates of these identified regions, it sends queries to an external environmental data API. The unit acquires time-series data on particulate matter concentration, UV index, or noise level for these regions. It resamples and combines the acquired environmental data along the same timeline as the patient's biological data. For dietary image data, it applies a semantic segmentation model. It outputs a probability tensor indicating which food category each pixel belongs to. From this probability tensor, it calculates the area of ​​each food component. Using area ratios and standard density data, it estimates the weight of each food component. From the estimated weight, it calculates the nutrient content as a numerical vector. This series of processes fully automates the quantification of environmental factors. It constructs an objective dataset independent of the patient's subjective reports. This Environmental Factors Acquisition Department is extremely useful in analyzing the epigenetic effects of environmental factors on health or in epidemiological surveys in public health.

[0056] The Long-Term Memory Management Department comprehensively accumulates and manages each patient's medical records, biological data, genetic information, and environmental factors in a time-series manner. For example, it records data such as blood pressure fluctuations over the past five years, seasonal heart rate variations, blood glucose changes before and after treatment interventions, weight changes before and after lifestyle improvement programs, recovery progress before and after surgery, sleep patterns before and after hospitalization, changes in health indicators accompanying life events such as job changes or relocations, medication adherence, and adverse reactions after vaccinations. All data is recorded in a time-series format. This allows for detailed tracking of long-term changes in each patient's health status, or changes in health indicators accompanying specific events (surgery, hospitalization, relocation, job change, changes in family composition, disasters, etc.). Specifically, the Long-Term Memory Management Department employs a distributed columnar database capable of processing petabyte-scale data with low latency. The department partitions data using patient ID and timestamp as composite primary keys. It performs high-speed scans of data within specific time windows for time-series data retrieval queries. The department utilizes incremental coding and run-length encoding algorithms for data compression. This long-term memory management system stores only the differences in continuous biological data. It significantly reduces the physical capacity of storage. It executes an event-driven data tagging engine in the background. For example, upon detecting a marker of surgical procedure in an electronic medical record, it automatically assigns metadata tags to biological data from several months prior and after that event. This system uses these metadata tags to optimize the execution speed of SQL queries that extract shifts in health indicators before and after specific events. It automatically implements data lifecycle management strategies. It transparently migrates older, less frequently accessed data to inexpensive cold storage. This system minimizes operational costs through hierarchical storage management. It securely and comprehensively maintains vast amounts of healthcare data spanning a lifetime. It provides an indispensable data foundation for subsequent deep learning models to learn long-term temporal dependencies. It plays a significant role in predicting long-term prognoses of chronic diseases and modeling physiological changes associated with aging.

[0057] The generative AI processing unit comprehensively analyzes all the above data. Generative AI combines multiple AI technologies, including natural language processing, image analysis, time series data analysis, pattern recognition, causal inference, and generative models. For example, natural language processing AI is used to extract disease risk from a patient's medical records; image recognition AI is used to analyze electrocardiogram waveforms or image data; and time series analysis AI is used to detect abnormal patterns in time series data such as blood pressure or heart rate. Furthermore, generative AI performs multivariate analysis on a patient's genetic factors, lifestyle habits, and environmental factors to predict future disease incidence or symptom exacerbation risks. For example, in the case of a specific combination of genetic markers and a history of hypertension, insufficient exercise, and smoking habits, AI automatically calculates that the risk of heart disease is increased several times higher than usual. In addition, generative AI learns each patient's long-term health patterns, extracting individual-specific patterns such as "a tendency for blood pressure to rise every winter," "a tendency for a sharp rise in blood sugar levels after taking specific medications," "a tendency for stress levels to worsen and sleep quality to decrease after changing jobs," "a tendency for respiratory function to decrease during hay fever season," and "a tendency for blood pressure to worsen during periods of low medication adherence." These analytical results are automatically generated as explanatory texts or reports that are easy for medical practitioners or patients to understand, using AI-based natural language generation capabilities. For example, specific suggestions are automatically generated such as, "Since your blood pressure tends to rise in winter, strengthen your low-salt diet and exercise starting in November," or "Since your sleep quality decreased during your last three job changes, strengthen sleep monitoring after changing jobs." As a specific example of prompts that the generative AI can read, consider this example: "Based on the following patient data, predict future heart disease risk and generate a lifestyle improvement plan to reduce that risk. Patient data: Age 55, male; Past medical history: hypertension, diabetes; Genetic information: positive for heart disease risk genes; Lifestyle habits: history of smoking, no exercise; Diet: high-fat diet; Latest blood pressure: 150 / 95; Latest blood glucose: 140 mg / dL; Residence: urban area; Occupation: night shift, high stress level." Specifically, this generative AI processing unit employs an ensemble learning architecture. This generative AI processing unit executes gradient boosting decision tree models and deep neural network models in parallel. This generative AI processing unit receives a high-dimensional feature tensor as input data, which combines scalar values ​​such as the patient's age or blood pressure, binary vectors of genetic information, and textual embedding vectors of past medical records. This feature tensor is then input into each model. The generative AI processing unit normalizes the logarithmic odds values ​​of the disease incidence probabilities output from each model using a softmax function. The weighted average of the normalized probability values ​​is then used to calculate the final risk score. If this risk score exceeds a predetermined threshold, the generative AI processing unit initiates the causal inference module.This generative AI processing unit uses a Bayesian network to determine, through probabilistic inference, which input feature contributes most to the increase in risk score. This unit then formats the determined contribution ranking and risk score into structured data. This structured data is dynamically inserted into the input prompts of a large language model. The large language model then performs autoregressive label generation. As output, this unit generates natural language text data that prompts specific behavioral changes for patients. This text data is then sent to the subsequent user interface. This unit does not black-box the model's inference process. It uses Shapley and interpretation methods to quantitatively visualize the contribution of each feature. This unit fully meets the requirements of interpretable AI for healthcare practitioners. It delivers extremely high technical effectiveness in improving the accuracy of diagnostic support and accelerating physician decision-making.

[0058] Based on the analysis results from the generative AI processing unit, the Preventive Medicine Plan Proposal Department automatically generates optimized preventive medicine plans for each patient. For example, for patients at high risk of heart disease, a plan combining "aerobic exercise three times a week," "practicing a low-sodium diet," "regular electrocardiogram checkups," and "participating in a stress management program" is proposed; for patients at high risk of diabetes, recommendations include "calorie restriction," "measuring blood sugar levels after each meal," "strengthening exercise habits," and "ensuring sufficient sleep." Furthermore, the plan content is flexibly adjusted according to the patient's living or working environment. For example, a "diet, exercise, and sleep plan that fits the work schedule" is automatically generated for night shift workers; a "muscle strength training program to prevent falls" is automatically generated for the elderly; and for local residents, "regular health checkups utilizing remote consultations" are automatically generated, with individualized optimizations. In addition, the plan considers the patient's hobbies or lifestyle, family structure, economic situation, religious restrictions, etc., such as proposing alternative ingredients when certain foods need to be avoided for religious reasons, proposing low-cost exercise plans considering economic burden, and proposing health programs that can be performed with family. As a specific example of a prompt text that generative AI can read, one could be: "Based on the following health risk analysis results for a patient, generate the optimal preventive medical plan in Japanese, taking into account the patient's living environment, hobbies, and economic situation. Patient information: Age 68, female; Past medical history: osteoporosis; Exercise habits: weekly walks; Family structure: living with husband; Economic situation: living on pension; Hobbies: primarily Japanese cuisine; Religious restrictions: cannot eat certain meats." Specifically, this preventive medical plan proposal department employs a retrieval-enhanced generative architecture that integrates knowledge graphs and large-scale language models. This department receives a multi-dimensional vector representing the patient's disease risk score, living environment parameters, and economic restrictions as input. It uses this input vector as a query to retrieve a vector database storing medical evidence or clinical guidelines. The department extracts a list of intervention tasks with the highest relevance using cosine similarity calculation. It combines this extracted task list with the patient's restrictions as context for the prompt. Finally, it inputs the combined prompt into a large-scale language model. This preventative healthcare plan proposal department uses a large-scale language model to generate text data containing specific scheduling or alternative solutions that meet certain constraints. The department applies rule-based filtering to the output text data. It automatically verifies for medically contraindicated combinations or logical contradictions. The department then pushes the verified text data as the final preventative healthcare plan to the patient's terminal. The department continuously monitors task completion markers or changes in biometric data sent from the patient's terminal.This preventative healthcare plan proposal department calculates a reward function to evaluate the effectiveness of the plan. Based on the value of this reward function, the department dynamically fine-tunes the retrieval weights or suggested parameters when generating the next plan. Through this self-optimization cycle, the department continuously improves the personalization accuracy of the plan. This department scientifically and efficiently guides patients to change their behavior. This department provides extremely powerful solutions for the prevention of lifestyle-related diseases and the extension of healthy lifespan.

[0059] The remote monitoring and collaboration department receives biometric data from patients' home medical devices or wearable devices in real time and sends it sequentially to the generative AI processing department. The generative AI analyzes the received data in real time and automatically sends alerts to healthcare professionals via their terminals when anomalies or increased risk are detected. For example, it immediately notifies healthcare professionals in cases of a sudden increase in heart rate, dangerously high blood sugar levels, or sleep apnea, allowing for remote consultation or emergency response if necessary. Furthermore, in cases where the patient has difficulty moving or resides in a particular location, the AI ​​automatically determines the feasibility of a remote consultation and notifies both the patient and healthcare professional of the scheduled appointment. A specific example of a prompt that the generative AI can read could be: "Analyze the following real-time biometric data. If anomalies or increased risk are detected, generate a notification to the healthcare professional in Japanese, along with recommended response strategies. Biometric data: Heart rate 120 bpm, blood pressure 160 / 100, blood sugar 180 mg / dL, sleep apnea event." Specifically, this remote monitoring collaboration unit has a data collection module that utilizes edge computing technology. This module receives biological signals asynchronously transmitted from multiple sensor nodes. This module converts the received analog signals into digital data. This module generates time-series array data at a sampling rate of 100 Hz. This module inputs the generated time-series array data into a noise removal filter. This filter uses a moving average method to reduce high-frequency noise. This module sends the preprocessed time-series array data to the generative AI processing unit. This generative AI processing unit is equipped with a multimodal neural network that integrates a large-scale language model and a time-series prediction model. This network adopts a Transformer architecture with a self-attention mechanism. The input data to this network is in the form of a multidimensional tensor. This input tensor contains numerical vectors representing heart rate, blood pressure, blood glucose level, and respiratory status. The dimension of this input tensor is, for example, 128 dimensions. As a specific example of this input data, there is a heart rate vector of 120 beats per minute, a blood pressure vector of 160 to 100, and a blood glucose vector of 180 milligrams per minute. This network maps the input tensor to a high-dimensional feature space. This network calculates the Mahalanobis distance between the past normal data distribution and the current input data. Based on this calculated distance, the network determines anomaly scores. The output data from this network is a combination of floating-point numbers representing anomaly probabilities and natural language text representing recommended actions. Specific examples of this output data include an anomaly probability score of 95%, text recommending emergency transport, and control commands instructing for remote consultation appointments. The system compares the output anomaly probability scores to pre-set thresholds. If the score exceeds the threshold, the system activates an alarm generation module. This module sends an instant notification data packet to the healthcare professional's terminal. The system drives the scheduler based on the output control commands.This scheduler retrieves idle time and automatically registers scheduled remote examinations into the database. Through this process, the system achieves real-time and high-precision anomaly detection without relying on human visual monitoring. This system significantly reduces the cognitive load on healthcare professionals. Because it sends only critical alerts without consuming excessive communication bandwidth, this system optimizes network resource consumption. This technology is extremely useful in use cases such as home healthcare monitoring of the elderly in sparsely populated areas.

[0060] The user interface (UI) provides functions such as data input, plan confirmation, alarm reception, and remote consultation participation on both patient and healthcare professional terminals. For example, patients input their daily health status or lifestyle habits into a smartphone application, and wearable devices automatically transmit biometric data. Healthcare professionals confirm AI-generated reports or plans on their personal computers and conduct remote consultations with patients when necessary. Furthermore, the UI automatically adjusts the explanatory text or plan content based on the patient's comprehension and implementation status, providing easy-to-understand explanations for the elderly and professional interpretations for healthcare professionals, achieving optimal information delivery based on user attributes. Specifically, this UI functions as a front-end module with a dynamic rendering engine. This module automatically optimizes the display layout based on the terminal's screen size or resolution. This module provides an API endpoint for receiving input data from the patient terminal. This module serializes the input text or biometric data. This module sends the serialized data to the server via a secure communication protocol. This server analyzes the received data and extracts the user's attribute information. This server drives a natural language generation model based on the extracted attribute information. This model receives the user's age, literacy level, and professional knowledge as input parameters. The input data consists of raw text containing medical terminology and vectors of categorical variables representing target attributes. Specific examples of this input data include pathological descriptions of myocardial infarction, marker variables representing the elderly, and marker variables representing non-medical practitioners. The model projects the input vectors into a latent space. It uses a decoder network to reconstruct the text while selecting easily understandable vocabulary. The output data from this model consists of natural language text with adjusted difficulty and metadata for display. Specific examples of this output data include easily understandable descriptions of cardiovascular blockage, control markers for enlarging text size, and phonetic data for voice reading. The user interface is based on the output metadata to perform screen rendering. It uses a color scheme with increased contrast for elderly users. It generates dashboards with overlaid time-series charts or detailed numerical data for medical practitioners. Through this processing, the system improves the accuracy of information delivery for diverse users. This system reduces the risk of medical accidents caused by user errors or misunderstandings. This system automates the generation of displayed content, eliminating the need for manual updates by system administrators. This technology can be applied to use cases such as multi-generational regional integrated nursing systems or telemedicine platforms where specialists and general patients share information.

[0061] The Security Management Department implements encryption for all data communication and storage, and employs security measures such as anonymization of patient information, access control management, multi-factor authentication, audit log recording, and unauthorized access detection. For example, TLS-based encryption is used for sending and receiving patient data, AES encryption is applied to the database on the server, and access permissions are meticulously set for each healthcare professional, preventing third parties from accessing data without the patient's consent. Furthermore, a security audit log is recorded for the entire system, and AI automatically detects signs of unauthorized access or information leakage and notifies the administrator. Specifically, the Security Management Department collaborates with the hardware security module to generate and store encryption keys. The department establishes mutual authentication based on public key infrastructure between the data source and destination. The department uses Transport Layer Security (TLS) protocols to encrypt the communication path end-to-end. The department applies Advanced Encryption Standard (AES) algorithms to each record stored in the database. The department uses hash functions to convert patients' personally identifiable information into irreversibly anonymized data. The department expands role-based access control lists into memory. For user authentication requests, the department performs a multi-factor authentication process including one-time passwords or biometric authentication. All transactions occurring within the system, along with their timestamps, are recorded as audit logs in an append-only database. An AI model for anomaly detection is installed to monitor the recorded log data in real time. This model employs an autoencoder-type neural network architecture. The input data to this model is a multi-dimensional feature vector containing access time, IP address, URI of the requested resource, and data transfer volume. Specific examples of this input data include an access time vector at 2 AM, a categorical variable representing an overseas IP address, and a query vector requesting a large amount of medical record data. The model compresses the input vector into a low-dimensional latent representation. The model reconstructs the original feature vector from the latent representation. The model calculates the reconstruction error between the input vector and the reconstructed vector. The output data from this model is a threat score based on the reconstruction error and a contribution vector representing anomalies. Specific examples of this output data include a 99% threat score, a flag indicating an anomaly in the IP address, and a flag indicating an anomaly in the data transfer volume. If the output threat score exceeds a specified threshold, the department immediately and forcibly terminates the session. The department sends an emergency alert to the administrator's terminal. This process prevents information leaks caused by unknown cyberattacks or internal breaches. The system reduces the latency from security incident detection to response to milliseconds. This technology is an indispensable foundational technology in use cases such as large-scale medical data centers handling confidential personal information or wide-area medical information networks where multiple medical institutions collaborate.

[0062] As described above, the system described in this embodiment is a highly practical and scalable integrated medical information system that collaboratively manages and analyzes diverse medical and health data for each patient. It utilizes generative AI as its core to achieve automated generation of individually optimized preventative medical plans, real-time monitoring, alarm notifications for anomalies, and automated scheduling of remote consultations. Specifically, this system coordinates the actions of multiple computing nodes based on a distributed processing architecture. It employs a microservice architecture where each part exchanges data via asynchronous message queues. The system executes each stage of data collection, accumulation, analysis, proposal, and notification as a pipeline. Generative AI is configured as the core of the inference engine. This AI is a hybrid model that integrates a large-scale language model with billions of parameters with a recurrent neural network specialized for time series prediction. The system integrates the diverse medical data of each patient as a high-dimensional tensor. The system inputs the integrated tensor into the AI ​​for parallel computation. The system dynamically generates individually optimized preventative medical plans. The system uses a stream processing engine to continuously evaluate real-time monitoring data. This system handles interruptions upon anomaly detection and immediately issues alarm notifications. It formulates the scheduling of remote consultations as a constraint satisfaction problem, which is then automatically solved using optimization algorithms. The system autonomously performs these complex computations without human intervention. This system dramatically reduces the workload of healthcare professionals. It provides preventative healthcare based on data-driven, objective indicators. Through dynamic allocation of computing resources, the system is easily scalable horizontally to accommodate increases in the number of patients. This technology is applied as a core system for building a national-scale healthcare information infrastructure or for envisioning next-generation smart hospitals.

[0063] (Implementation steps) Step 1: Diverse Collection of Patient Information. In this step, the patient's medical record acquisition department collaborates with electronic medical record systems or clinical databases to comprehensively acquire diagnostic history, treatment history, past medical history, medication history, allergy information, hospitalization history, surgical history, examination results, image diagnostic data, vaccination history, rehabilitation records, etc. In addition, the biometric data acquisition department automatically collects real-time data such as blood pressure, heart rate, blood glucose levels, body temperature, blood oxygen saturation, steps, calories burned, sleep status, respiratory rate, weight, body fat percentage, and muscle mass from wearable devices, home medical machines, and smartphone applications. The genetic information acquisition department obtains information on risk genes for heart disease, diabetes, drug-metabolizing enzymes, cancer, and Alzheimer's disease from genetic testing services or medical institutions. In addition, the Environmental Factors Acquisition Department collects information from multiple perspectives, including lifestyle habits (smoking history, drinking history, exercise habits, diet, sleep habits, stress levels, medication adherence), living environment (air pollution, noise, temperature, humidity, housing performance, visits to medical institutions), and occupational environment (work style, stress, work environment, night shifts, long working hours, safety and hygiene conditions). Patients input their daily lifestyle habits or dietary information into a smartphone application, or the system automatically retrieves meteorological or environmental data from external databases. Specifically, in this step, the system initiates a data ingestion pipeline that asynchronously collects data from multiple heterogeneous data sources. The Medical Records Acquisition Department issues standardized queries to the hospital information system's database. The department parses the returned structured data. The department uses a natural language processing model to extract entities from unstructured text. The Biological Data Acquisition Department establishes a pairing with the wearable device via a near-field communication protocol. The department temporarily stores the biological signals streaming from the device in a buffer. The department applies a Fast Fourier Transform to extract frequency components from the waveform data acquired via an analog-to-digital converter. The Genetic Information Acquisition Department receives variant call files from an external genome analysis center using a secure file transfer protocol. This unit parses the base sequence data within the file and converts it into a binary vector representing a specific single nucleotide polymorphism. The environmental factors acquisition unit obtains coordinates from the smartphone's location sensor. This unit sends the acquired coordinates to an external meteorological information system to obtain environmental data vectors for the region. This system maps data with these diverse modalities to a common data pattern. This system performs data cleaning processes such as imputation of missing values ​​and removal of outliers. This system constructs a multidimensional patient state tensor from the preprocessed data. Through this processing, this system transforms massive amounts of medical data in various formats into a unified, computer-computable data structure. This system significantly improves computational efficiency and inference accuracy in subsequent AI analysis. This technology is applied to use cases such as building personal health record systems that integrate an individual's lifetime health data.

[0064] Step 2: Accumulation of Time-Series Data Based on Long-Term Memory Management. In this step, the Long-Term Memory Management Department accumulates and manages each patient's medical records, biological data, genetic information, and environmental factors in a unified, time-series manner. For example, this includes data such as blood pressure fluctuations over the past few years, seasonal heart rate variations, blood glucose changes before and after treatment interventions, weight changes before and after lifestyle improvement programs, recovery progress before and after surgery, sleep patterns before and after hospitalization, changes in health indicators accompanying life events such as job changes or relocations, medication adherence, and adverse reactions after vaccinations. All data is recorded in time series, enabling detailed tracking of long-term changes in each patient's health status or changes in health indicators accompanying specific events. Specifically, in this step, the Long-Term Memory Management Department drives a distributed time-series database engine. The department assigns timestamps to the patient state tensors generated in Step 1. The department uses the timestamps as index keys to write the data to the storage cluster. The department applies data compression algorithms to optimize storage capacity during data writing. The department aligns data acquired at different sampling rates along the timeline. This system uses linear or spline interpolation to estimate missing intervals of data, generating continuous time series arrays. To efficiently retrieve long-term data, a multidimensional index tree is constructed. Event markers representing the timing of specific life events or treatment interventions are overlaid on the time series data. A sliding window process is performed to extract data intervals before and after events. The system dynamically updates patients' baseline status using accumulated time series data. The system continuously learns a statistical model in the background to calculate deviations from baseline. Through this processing, the system can seamlessly manage long-term data variations (years in length) and short-term data variations (seconds in length) on the same platform. The system improves data read / write speeds, reducing query response times for large-scale data to the millisecond level. This technology is extremely important in use cases such as generating learning data for chronic disease progression prediction models or managing clinical trial data for long-term efficacy evaluation of new drugs.

[0065] Step 3: Comprehensive Analysis and Risk Prediction Based on Generative AI. In this step, the generative AI processing unit comprehensively analyzes all the data. Generative AI combines natural language processing, image analysis, time series data analysis, pattern recognition, causal reasoning, and generative models to extract disease risks from patients' medical records, detect abnormal patterns from electrocardiogram waveforms or image data, and analyze time series data such as blood pressure or heart rate. It also performs multivariate analysis on genetic factors, lifestyle habits, and environmental factors to predict future disease incidence or symptom exacerbation risks. For example, it extracts individual-specific patterns such as "a tendency for blood pressure to rise every winter," "a tendency for blood sugar levels to rise sharply after taking specific medications," and "a tendency for stress levels to worsen and sleep quality to decrease after changing jobs." As a specific example of a prompt text for generative AI to read, one could be: "Based on the following patient data, predict future heart disease risk and generate a lifestyle improvement plan to reduce risk. Patient data: Age 55, male; Past medical history: hypertension, diabetes; Genetic information: positive for heart disease risk genes; Lifestyle habits: history of smoking, no exercise; Diet: high-fat diet; Latest blood pressure: 150 / 95; Latest blood glucose: 140 mg / dL; Residence: urban area; Occupation: night shift, high stress level." Specifically, in this step, the generative AI processing unit activates a comprehensive inference engine integrating multiple specialized neural networks. This engine includes a Long Short-Term Memory network for processing time-series data, a Convolutional Neural Network for processing image data, and a Transformer model for processing text data. The input data to this engine is a multidimensional time-series tensor extracted from the Long-Term Memory Management Unit, a pixel matrix of medical images, and a token embedding vector of natural language text. As a specific example of the input data, there exists a one-dimensional array representing blood pressure changes over the past month, a two-dimensional tensor of chest X-ray images, and a 768-dimensional embedding representation of the electronic medical record text vectorized from the electronic medical record. This engine inputs the feature vectors output from each network into a combination layer to generate a single comprehensive feature vector. This engine then inputs the comprehensive feature vector into a fully connected layer for nonlinear transformation. This engine applies the Softmax function to calculate the incidence probability distribution of each disease. The output data from this engine consists of a predicted probability score for each disease, an attention weight matrix representing the contribution of risk factors, and natural language text explaining the basis for the prediction. As a specific example of the output data, there exists numerical data showing an 85% incidence probability of myocardial infarction, an attention map assigning high weight to the upward trend in blood pressure, and text explaining that rising blood pressure in winter increases the risk. This system inputs the output probability scores into a risk stratification algorithm to determine the patient's risk level. This system analyzes the attention weight matrix to determine the feature quantity with the highest contribution. Through this processing, this system mathematically elucidates the complex nonlinear correlations between variables that are beyond the comprehension of human doctors.This system significantly improves prediction accuracy compared to existing methods that rely on a single data source. This technology can be applied to use cases such as early screening of high-risk patients in preventative medicine and supporting the determination of optimal treatment strategies in personalized medicine.

[0066] Step 4: Automatic Generation of Individually Optimized Preventive Care Plans. In this step, the Preventive Care Plan Proposal Department automatically generates an optimized preventive care plan for each patient based on the analysis results from the generative AI processing department. For example, a plan combining "aerobic exercise three times a week," "practicing a low-sodium diet," "regular electrocardiogram checkups," and "participating in a stress management program" is proposed for patients at high risk of heart disease; while "calorie-restricted diet," "measuring blood glucose levels after each meal," "strengthening exercise habits," and "ensuring sufficient sleep" are recommended for patients at high risk of diabetes. The plan also considers the patient's living or working environment, hobbies, family structure, economic situation, religious restrictions, etc., and can provide detailed customizations such as "proposing alternative ingredients when certain foods need to be avoided for religious reasons," "proposing a low-cost exercise plan considering economic burden," and "proposing health programs that can be performed with family." As a specific example of a prompt text for generative AI to read, one could be: "Based on the following health risk analysis results for a patient, generate the optimal preventive medical plan in Japanese, taking into account the patient's living environment, hobbies, and economic situation. Patient information: Age 68, female; Past medical history: Osteoporosis; Exercise habits: Weekly walks; Family structure: Living with husband; Economic situation: Relying on pension; Hobbies: Primarily Japanese cuisine; Religious restrictions: Cannot eat certain meats." Specifically, in this step, the preventive medical plan proposal department drives a plan generation agent equipped with a reinforcement learning algorithm. This agent acts within an environment model based on a Markov decision process. The input data to this agent consists of the disease prediction probability score output in step 3, the patient's attribute information vector, and a binary mask matrix representing the constraints. As a specific example of this input data, there is a score indicating a 90% probability of developing diabetes, an attribute vector representing advanced age and low income, and a constraint mask representing specific food allergies. This agent calculates an action value function based on the input data. This agent selects the optimal sequence of actions from multiple available interventions to maximize future health while minimizing costs. The agent then inputs this selected sequence into a natural language generation model. This model converts the sequence into concrete, human-understandable instructions. The output data from this model includes a scheduled task list, structured data of recommended dietary menus, and motivational message text. Specific examples of this output data include stretching tasks to be performed daily at 10 AM, a diet plan with less than two grams of salt, and encouraging text to promote continued practice. The system pushes and distributes the output plan data to the patient's terminal application. The system then initiates a tracking module to monitor the plan's execution. Through this process, instead of providing generic guidelines, the system dynamically synthesizes executable plans perfectly suited to the individual's context. This system improves patient adherence and enhances the effectiveness of preventative medicine.This technology can be applied to use cases such as automating health enhancement programs for enrollees by health insurance portfolios or automatically generating personal training menus in fitness clubs.

[0067] Step 5: Real-time Remote Monitoring and Alarm Notification. In this step, the remote monitoring collaboration unit receives biometric data from the patient's home medical device or wearable device in real time and sends it sequentially to the generative AI processing unit. The generative AI analyzes the received data in real time and automatically sends alarm notifications to healthcare professionals via their terminals when anomalies or increased risks are detected. For example, in cases of a sudden increase in heart rate, blood sugar levels reaching dangerous levels, or sleep apnea detected, the system immediately notifies healthcare professionals and dispatches remote consultations or emergency responses when necessary. A specific example of a prompt message for the generative AI could be: "Analyze the following real-time biometric data. If anomalies or increased risks are detected, generate a notification to healthcare professionals in Japanese along with recommended response strategies. Biometric data: Heart rate 120 bpm, blood pressure 160 / 100, blood sugar 180 mg / dL, sleep apnea event detected." Specifically, in this step, the remote monitoring collaboration unit constructs a stream processing pipeline that utilizes an in-memory database. This unit continuously receives biological data packets arriving in milliseconds. It analyzes the header information of the received packets and extracts numerical data from the payload. The extracted data is stored in a sliding window, and real-time feature quantities are calculated. These calculated feature quantities are input into a lightweight anomaly detection model deployed at the edge. This model employs a one-dimensional convolutional neural network architecture. The input data to this model consists of time-series tensors of heart rate and blood oxygen saturation over the past five minutes. As a specific example of this input data, there exists a heart rate array with 300 sampling points and an equal number of blood oxygen saturation arrays. The model repeatedly applies convolution and pooling operations to the input tensors. The model uses the sigmoid function to output the probability of anomalies. The output data from this model consists of anomaly probability values ​​ranging from zero to one and classification labels representing the types of anomalies. As a specific example of this output data, there exists label data indicating a sleep apnea episode with a probability value of 0.95. If the output probability value exceeds a set threshold, the system immediately initiates an interrupt handler. This handler uses a push notification protocol to send an alert message to the healthcare worker's terminal. This processing procedure also performs an API call to the electronic medical record system to raise an emergency flag. Through this processing, the system minimizes the delay from data reception to anomaly detection and notification. The system ensures no life-threatening deterioration is missed, enabling rapid medical intervention. This technology offers significant value in use cases such as continuous monitoring of patients transferred from the intensive care unit to general wards and systems for preventing the solitary death of elderly people living alone.

[0068] Step 6: Providing Information and Collecting Feedback via the User Interface. In this step, the user interface provides functions such as data input, plan confirmation, alarm reception, and remote consultation participation on both the patient's and healthcare professional's terminals. Patients input their daily health status or lifestyle habits into a smartphone application, and biometric data is automatically sent via wearable devices. Healthcare professionals confirm AI-generated reports or plans on their personal computers and conduct remote consultations with patients when necessary. The explanatory text or plan content is automatically adjusted based on the patient's understanding and implementation status, using easily understandable language for the elderly and providing professional explanations for healthcare professionals, achieving optimal information delivery based on user attributes. Specifically, in this step, the user interface establishes a WebSocket connection that enables bidirectional communication. It asynchronously captures input events from the patient's touchscreen or microphone. The input voice data is fed into a speech recognition engine and converted into text data. The input text data or logs of selection operations are sent to the server as structured objects. The server inputs the received feedback data into a natural language understanding model. This model extracts emotional polarity or dissatisfaction with the plan from the text. The input data consists of free text input by the patient and a vector representing past plan execution history. Specific examples of this input data include text indicating that the exercise was too strenuous to continue, and a numerical vector representing an execution rate of 30%. The model analyzes the input data and scores the degree of plan modification required. The output data from this model includes an emotional score, a modification requirement flag, and a set of parameters for generating alternatives. Specific examples of this output data include a negative emotional score, a flag indicating that modification is enabled, and parameters for reducing exercise intensity. The system feeds back the output parameters to the preventive medical plan proposal department, triggering a recalculation of the plan. The department dynamically redraws the recalculated plan on the terminal screen. Through this processing, the system incorporates the user's reactions into its internal control loop in real time. This system evolves a static information prompting system into an adaptive dialogue system. This technology is applied in use cases such as self-care support applications for patients with chronic diseases and digital therapies aimed at improving medication adherence.

[0069] Step 7: Security Management and Privacy Protection. In this step, the Security Management Department implements encryption for all data communication and data storage, and implements security measures such as anonymization of patient information or access control management, multi-factor authentication, audit log recording, and unauthorized access detection. TLS-based encryption is used for sending and receiving patient data, and AES encryption is implemented for the database on the server. Access permissions are set in detail for each medical practitioner, and third parties are not allowed to view the data without the patient's consent. Security audit logs for the entire system are recorded, and AI automatically detects signs of unauthorized access or information leakage and notifies the administrator. Specifically, in this step, the Security Management Department installs an access control mechanism based on a zero-trust network architecture. The department performs signature verification on all API requests. The department strictly checks the validity period and issuance source of tokens. The department enables transparent data encryption for each column of the database. The department drives a key management service that automates the rotation of encryption keys. The department applies a differential privacy algorithm to patient data export requests. This algorithm mathematically makes it difficult to specify individuals by adding Laplace noise to the dataset. The department mirrors network traffic and inputs it into the intrusion detection system. This system incorporates a deep learning-based packet analysis model. The input data to this model is a feature vector containing packet header information, payload byte sequences, and session statistics. Specific examples of this input data include the source port number, payload entropy, and the number of packets per unit time. The model calculates the deviation from normal traffic patterns. The output data from this model is an alert level indicating signs of malware infection or data escaping, along with a list of IP addresses that should be blocked. Specific examples of this output data include critical alert levels and specific external IP addresses. The system dynamically appends the output IP addresses to the firewall's blacklist, immediately blocking communication. Through this process, the system robustly protects confidential medical data from sophisticated cyberattacks. This system provides a secure data utilization environment while meeting compliance requirements. This technology becomes an essential security foundation in use cases such as data sharing platforms between medical institutions or cloud-based electronic medical record systems.

[0070] (specific use case) For example, when applying the system of the present invention to elderly patients with chronic heart disease, the medical record acquisition unit automatically retrieves the patient's past history of heart disease diagnosis, multiple hospitalizations, cardiac catheterization treatments, medication history, allergy information, past electrocardiogram or echocardiogram image diagnostic data, blood test results, rehabilitation records, etc., from the electronic medical record system. The biometric data acquisition unit retrieves daily blood pressure, heart rate, blood oxygen saturation, weight, steps, sleep time, nighttime heart rate fluctuations, peak heart rate during exercise, respiratory rate, body temperature, etc., from blood pressure monitors, electrocardiographs, pulse oximeters, scales, and wearable devices installed in the patient's home, and sends this data to the server. The Genetic Information Acquisition Department retrieves information from the medical institution's gene testing database, including genes for heart disease risk, drug-metabolizing enzymes risk, diabetes risk, and lipid dyslipidemia risk. The Environmental Factors Acquisition Department collects information from multiple perspectives, including the patient's smoking history, alcohol consumption history, exercise habits, diet, sleep habits, stress levels, medication adherence, local air pollution levels, temperature, humidity, noise, home insulation, visits to medical institutions, and occupational environment (retired, at home during the day, family composition). The Long-Term Memory Management Department comprehensively accumulates all this information chronologically, including detailed records of blood pressure fluctuations over the past five years, seasonal variations in heart rate, changes in blood glucose or weight before and after treatment interventions, recovery progress, sleep patterns before and after hospitalization, changes in health indicators associated with relocation or family composition, changes in medication adherence, and adverse reactions after vaccination. Specifically, in this embodiment, the system enables multiple data collection agents to run in parallel. The Medical Record Acquisition Department connects to the hospital's electronic medical record server using a standard protocol. This department downloads medical records from the past ten years using the patient's identifier as the key. This unit extracts key indicators such as echocardiographic ejection fraction or BNP levels from the downloaded data and stores them in a structured database. The biological data acquisition unit maintains a constant connection with a gateway located in the patient's home. This unit receives millivolt-level potential data from a wearable electrocardiograph at a sampling rate of 100 Hz. This unit applies a bandpass filter to the received waveform data to remove baseline drift or electromyographic noise. This genetic information acquisition unit sends secure API requests to external genomic databases based on the patient's consent profile. This unit extracts allele information related to specific disease risks from the acquired gene sequence data. This environmental factors acquisition unit obtains data from gyroscope or accelerometer sensors via a dedicated application installed on the patient's smartphone to estimate activity levels. This unit obtains time-series data on temperature, humidity, air pressure, and PM2.5 concentration in the patient's residential area from external environmental data providers. This long-term memory management unit maps these heterogeneous data sets onto the same timeline.This system assigns a reliability score to each data point and stores it as a multidimensional time-series tensor in a distributed memory. Through this processing, the system comprehensively digitizes all variables affecting a patient's health status. This system eliminates data silos and builds an inclusive analytical foundation. This technique is highly effective in use cases such as readmission prevention procedures for heart failure patients or inclusive monitoring to prevent the progression of frailty in the elderly.

[0071] The generative AI processing unit comprehensively analyzes all this data. Natural language processing AI extracts disease risk or treatment history from medical records; time series analysis AI detects abnormal patterns or seasonal variations from data such as blood pressure, heart rate, weight, and sleep; image recognition AI extracts abnormalities from electrocardiogram or echocardiogram images; causal inference AI analyzes the correlation between genetic factors, lifestyle habits, environmental factors, and health status; and generative model AI predicts future risk of heart disease recurrence or symptom worsening. For example, it extracts individual-specific patterns such as a tendency for blood pressure to rise in winter, a tendency for weight gain and increased heart rate when there is insufficient exercise, a tendency for blood pressure to worsen during periods of decreased medication adherence, and a tendency for decreased sleep quality and increased heart rate variability after moving. Specifically, in this embodiment, the generative AI processing unit drives a large neural network based on a multimodal learning architecture. This network has independent encoder groups for each modality, processing text, time series numerical data, and images. This natural language processing AI adopts a Transformer-based architecture. The data input to this AI is a vector of text tokenized from the discharge summary of the electronic medical record. This AI extracts the history of heart failure exacerbations or medication changes from text. This time series analysis AI is a hybrid model combining an autoregressive integral moving average model and a long short-term memory network. The input data to this AI is the daily variation tensor of blood pressure, weight, and heart rate over the past six months. This AI extracts patterns of slight weight increases synchronized with increases in heart rate. This image recognition AI employs a residual network architecture. The input data to this AI is the frame image tensor of echocardiogram videos. This AI segments the region of reduced motion in the left ventricular wall at the pixel level. This causal inference AI uses a Bayesian network to model the causal relationships between variables. The input data to this AI is the feature vector group output from each encoder. This AI determines the causal path of decreased temperature and increased salt intake leading to increased blood pressure. This generative model AI integrates these analysis results to calculate future state transition probabilities. The output data from this model are the heart failure recurrence probability score over 30 days, the contribution vector of risk factors, and explanatory text representing the basis of the prediction. As a specific example of the output data, there is a 78% relapse probability, vectors with high weights for weight gain and missed medication, and text indicating warning signs of fluid retention. The system inputs the output probability scores into a threshold determination module to determine the necessity of intervention. Through this processing, the system achieves high-precision risk prediction based on causality rather than simple correlation. This system supplements doctors' empirical judgment, preventing oversights. This technology can be applied to use cases such as chronic disease exacerbation prediction systems or clinical decision support systems that optimize the timing of preventative interventions.

[0072] Based on the analysis results of the generative AI processing unit, the preventive medical plan proposal department automatically generates plans such as "strengthening a low-salt diet and aerobic exercise in winter, recommending walking three times a week," "thoroughly measuring and recording blood pressure every morning, and automatically notifying healthcare professionals in case of persistent abnormal values," "utilizing medication alerts to improve medication adherence," "recommending bedtime stretching or bathing to improve sleep quality," "proposing exercise programs that can be implemented with family members," "prioritizing free or low-cost health programs considering economic burden," and "proposing dietary plans that take into account religious constraints." It also considers individual optimization plans based on the patient's living environment or hobbies, family structure, economic status, and religious constraints. Specifically, in this embodiment, the preventive medical plan proposal department activates an evolutionary algorithm engine for solving constrained optimization problems. This engine inputs the risk factor vector output from the generative AI processing unit and the patient's attribute profile into the objective function. The data input to this engine includes a medical requirement vector for reducing the risk of heart failure recurrence, a categorical variable representing the patient's economic status, and a binary flag representing religious constraints. As a specific example of the input data, there are vectors requiring salt restriction and aerobic exercise, variables representing low-income groups, and a marker indicating that pork consumption is prohibited. The engine generates an initial population consisting of thousands of combinations of intervention actions. The engine repeatedly applies crossover and mutation operators to explore the optimal solution that maximizes the objective function. The optimized action sequence is then input into a large-scale language model. This model uses prompting engineering techniques to generate natural language text that is easy for patients to understand and act upon. The output data from this model consists of a date-specific list of action instructions, recipe data using alternative ingredients, and support guidelines for family members. As a specific example of the output data, there is text instructing patients to weigh themselves immediately upon waking each morning, low-sodium recipe data using chicken, and text prompting family members to confirm medication use. The system synchronously distributes the output plan data to the terminals of patients and their families. The system scores the implementation status of the plan and uses this as input data for the next optimization cycle. Through this process, the system automatically generates plans that meet both medical appropriateness and the patient's lifestyle. This system reduces the enormous time that healthcare practitioners spend manually creating plans. This technology has proven highly effective in use cases such as home care support systems after hospital discharge or in the prevention of severe illnesses caused by lifestyle-related diseases.

[0073] The remote monitoring collaboration unit receives biological data from the patient's home medical devices or wearable devices in real time and sends it sequentially to the generative AI processing unit. The generative AI analyzes the received data in real time, detecting abnormalities such as heart rate exceeding 120 bpm, blood pressure exceeding 160 / 100, respiratory arrest during sleep, and blood glucose exceeding 180 mg / dL. When abnormal values ​​or increased risk are detected, it automatically sends alerts to healthcare professionals via their terminals and automatically dispatches remote diagnostics or emergency responses when necessary. Specifically, in this embodiment, the remote monitoring collaboration unit builds a lightweight publish-subscribe communication network between edge devices and a cloud server. This unit subscribes to ECG waveform data or blood oxygen saturation data sent every second from the patient's wearable device. The unit buffers the received data stream into a time-series database in memory. The unit cuts out the buffered data at a specific window size and forwards it to the generative AI processing unit. The generative AI processing unit performs real-time inference and executes an autoencoder model specialized for anomaly detection. The data input to this model is the ECG waveform tensor and respiratory waveform tensor of the most recent minute. As a specific example of the input data, there is an array of electrocardiograms with 6,000 data points and an equal number of arrays of respiratory waveforms. This model compresses the input tensor into the latent space and restores it to the original waveform. The model calculates the mean square error between the input waveform and the restored waveform. The output data from this model consists of anomaly scores based on the reconstruction error and classification vectors representing the types of anomalies. As a specific example of the output data, there are anomaly scores significantly exceeding a threshold and a classification vector representing ventricular fibrillation. When the system detects that the output anomaly score exceeds an emergency threshold, it immediately activates an event-driven alarm manager. This manager sends a high-priority push notification to the attending physician's smartphone. Simultaneously, the manager automatically sends an API request to the emergency transport system, providing the patient's location information and recent vital signs data. Through this process, the system instantly detects the onset of lethal arrhythmias or acute heart failure and automatically initiates a life-saving chain of events. This system completely eliminates the time lag caused by human monitoring. This technology provides an indispensable safety net in use cases such as home-based end-of-life care for critically ill heart patients or early discharge support procedures for postoperative patients.

[0074] The user interface unit, located on both patient and healthcare professional terminals, allows patients to input their daily health status or lifestyle habits via a smartphone application. This data is automatically transmitted through wearable devices, and healthcare professionals confirm AI-generated reports or plans, enabling remote consultations with patients when necessary. The unit automatically adjusts the explanatory text or plan content based on patient comprehension and implementation status, using easily understandable language for the elderly and providing professional explanations for healthcare professionals, thus achieving optimal information delivery based on user attributes. Specifically, in this embodiment, the user interface unit provides a front-end application based on a responsive programming paradigm. On the patient terminal, the unit renders a screen with large buttons and high-contrast color schemes to support accessibility. The unit maintains the patient's answers to health status questionnaires or the status of medication checks in a status management library. When a status change is detected, the unit immediately performs asynchronous communication with the back-end server. On the healthcare professional terminal, the unit generates a dashboard component displaying real-time data for multiple patients. The unit parses and displays report data created by generative AI. The unit inputs user attribute information into the decision module. The input data to this module consists of the user's age, occupation, and a vector representing their past system usage history. Specific examples of this input data include an 80-year-old age, occupational markers, and a vector representing low usage frequency. This module determines the optimal display mode based on the input data. The output data from this module includes control parameters specifying the display mode and cue variables for adjusting text difficulty. Specific examples of this output data include parameters specifying an elderly mode and cue variables excluding technical terms. The system dynamically reconstructs the screen layout and text presentation based on the output parameters. Through this process, the system provides an optimal user experience for all users, from elderly patients with low IT literacy to specialists seeking advanced information analysis. This system reduces input errors and improves overall data quality. This technology can be applied to use cases such as multi-professional collaborative regional healthcare network systems or patient-participatory personal health record applications.

[0075] The Security Management Department implements encryption for all data communication and storage, and implements security measures such as anonymization of patient information or access control, multi-factor authentication, audit log recording, and unauthorized access detection. Specifically, in this embodiment, the Security Management Department operates an authentication foundation incorporating a zero-knowledge proof protocol. For all communication sessions between the patient's terminal and the server, the department uses the elliptic curve Diffie-Hellman key exchange algorithm to generate temporary session keys. The department uses the generated keys to encrypt the payload. The department applies a salted hash function to convert direct identifiers such as the patient's name or address stored in the database into irreversible tokens. The department evaluates attribute-based access control policies for access requests from healthcare professionals. The department enforces a multi-factor authentication process combining smart card and biometric authentication. The department records all data read and write operations performed within the system in an immutable blockchain-based audit log system. The department drives a machine learning-based intrusion detection agent that continuously monitors network traffic and system logs. The data input to this agent is a feature vector representing the IP address of the access source, the frequency of the request, and the type of data requested. As a specific example of the input data, there are overseas IP addresses that differ from the norm, and data showing a high frequency of requests for large amounts of medical record data within a short period. This agent compares the input vector with the normal configuration file and calculates the Mahalanobis distance. The output data from this agent consists of a score representing the probability of unauthorized access and an identifier for the session. As a specific example of the output data, there is a 99% probability of unauthorized access, and a string representing the session ID. If the output score exceeds a threshold, the system immediately terminates the session and automatically sends an event report to the administrator. Through this process, the system fully meets the stringent privacy protection requirements for medical information. This system minimizes the risk of data leakage due to ransomware attacks or insider crimes. This technology becomes an essential feature in use cases such as cloud-based electronic medical record systems or medical databases that share data across multiple research institutions.

[0076] As a specific example of a prompt text that can be read by generative AI, it could be: "Based on the following patient data, predict the future risk of heart disease recurrence and generate a lifestyle improvement plan to reduce the risk. Patient data: Age 75, male; Past medical history: myocardial infarction, hypertension; Genetic information: positive for heart disease risk genes; Lifestyle habits: history of smoking, no exercise; Diet: high-fat diet; Latest blood pressure: 155 / 98; Latest blood glucose: 145 mg / dL; Residence: urban area; Occupation: retired; Family structure: living with spouse; Economic status: living on pension; Hobbies: mainly Japanese cuisine; Religious restrictions: cannot eat certain meats; Medication adherence rate: 80%; Sleep duration: 5 hours; High stress level; Attached time-series data on blood pressure, heart rate, weight, sleep, exercise, and medication over the past 3 years." Specifically, in this embodiment, the system drives the prompt engineering module to dynamically construct the input text for a large-scale language model. This module performs the process of embedding the structured patient data extracted from the database into a template of natural language. This module concatenates multidimensional data such as patient age, gender, medical history, genetic information, lifestyle habits, latest vital signs, environmental factors, and medication adherence rate as strings. It also converts time-series data from the past three years into text representing statistical summary values ​​or trends, which are then appended to the prompt. As a specific example of the prompt text generated by this module, the following text data was constructed: "Based on the following patient data, predict the future risk of heart disease recurrence and generate a lifestyle improvement plan to reduce the risk. Patient data: Age 75, male; Past medical history: myocardial infarction, hypertension; Genetic information: positive for heart disease risk genes; Lifestyle habits: history of smoking, no exercise; Diet: high-fat diet; Latest blood pressure: 155 / 98; Latest blood glucose: 145 mg / dL; Residence: urban area; Occupation: retired; Family structure: living with spouse; Economic status: living on pension; Hobbies: mainly Japanese cuisine; Religious restrictions: cannot eat certain meats; Medication adherence rate: 80%; Sleep duration: 5 hours; High stress level; Blood pressure, heart rate, weight, sleep, exercise, and medication data for the past three years are appended in time series." The system inputs the constructed prompt text into the encoder of a large-scale language model. This model uses a self-attention mechanism to analyze the context of the input text. The output data from this model is structured text containing a quantitative assessment of relapse risk, a list of specific improvement actions, and messages aimed at enhancing patient motivation. Specific examples of this output data include an assessment of high relapse risk, instructions to limit salt intake to less than six grams per day, and a proposal to walk with one's spouse. The system parses the output text and renders it on a screen via the user interface.This process automates advanced reasoning that integrates complex medical data and patients' social context. The system mimics the thought process of a specialist physician to provide high-quality, personalized medicine. This technology is applicable to use cases such as family physician support systems and patient-centric personalized healthcare applications.

[0077] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0078] (Implementation methods for carrying out the present invention) The embodiments of this invention constitute an advanced nursing support system that collects and integrates diverse and time-series data on the health status, living conditions, environmental factors, nursing history, medication status, and lifestyle habits of the caregiver. It uses generative AI to automatically generate and propose individually optimized preventative care and life support plans, and uniformly implements automatic notifications and emergency responses until remote monitoring or abnormal situations occur. The following provides a more detailed and specific description of each component and its specific functions and processes. Specifically, the nursing support system involved in this embodiment operates as a distributed processing platform built on a cloud computing environment. This system has a data ingestion engine that processes streaming data asynchronously sent from multiple data collection nodes. This system collaborates with an IoT sensor array that acquires the caregiver's biological or environmental information. This system standardizes the collected data and converts it into a multi-dimensional tensor data structure. This system inputs the converted tensor data into a hybrid AI engine that includes a large-scale language model and graph neural networks. The input data to this AI engine consists of the caregiver's past nursing records, real-time vital sign sequences, and sensor data vectors from the living environment. Specific examples of input data include daily report texts entered by home caregivers, heart rate sequences measured at one-minute intervals, and numerical vectors representing indoor temperature and humidity. This AI engine processes the input data in a high-dimensional space to extract early signs of changes in the patient's condition. Output data from this AI engine includes a probability score indicating the risk of future deterioration in care needs, structured data for recommended care intervention plans, and alarm control commands for anomaly detection. Specific examples of output data include an 80% fall risk score, plan data proposing the installation of handrails or additional rehabilitation, and commands to activate the emergency notification system. The system distributes the output plan data to the care manager's terminal and executes automatic notification processing based on alarm control commands. Through this processing, the system fundamentally improves the care planning process from one reliant on past rules of thumb to a data-driven, scientific approach. This system significantly reduces the workload of caregivers while improving the patient's quality of life (QOL) and preventing severe illness. This technology plays a central role in use cases such as resident management systems in special care homes and home care support platforms in regional integrated support centers.

[0079] The Patient Information Acquisition Department automatically retrieves each patient's health records, chronic disease history, fall history, fracture history, hospitalization history, medication information, allergy information, past medical conditions, surgical history, vaccination history, past infectious disease history, rehabilitation records, nutritional status assessment, swallowing function assessment, and pressure ulcer history from a variety of information sources, including electronic medical records, nursing records, home care records, rehabilitation records, medication management applications, nutrition management systems, and pressure ulcer management systems. For example, in cases with a history of heart disease or stroke, detailed records are kept of the diagnosis date, treatment process, medication information, relapse history, rehabilitation progress, and post-discharge observation records. Regarding fall history, the department retrieves the date, time, location, cause, post-fall response, presence of fractures or hospitalization, and the implementation status of fall prevention measures in chronological order. Regarding medication information, the department comprehensively manages the types, dosages, dosing schedules, medication adherence rates, records of missed or incorrect doses, medication change history, side effects, and any drug allergies. In addition, as a nutritional status assessment, serum albumin levels, BMI, weight changes, food intake, presence of oral indigestion, and utilization of tube or parenteral nutrition are obtained. As a swallowing function assessment, swallowing screening test results, history of aspiration pneumonia, and history of changes in dietary patterns are recorded. Specifically, this patient information acquisition department drives a dedicated API gateway for collaboration with multiple external systems. Using secure authentication tokens, this department periodically issues data extraction queries to the databases of the electronic medical record system, the long-term care insurance request system, and home care stations. This department parses the acquired raw data, using a natural language processing model to extract important medical and nursing entities from unstructured text. This department associates the extracted data with the patient's unique identifiers and maps them to tables in the relational database. Regarding past medical histories of heart disease or stroke, this department generates text vectors containing diagnostic codes, onset dates, and treatment details. Regarding fall history, this department generates timestamps for the date and time of occurrence, categorical variables representing the location, and text vectors representing the cause. Regarding medication administration, this system generates drug identification codes, numerical data representing dosage, and percentage data representing medication adherence. For nutritional status assessment, it generates time-series arrays of serum albumin values ​​or BMI. For swallowing function assessment, it generates score vectors for screening tests. This system integrates these generated multi-dimensional data to construct a comprehensive feature tensor encompassing the current state and past history of the person being cared for. Specific examples of input data to this feature tensor include binary flags representing a history of stroke, integer values ​​representing the number of falls in the past year, and floating-point arrays representing BMI changes. This system forwards the constructed feature tensor to the subsequent generative AI processing unit. Through this processing, the system integrates scattered, fragmented nursing and medical information into a single computable data structure.This system prevents information omissions or miscommunications, establishing a foundation for high-precision risk prediction based on AI. This technology is applied in use cases such as an information sharing platform in multiple home care cases involving service provider intervention, and an automated assessment system for new residents.

[0080] The Daily Living Activities Data Acquisition Department acquires data on walking ability (independent, using crutches, walkers, wheelchairs), elimination independence (independent, partially assisted, fully assisted), eating status (independent, assisted, managed nutrition), bathing or dressing independence, getting out of bed or standing independence, toilet mobility independence, stair climbing ability, outdoor mobility, shopping or cooking independence, and money management or medication management independence from caregivers' records or sensors, motion capture devices, bed sensors, toilet sensors or elimination recording applications, etc. For example, accelerometers or gyroscopes are used to detect the risk of swaying or falling while walking; toilet sensors or elimination recording applications automatically record the number of toilet uses or elimination independence; staff input food intake or whether food is being provided on a tablet terminal; GPS devices track outdoor mobility or outing frequency; and family members or staff periodically evaluate money management or medication management independence. Specifically, this acquisition department converts analog signals obtained from various sensors into digital data. This acquisition unit generates time-series arrays of acceleration and angular velocity at a sampling rate of 100 Hz. It applies a low-pass filter to the generated time-series arrays to remove high-frequency noise. The acquisition unit then segments the noise-removed time-series arrays into a window of a specific time size. The acquisition unit constructs a three-dimensional tensor from the segmented data. This constructed three-dimensional tensor is then input into a deep convolutional neural network (DNN). The input to the DNN is a multi-dimensional array with time, sensor, and feature axes. The DNN extracts spatial and temporal features, generating a 256-dimensional feature vector. This acquisition unit inputs the extracted feature vector into a support vector machine (SVM) via a fully connected layer. The SVM outputs a classification label for the walking state. The classification label represents specific states such as normal walking, shuffling, swaying, and fall warning signs. The acquisition unit calculates a fall risk score based on the output probability of the classification label. The fall risk score is a continuous value from zero to one hundred. If the fall risk score exceeds a pre-set safety threshold, the acquisition unit immediately sends a trigger signal to the subsequent warning processing module. Therefore, this system achieves objective and high-precision motion analysis without relying on human visual observation. This system significantly reduces manual recording tasks performed by caregivers. The system also achieves high speed through real-time processing of sensor data and improved accuracy in state determination through machine learning models. This technology is widely used in fall prevention monitoring systems in facilities for the elderly and in automated assessment platforms for motor function in home rehabilitation.

[0081] The vital signs data acquisition unit acquires vital signs data in real time from home medical devices or wearable devices, smartwatches, activity meters, scales, body composition analyzers, blood glucose meters, electrocardiographs, respiratory sensors, sleep sensors, etc., including blood pressure, heart rate, body temperature, blood oxygen saturation, respiratory rate, weight, body fat percentage, blood glucose level, electrocardiogram data, breathing patterns, sleep apnea events, activity level, steps, calories burned, resting heart rate, maximum heart rate, exercise heart rate, and nighttime heart rate fluctuations. For example, it automatically collects vital signs data from various everyday scenarios, such as daily morning blood pressure or heart rate measurements, nighttime heart rate fluctuations, daytime activity or steps, weekend changes in weight and body fat percentage, post-meal blood glucose levels, abnormal respiratory rate or blood oxygen saturation values, abnormal electrocardiogram waveforms, and sleep apnea event detection, and accumulates this data in a time-series manner. Specifically, this acquisition unit receives biosignals asynchronously transmitted from multiple wearable devices via a wireless communication interface. This acquisition unit uses a network time protocol to synchronize the timestamps of received biological signals in milliseconds. It automatically completes missing values ​​in data caused by communication errors using a linear interpolation algorithm. The acquired data is then standardized using a standardization process. This standardized time-series data is input into a Long Short-Term Memory (LSTM) network. This LSM network has pre-learned normal patterns of heart rate, respiratory rate, and blood oxygen saturation over the past seven days. The input to this LSM network is a 128-dimensional time-series feature vector. The LSM network calculates the deviation between the input data and the learned patterns. The output of this LSM network represents a score indicating the probability of anomalies occurring in the next 24 hours. The anomaly probability score is a continuous value from zero to one. If the anomaly probability score exceeds a confidence threshold, the acquisition unit generates an anomaly detection flag. This flag is immediately forwarded to the remote monitoring collaboration unit. Thus, this system captures sudden biological changes in real time. This system can perform objective anomaly detection independently of human rules of thumb or subjective judgment. This system achieves high-precision monitoring by offloading primary processing on the edge device side and deep learning processing on the server side, thus minimizing computational resource consumption. This technology is applied in early detection systems for acute illnesses in telemedicine and in automated screening devices for sleep apnea syndrome.

[0082] The Cognitive Function Assessment Data Acquisition Department acquires regular dementia assessment scores (MMSE, HDS-R, CDR, FAST, MoCA, etc.), AI-automated assessments from daily conversations or behavioral patterns, observation records from family members or individuals, and warning signs of cognitive decline (forgetfulness, wandering, impaired judgment, decreased willingness, aphasia, agnosia, apraxia, disorientation, etc.). For example, it records monthly cognitive function test results, uses speech recognition AI to analyze the content or reaction speed of daily conversations, uses caregiver cameras or sensors to detect wandering or abnormal behavior, and inputs changes noticed by family members or individuals into an application. The AI ​​compares these changes with past cognitive function assessments to automatically determine the degree of progress or trend of change. Specifically, this acquisition department possesses a multimodal artificial intelligence architecture that integrates speech recognition and natural language processing models. The department converts speech waveform data acquired from a microphone into frequency region data using Fast Fourier Transform. The converted data is then extracted as Mel-frequency cepstral coefficients. The extracted acoustic features are input into a speech recognition model based on a recurrent neural network. The speech recognition model outputs the spoken content as high-precision text data. This acquisition unit tokenizes text data through lexical analysis. The tokenized sequence is then input into a Transformer-based natural language processing (NLP) model. The input to the NLP model consists of a 512-dimensional word embedding vector and the sum of positional encodings. The NLP model uses a self-attention mechanism to analyze the context. The NLP model outputs a cognitive function score representing the logical consistency or lexical diversity of speech. This cognitive function score is represented as a multidimensional probability distribution. The acquisition unit then inputs image tensors representing facial expressions or gaze movements from a camera into a convolutional neural network (CNN). The CNN outputs a behavioral feature vector representing emotional state or reaction delay. The acquisition unit combines the cognitive function score with the behavioral feature vector. This combined vector is then input into a gradient boosting decision tree. The gradient boosting decision tree outputs a final evaluation label representing the degree of cognitive function decline. The acquisition unit stores this final evaluation label in the long-term memory management unit. This system supplements regular testing conducted by specialist physicians. This system enables early detection of subtle changes in cognitive function through continuous daily evaluation. This technology is applied to early detection systems for dementia and to algorithms for automatically adjusting individual care plans in nursing facilities.

[0083] The Environmental Factors Acquisition Department collects information such as the location of steps or handrails within a residence, whether the resident lives alone or in a shared home, the presence of pets, local temperature or humidity, noise levels, insulation performance, accessibility to nearby medical facilities, disaster risks, accessibility to the surrounding area, renovation history, family structure, frequency of communication with neighbors, local care system, and the availability of care services in the local government. For example, it acquires information on the location of steps or handrails from residential plans or on-site survey data; automatically acquires environmental data such as temperature, humidity, and noise from IoT sensors or external meteorological databases; obtains information on whether the resident lives alone or in a shared home, family structure, or the presence of pets from a family information database; and acquires information on the frequency of communication with neighbors or the local care system from the database of the local comprehensive support center. Specifically, this acquisition department continuously collects heterogeneous data from a diverse array of environmental sensors and external meteorological application programming interfaces. The department maps the collected data to a common three-dimensional spatial coordinate system. Using distance data obtained from laser sensors, the department generates three-dimensional point cloud data representing the height of steps or the location of handrails within the residence. The department then inputs the three-dimensional point cloud data into a graph neural network. The input to the graph neural network is spatial graph structure data with feature points in space as nodes and physical adjacency relationships as edges. The graph neural network outputs a heat map tensor representing the physical hazards such as falls in each area of ​​the residence. This acquisition unit inputs time-series environmental data such as temperature, humidity, and noise levels into a multilayer perceptron. The multilayer perceptron outputs an environmental stress score representing the risk of heatstroke or sleep disturbance factors. This acquisition unit integrates the heat map tensor and the environmental stress score through tensor combination operations. This acquisition unit generates an integrated environmental risk vector. This acquisition unit sends the environmental risk vector to the subsequent generative artificial intelligence processing unit. Thus, this system comprehensively evaluates the structural risks of the static living environment and the dynamic changes in meteorological conditions. This system quantifies complex environmental risks that are difficult to grasp through human intuition or visual investigation. This system automates the integration process of heterogeneous data, significantly improving computational efficiency. This technology is applied to automated environmental control systems for smart homes for the elderly and building support systems proposing optimal locations for barrier-free renovations.

[0084] The Nursing History Acquisition Department records the utilization of home care or day care, rehabilitation visit history, past nursing accidents or near-miss events, nursing plan change history, nursing staff handover history, nursing service satisfaction evaluations, nursing insurance service utilization, changes in nursing classification, nursing payment performance, nursing consultation history, and requests or complaints from families. For example, it manages detailed usage history for home care visits (3 times per week), day care visits (2 times per month), rehabilitation content or frequency, details of past falls or near-miss events, reasons for or effects of nursing plan changes, handover content during staff transitions, service satisfaction survey results, nursing classification change history, nursing payment performance, and requests or complaints from families. Specifically, this acquisition department converts unstructured text data records or near-miss event reports into machine-readable structured data. This acquisition department uses natural language processing technology to extract named entities from the text data. The extracted named entities are then input into a pre-learned word embedding model. The input to the word embedding model is a tensor containing an array of word indices and contextual information. The word embedding model outputs a 768-dimensional dense vector. This acquisition unit performs cosine similarity calculations on the generated dense vector group. This acquisition unit uses the K-means algorithm to cluster past events with high similarity. Based on the clustering results, this acquisition unit constructs a causal relationship graph representing the effects of specific nursing interventions. This acquisition unit uses a Bayesian network to calculate the transition probabilities between nodes in the causal relationship graph. This acquisition unit updates the graph using the calculated transition probabilities as weights. This acquisition unit stores the updated causal relationship graph in a graph database. Thus, this system automatically extracts useful insights or warning patterns of incidents from a vast amount of past records. This system achieves objective historical analysis independent of nurses' personal experience or memory. This system significantly reduces the computational cost of text retrieval and minimizes information retrieval latency. This technology is applied to a nursing incident prediction and alert system and a data-driven educational support system for new nurses.

[0085] The medication status acquisition unit obtains information from medication support devices or medication record applications, including whether medication was missed or taken incorrectly, medication adherence rate, medication alarm activation status, medication side effects, medication change history, any medication allergies, medication costs, medication storage status, medication remaining quantity management, and medication instruction history. For example, the medication support device records medication times and automatically records cases of missed or incorrect medication. It also manages details such as whether medication alarms activated correctly, side effects after medication changes, automatic notifications when medication remaining quantity decreases, medication cost management, medication storage status, and medication instruction history. Specifically, this acquisition unit analyzes event logs sent from the medication support device in real time. It extracts the medication time, medication type, and medication completion flag from the event logs. The extracted data is converted into a time-series binary matrix. This binary matrix is ​​then input into a Markov chain model. The input to the Markov chain model is a transition probability matrix of medication adherence over the past thirty days and the patient's lifestyle data. The Markov chain model outputs the predicted probability of forgetting to take medication at the next scheduled time. The predicted probability is expressed as a numerical value from zero to 100%. If the predicted probability exceeds a set threshold, the system generates a pre-emptive warning alert. This alert is sent to the patient's terminal and the caregiver's terminal. The system inputs the remaining medication quantity data into the inventory management algorithm. Based on past consumption rates and the predicted medication schedule, the algorithm calculates the optimal reorder date. Thus, this system prevents a decline in medication adherence before it occurs. This system reduces serious health damage caused by overdosing or forgetting to take medication. This system achieves business efficiency through the automation of inventory management and ordering processes. This technology is applied to medication management systems for patients with chronic diseases and prescription optimization modules in online medical platforms.

[0086] The Life Information Acquisition Department obtains information from sensors or applications and personnel records, including food intake or calories consumed, frequency of bowel movements or incontinence, sleep duration or number of awakenings, activity levels or exercise habits, hobbies or social participation, daytime residence or movement history, number of baths, frequency of outings, household chores performed, independence in shopping or cooking, frequency of communication with friends or family, television or radio viewing time, newspaper or book reading time, and participation in religious or community activities. For example, it can record food intake with photos and use AI to estimate nutrients, automatically record toilet usage or incontinence with bowel movement sensors, measure sleep duration or number of awakenings with sleep sensors, obtain step count or exercise intensity with activity meters, record hobbies or frequency of outings with applications, have personnel evaluate household chores performed or independence in shopping and cooking, and analyze the frequency of communication with friends or family using voice recognition AI. Specifically, this acquisition department has a multimodal data processing pipeline that enables image recognition AI and motion recognition AI to collaborate. This acquisition unit resizes and standardizes the image data of eating acquired from the camera. The preprocessed image tensor is then input into a deep convolutional neural network (DNN). The input to the DNN is a 224-pixel square, three-channel image tensor. The DNN segments the food regions within the image pixel by pixel. The DNN outputs structured data representing the estimated weight of each food item and its calorie, protein, and other nutrient content. This acquisition unit also performs a Fast Fourier Transform on the three-axis acceleration data obtained from an activity meter. The acquisition unit inputs the frequency region features into a random forest classifier. The random forest classifier outputs action category labels such as sleep, walking, and housework. This acquisition unit integrates the nutrient content data with the action category labels. The acquisition unit calculates an index representing the balance between daily energy intake and expenditure. This acquisition unit sends the index value to the long-term memory management unit. Thus, this system achieves accurate and objective collection of lifestyle information without relying on self-reporting. This system can accurately grasp lifestyle patterns through the integration of multimodal data. This technology is applied to personalized nutrition guidance systems and activity monitoring platforms for frailty prevention.

[0087] The Long-Term Memory Management Department comprehensively accumulates all this information in a time-series manner, meticulously recording and managing long-term changes in each patient's health status or living conditions, fluctuations in health indicators accompanying specific events (hospitalization, falls, changes in family composition, home renovations, changes in nursing services, disasters, infectious disease outbreaks, etc.), the effects of nursing interventions, seasonal variations or changes in lifestyle, and changes in social isolation or economic status. For example, it meticulously records blood pressure fluctuations over the past five years, changes in cognitive function scores, changes in walking ability or ADL (Activities of Daily Living), functional changes before and after rehabilitation interventions, the correlation between fall history and living environment, changes in medication adherence, seasonal variations in sleep duration or activity levels, changes in health indicators accompanying changes in family composition or nursing services, deterioration of health status during disasters, and changes in lifestyle during infectious disease outbreaks. Specifically, this management department employs a hybrid storage architecture combining a distributed time-series database and a graph database. This management department correlates heterogeneous data received asynchronously from various acquisition departments using a common patient identifier and timestamp. This management department compresses and stores time-series data in a columnar format to optimize storage capacity. This management department represents the relationships between specific events and changes in health indicators as nodes and edges in a graph database. It performs correlation analysis on data in a high-dimensional vector space. Principal component analysis (PCA) is used to reduce the dimensionality of multidimensional data. Anomaly detection using the isolated forest algorithm is performed in the reduced feature space. The anomaly detection algorithm detects singularities that deviate from seasonal variations or long-term changes in lifestyle rhythms. Indexes representing these singularities are provided to the generative artificial intelligence processing unit. Thus, this system balances long-term data accumulation (on a scale of several years) with high-speed retrieval and query processing. It optimizes the physical capacity of data storage while maintaining data integrity. This system achieves comprehensive historical management that surpasses the limits of human memory. This technology can be applied to lifelong health record management platforms or as the foundation for large-scale data analysis in epidemiological surveys.

[0088] The generative AI processing unit comprehensively analyzes all the aforementioned data. Natural language processing AI extracts risk factors or processes from medical and nursing records; time series analysis AI detects changes or abnormal patterns in vital signs, ADL (Activities of Daily Living), cognitive function, and life information; image recognition AI extracts risk factors from dietary content or behavioral patterns and images of the living environment; causal inference AI analyzes the relationship between living environment or nursing history, medication status, and health status; and generative model AI predicts future risks such as fall risk, dementia progression risk, reduced functioning risk, dehydration or infection risk, social isolation risk, and economic hardship risk. For example, AI automatically extracts tendencies toward declining walking ability or cognitive function, changes in medication adherence, seasonal changes in physical condition, correlations between fall history and living environment, associations between sleep deprivation and cognitive decline, associations between social isolation and depressive tendencies, and associations between deteriorating economic conditions and declining nutritional status, learning individual-specific patterns. Specifically, this processing unit possesses an integrated learning architecture that integrates multiple specialized artificial intelligence models. This processing unit inputs time series data extracted from the long-term memory management unit into a Transformer-based time series analysis model. The input to the time series analysis model is a multidimensional time series tensor containing vital signs and activity levels over the past month. The time series analysis model uses a self-attention mechanism to learn long-term dependencies between data. The output of the time series analysis model represents a shift in risk scores indicating future deterioration of health status. This processing unit inputs textual data from medical records into a large-scale language model. The input to the large-scale language model is a text sequence of prompted patient medical history and current symptoms. The large-scale language model outputs a structured list of risk factors for potential complications. This processing unit executes a causal inference algorithm using Bayesian networks. This processing unit calculates the strength of the causal relationship between environmental factors and changes in health status as probability values. This processing unit integrates these outputs and inputs them into a generative model based on generative adversarial networks. The generative model outputs probability distributions for the risks of falls, dementia progression, and social isolation. If the probability distribution exceeds a predetermined warning threshold, this processing unit sends an analysis result vector to the preventive care plan proposal department. Thus, this system achieves a high degree of predictive analysis beyond simple statistical summarization of data. This system quantitatively separates and evaluates complex and intertwined risk factors. It utilizes parallel processing of computational resources to perform large-scale inference tasks at high speed. This technology can be applied to risk stratification systems in preventative medicine or decision support systems for personalized medicine.

[0089] Based on the analysis results from the generative AI processing unit, the Preventive Care Plan Proposal Department automatically generates individually optimized plans that take into account the caregiver's living environment or hobbies, family structure, economic status, religious constraints, social background, etc. Examples include eliminating steps or installing handrails in the living environment, introducing walking training or balance exercises, brain training or dialogue programs to inhibit the progression of dementia, introducing medication support devices, strengthening care through nursing staff, adjusting the lifestyle rhythm to improve sleep, reassessing dietary content or promoting water intake, recommending hobby activities or social participation, strengthening cooperation with family or community, proposing economic support policies, promoting community participation to prevent social isolation, and strengthening the safety confirmation system during disasters. For example, this proposal combines various preventative care and life support strategies, such as strengthening remote care and emergency reporting systems for elderly people living alone; introducing dementia prevention programs when there is a tendency for cognitive decline; strengthening home renovations or walking training when there is a high risk of falls; introducing medication support devices when there are frequent medication forgetfulness issues; proposing sleep improvement programs when there are sleep disorders; utilizing nutritional supplements or dietary support services when food intake is low; recommending community activities or volunteer participation when there is a tendency for social isolation; proposing the utilization of public support systems when there is economic hardship; and strengthening safety confirmation systems or evacuation support strategies during disasters. Specifically, this proposal department has a plan generation engine that combines reinforcement learning algorithms and large-scale language models. This proposal department defines the vector of analysis results received from the generative artificial intelligence processing department as the state space of a Markov decision process. This proposal department sets the history of past care interventions and the degree of improvement in health status achieved as the reward function. This proposal department explores the optimal sequence of intervention actions that maximizes the reward by using a reinforcement learning agent with a Q-learning algorithm. This proposal department inputs the explored sequence of intervention actions into a large-scale language model. The input to the large-scale language model is a sequence of codes containing intervention actions and a context vector containing patient profile information. The large-scale language model outputs text data of an individually optimized plan described in natural language. The output plan includes specific details of recommended home modifications, a walking training schedule, and steps for implementing medication support devices. This proposal department, in conjunction with healthcare practitioners, converts the generated plan into Hypertext Markup Language (HML) format for display on a terminal. The proposed department then sends the converted data from the server to the terminal via a secure communication protocol. Thus, this system automatically generates plans perfectly tailored to the individual patient, rather than a one-size-fits-all approach. This system significantly reduces the time required for nursing managers to create plans. The system continuously learns from past successes to improve plan quality. This technology can be applied to nursing plan creation support systems within long-term care insurance systems or automated construction platforms for rehabilitation programs.

[0090] The remote care collaboration unit receives biometric data and life information in real time from the caregiver's home medical devices or wearable devices, care cameras, fall sensors, door opening / closing sensors, sleep sensors, wandering detection sensors, emergency notification buttons, GPS devices, etc., and sends them sequentially to the generative AI processing unit. The generative AI analyzes the received data in real time and automatically sends alerts to caregivers or family members in case of anomalies, such as a sudden increase in blood pressure or heart rate, fall sensor activation, respiratory arrest detected during sleep, repeated forgetting of medication, detection of wandering or abnormal behavior, door opening / closing sensors activating late at night, or emergency notification button being pressed. It also automatically dispatches remote medical examinations or emergency responses when necessary. For example, if an elderly person living alone falls at night, family members or caregivers are immediately notified, and emergency transport or remote medical examinations are arranged if needed. If wandering is detected, the GPS device can track the location information and notify family members or the regional care network. Specifically, this collaboration unit adopts a distributed processing architecture that integrates edge computing and cloud computing technologies. This collaborative unit performs initial filtering of biological data and lifestyle information at edge gateways deployed near various sensor devices. The noise-removed streaming data is then sent to a stream processing engine on a cloud server. The stream processing engine analyzes the received data in real-time using micro-batch processing. The data from the analyzed object is then input into an anomaly detection algorithm. The input to the anomaly detection algorithm is the heart rate variability over the past five minutes and the three-dimensional vector from the accelerometer. The anomaly detection algorithm calculates the Mahalanobis distance from the normal state. If the Mahalanobis distance exceeds a dynamically set threshold, the anomaly detection algorithm outputs a binary flag indicating an anomaly. Upon receiving the binary flag, this collaborative unit immediately generates an alarm message. This collaborative unit uses a push notification protocol to send the alarm message to the mobile terminal of caregivers. Simultaneously, this collaborative unit invokes the application programming interface (API) of a remote diagnostic system. This collaborative unit automatically establishes a video call session. Thus, the system minimizes delays in human intervention for emergency situations. This system achieves low-latency communication while optimizing network bandwidth consumption. This technology is applicable to emergency notification systems for elderly people living alone or patient monitoring systems in intensive care units.

[0091] Thus, the embodiments of this invention comprehensively manage the diverse and time-series health and life data of each caregiver. With generative AI at its core, it is an extremely practical and highly scalable nursing support system capable of automatically generating individually optimized preventative care, life support plans, real-time remote monitoring, automatic notification of anomalies, emergency response, responses to social isolation or economic hardship, and strengthening safety confirmation mechanisms during disasters. It makes a significant contribution to reducing the burden on nursing care sites, improving caregivers' quality of life (QOL), preventing nursing accidents, promoting integrated nursing care in the region, and addressing social issues. Specifically, this system achieves high scalability and fault tolerance by enabling multiple microservices to collaborate in a distributed computing environment. This system integrates multidimensional data collected from various acquisition units into a central data lake. This system constructs a data pipeline centered on a generative artificial intelligence processing unit. This system automates the entire process of data collection, analysis, prediction, and plan proposal. This system proactively executes the control of physical devices or establishes communication sessions based on the inference results of machine learning models. This system breaks the limits of human cognitive abilities or information processing through manual operations. This system reduces data processing latency to milliseconds. This system reduces unnecessary medical interventions by improving prediction accuracy. It optimizes energy efficiency by dynamically allocating overall system computing resources. This technology has wide applications in the construction of integrated regional care platforms in smart cities or next-generation healthcare information infrastructure.

[0092] (System Composition) The system involved in this embodiment includes a patient information acquisition unit, a daily living activities data acquisition unit, a vital signs data acquisition unit, a cognitive function evaluation data acquisition unit, a living environment factors acquisition unit, a nursing history acquisition unit, a medication status acquisition unit, a life information acquisition unit, a long-term memory management unit, a generative AI processing unit, a preventive nursing plan proposal unit, and a remote care collaboration unit.

[0093] The Patient Information Acquisition Department automatically retrieves each patient's health records, chronic disease history, fall history, fracture history, hospitalization history, medication content, allergy information, past medical conditions, surgical history, vaccination history, infectious disease history, rehabilitation records, nutritional status assessment, swallowing function assessment, and pressure ulcer history from a variety of information sources, including electronic medical records, nursing records, home care records, rehabilitation records, medication management applications, nutrition management systems, and pressure ulcer management systems. For example, in cases with a history of heart disease or stroke, detailed records are kept of the diagnosis date, treatment process, medication, recurrence history, rehabilitation process, and post-discharge observation records. Regarding fall history, records are obtained chronologically, including the date, time, location, cause, post-fall response, presence of fractures or hospitalization, and the implementation status of fall prevention measures. Regarding medication, comprehensive management is maintained of prescription drug types, dosages, medication schedules, medication adherence rates, records of missed or incorrect doses, medication change history, side effects, and any drug allergies. For nutritional status assessment, serum albumin levels, BMI, weight changes, dietary intake, presence of oral indigestion, and utilization of managed or intravenous nutrition are obtained. For swallowing function assessment, swallowing screening test results, history of aspiration pneumonia, and changes in dietary patterns are recorded. Specifically, this acquisition unit possesses a data crawler that asynchronously extracts data from multiple heterogeneous databases. This acquisition unit applies a natural language processing pipeline to the extracted text data. This acquisition unit performs morphological analysis and syntactic analysis. This acquisition department standardizes synonyms using a medical terminology dictionary. The standardized text data is then input into the information extraction model. The input to the information extraction model is a sequence of labeled electronic medical record texts. The model outputs disease name, onset date, and treatment details as key-value pairs. The acquisition department uses the extracted structured data to construct a knowledge graph. Patients, diseases, and medications are configured as nodes in the knowledge graph. Relationships between nodes are defined as edges. The knowledge graph is stored in a graph database. Subsequent analysis and processing involve querying the knowledge graph. Thus, this system transforms diverse medical records with varying formats into a unified data structure. This system significantly improves data retrieval speed and reference efficiency. It completely eliminates data input errors caused by manual processing. This technology is applied to improve the interoperability of electronic medical record systems or to build the foundation for medical big data analysis.

[0094] The Daily Living Activities Data Acquisition Department acquires data on walking ability (independent, using crutches, walkers, wheelchairs), elimination independence (independent, partially assisted, fully assisted), dietary intake (independent, assisted, managed nutrition), bathing or dressing independence, getting out of bed or standing independence, toilet mobility independence, ability to climb stairs, outdoor mobility, shopping or cooking independence, and money management or medication management independence from caregivers' records or sensors, motion capture devices, bed sensors, toilet sensors or elimination recording applications, etc. For example, accelerometers or gyroscopes are used to detect the risk of swaying or falling while walking; toilet sensors or elimination recording applications automatically record the number of toilet uses or elimination independence; staff input dietary intake or whether dietary assistance is provided on a tablet terminal; GPS devices track outdoor mobility or outing frequency; and family members or staff periodically evaluate money management or medication management independence. Specifically, this acquisition department converts analog signals obtained from various sensors into digital data. This acquisition unit generates time-series arrays of acceleration and angular velocity at a sampling rate of 100 Hz. It applies a low-pass filter to the generated time-series arrays to remove high-frequency noise. The acquisition unit then segments the noise-removed time-series arrays into windows of a specific time size. The acquisition unit constructs a three-dimensional tensor from the segmented data. This three-dimensional tensor is then input into a deep convolutional neural network (DNN). The input to the DNN is a multi-dimensional array with time, sensor, and feature axes. The DNN extracts spatial and temporal features, generating a 256-dimensional feature vector. The acquisition unit inputs the extracted feature vector into a support vector machine (SVM) via a fully connected layer. The SVM outputs a classification label for the walking state. This label represents specific states such as normal walking, shuffling, swaying, and signs of impending fall. Based on the output probability of the classification label, the acquisition unit calculates a fall risk score. The fall risk score is a continuous value from zero to one hundred. If the fall risk score exceeds a pre-set safety threshold, the acquisition unit immediately sends a trigger signal to the subsequent warning processing module. Therefore, this system achieves objective and high-precision motion analysis without relying on human visual observation. This system significantly reduces manual recording work by caregivers. Simultaneously, it achieves high-speed real-time processing of sensor data and improved accuracy in state determination through machine learning models. This technology is widely used in fall prevention monitoring systems in facilities for the elderly or as an automated assessment platform for motor function in home rehabilitation.

[0095] The vital signs data acquisition unit acquires vital signs data in real time from home medical devices or wearable devices, smartwatches, activity meters, scales, body composition analyzers, blood glucose meters, electrocardiographs, respiratory sensors, sleep sensors, etc., including blood pressure, heart rate, body temperature, blood oxygen saturation, respiratory rate, weight, body fat percentage, blood glucose level, electrocardiogram data, breathing patterns, sleep apnea events, activity level, steps, calories burned, resting heart rate, maximum heart rate, exercise heart rate, and nighttime heart rate variability. For example, it automatically collects vital signs data from various daily life scenarios, such as daily morning blood pressure or heart rate measurements, nighttime heart rate variability, daytime activity level or steps, weekend changes in weight or body fat percentage, postprandial blood glucose levels, abnormal respiratory rate or blood oxygen saturation values, abnormal electrocardiogram waveforms, and detection of sleep apnea events, and accumulates this data in a time series. Specifically, this acquisition unit receives biosignals asynchronously transmitted from multiple wearable devices via a wireless communication interface. This acquisition unit uses a network time protocol to synchronize the timestamps of received biological signals in milliseconds. It automatically fills in missing values ​​in the data due to communication errors using a linear interpolation algorithm. The acquisition unit then normalizes the supplemented time-series data through standardization processing. This normalized time-series data is input into a Long Short-Term Memory (LSTM) network. This LSM network has pre-learned normal fluctuation patterns of heart rate, respiratory rate, and blood oxygen saturation over the past seven days. The input to this LSM network is a 128-dimensional time-series feature vector. The LSM network calculates the deviation of the input data from the learned patterns. The output of this LSM network represents a score indicating the probability of anomalies occurring in the next 24 hours. The anomaly probability score is a continuous value from zero to one. If the anomaly probability score exceeds a confidence threshold, the acquisition unit generates an anomaly detection flag. This flag is immediately forwarded to the remote monitoring collaboration unit. Thus, this system captures sudden biological changes in real time. This system can perform objective anomaly detection without relying on human rules of thumb or subjective judgment. This system achieves high-precision monitoring by offloading primary processing on the edge device side and deep learning processing on the server side, thus minimizing computational resource consumption. This technology can be applied to early detection systems for acute illnesses in telemedicine or to automated screening devices for sleep apnea syndrome.

[0096] The Cognitive Function Assessment Data Acquisition Department acquires regular dementia assessment scores (MMSE, HDS-R, CDR, FAST, MoCA, etc.), AI-automated assessments from daily conversations or behavioral patterns, observation records from family members or individuals, and signs of cognitive decline (forgetfulness, wandering, impaired judgment, decreased willingness, aphasia, agnosia, apraxia, disorientation, etc.). For example, it records monthly cognitive function test results, uses speech recognition AI to analyze the content or reaction speed of daily conversations, uses caregiver cameras or sensors to detect wandering or abnormal behavior, and inputs changes noticed by family members or individuals into an application. The AI ​​compares these changes with past cognitive function assessments to automatically determine the degree of progress or trend of change. Specifically, this acquisition department possesses a multimodal artificial intelligence architecture that integrates speech recognition and natural language processing models. The department converts speech waveform data acquired from the microphone into frequency domain data using Fast Fourier Transform. The converted data is then extracted as Mel-frequency cepstral coefficients. The extracted acoustic features are input into a speech recognition model based on a recurrent neural network. The speech recognition model outputs the spoken content as high-precision text data. This acquisition unit tokenizes text data through morphological analysis. The tokenized sequence is then input into a Transformer-based natural language processing (NLP) model. The input to the NLP model consists of a 512-dimensional word embedding vector and the sum of positional encodings. The NLP model uses a self-attention mechanism to analyze the context. The NLP model outputs a cognitive function score representing the logical consistency or lexical diversity of speech. This cognitive function score is represented as a multidimensional probability distribution. The acquisition unit then inputs image tensors representing facial expressions or gaze movements captured from a camera into a convolutional neural network (CNN). The CNN outputs behavioral feature vectors representing emotional states or reaction delays. The acquisition unit combines the cognitive function score and the behavioral feature vectors. The combined vector is then input into a gradient boosting decision tree. The gradient boosting decision tree outputs a final evaluation label representing the degree of cognitive function decline. This final evaluation label is stored in the long-term memory management unit. Thus, this system supplements regular testing conducted by specialist physicians. This system enables early detection of subtle changes in cognitive function through continuous daily evaluation. This technology is applied to algorithms for automatically adjusting individual care plans in early detection systems for dementia or in care facilities.

[0097] The Environmental Factors Acquisition Department collects information such as the location of steps or handrails within a residence, whether the resident lives alone or in a shared home, the presence of pets, local temperature or humidity, noise levels, insulation performance, accessibility to nearby medical facilities, disaster risks, accessibility, renovation history, family composition, frequency of communication with neighbors, local care system, and the availability of care services in the local government. For example, it acquires information on the location of steps or handrails from residential plans or on-site survey data; automatically acquires environmental data such as temperature, humidity, and noise from IoT sensors or external meteorological databases; obtains information on whether the resident lives alone or in a shared home, family composition, or the presence of pets from family information databases; and acquires information on the frequency of communication with neighbors or the local care system from the database of the local comprehensive support center. Specifically, this acquisition department continuously collects heterogeneous data from a diverse array of environmental sensors and external meteorological application programming interfaces. The department maps the collected data to a common three-dimensional spatial coordinate system. Using distance data obtained from laser sensors, the department generates three-dimensional point cloud data representing the height of steps or the location of handrails within the residence. The department then inputs the three-dimensional point cloud data into a graph neural network. The input to the graph neural network is spatial graph structure data with feature points in space as nodes and physical adjacency relationships as edges. The graph neural network outputs a heat map tensor representing the physical hazards such as falls in each area of ​​the residence. This acquisition unit inputs time-series environmental data such as temperature, humidity, and noise levels into a multilayer perceptron. The multilayer perceptron outputs an environmental stress score representing the risk of heatstroke or sleep disturbance factors. This acquisition unit integrates the heat map tensor and the environmental stress score through tensor combination operations. This acquisition unit generates an integrated environmental risk vector. This acquisition unit sends the environmental risk vector to the subsequent generative artificial intelligence processing unit. Thus, this system comprehensively evaluates the structural risks of the static living environment and the dynamic changes in meteorological conditions. This system quantifies complex environmental risks that are difficult to grasp through human intuition or visual investigation. This system automates the integration and processing of heterogeneous data, significantly improving computational efficiency. This technology can be applied to automated environmental control systems for smart homes for the elderly or building support systems proposing optimal locations for accessibility modifications.

[0098] The Nursing History Acquisition Department records the utilization of home care or day care, rehabilitation visit history, past nursing accidents or near-miss events, nursing plan change history, nursing staff turnover history, nursing service satisfaction evaluations, nursing insurance service utilization, changes in nursing classification, nursing payment performance, nursing consultation history, and requests or complaints from families. For example, it manages detailed usage history for home care visits (3 times per week), day care visits (2 times per month), rehabilitation content or frequency, details of past falls or near-miss events, reasons for or effects of nursing plan changes, handover details during staff turnover, service satisfaction survey results, nursing classification change history, nursing payment performance, and requests or complaints from families. Specifically, this acquisition department converts unstructured text data records or near-miss event reports into machine-readable structured data. This acquisition department uses natural language processing technology to extract named entities from the text data. The extracted named entities are then input into a pre-learned word embedding model. The input to the word embedding model is a tensor containing a sequence of word indices and contextual information. The word embedding model outputs a 768-dimensional dense vector. This acquisition unit performs cosine similarity calculations on the generated dense vector clusters. It then uses the K-means algorithm to cluster past events with high similarity. Based on the clustering results, it constructs a causal relationship graph representing the effects of specific nursing interventions. A Bayesian network is used to calculate the transition probabilities between nodes in the causal relationship graph. These calculated transition probabilities are used as weights to update the graph. The updated causal relationship graph is stored in a graph database. Thus, this system automatically extracts useful insights or warning patterns of incidents from a vast amount of past records. This system achieves objective historical analysis independent of nurses' personal experience or memory. It significantly reduces the computational cost of text retrieval and minimizes information retrieval latency. This technology can be applied to predictive alert systems for nursing incidents or data-driven educational support systems for new nurses.

[0099] The medication status acquisition unit obtains information from medication support devices or medication record applications, including whether medication was missed or taken incorrectly, medication adherence rate, medication alarm activation status, medication side effects, medication change history, any medication allergies, medication costs, medication storage status, medication remaining quantity management, and medication instruction history. For example, the medication support device records medication times and automatically records cases of missed or incorrect medication. It also manages details such as whether medication alarms activated correctly, side effects after medication changes, automatic notifications when medication remaining quantity decreases, medication cost management, medication storage status, and medication instruction history. Specifically, this acquisition unit analyzes event logs sent from the medication support device in real time. It extracts the medication time, medication type, and medication completion flag from the event logs. The extracted data is converted into a time-series binary matrix. This binary matrix is ​​then input into a Markov chain model. The input to the Markov chain model is a transition probability matrix of medication adherence over the past thirty days and the patient's lifestyle data. The Markov chain model outputs a predicted probability of missed medication at the next scheduled dose. This predicted probability is represented as a numerical value from zero to 100%. If the predicted probability exceeds a set threshold, the system generates a pre-emptive warning alert. This alert is sent to both the patient's and caregiver's terminals. The system inputs remaining medication data into an inventory management algorithm. Based on past consumption rates and predicted medication schedules, the algorithm calculates the optimal reorder date. Thus, the system proactively prevents a decline in medication adherence. This system reduces serious health damage caused by overdosing or missed medication. This system achieves operational efficiency through the automation of inventory management and ordering processes. This technology is applied to medication management systems for patients with chronic diseases or to prescription optimization modules in online medical platforms.

[0100] The Life Information Acquisition Department obtains information from sensors or applications and personnel records, including food intake or calories consumed, frequency of excretion or incontinence, sleep duration or number of awakenings, activity level or exercise habits, hobbies or social participation, daytime residence or movement history, number of baths, frequency of outings, household chores performed, independence in shopping or cooking, frequency of communication with friends or family, television or radio viewing time, newspaper or book reading time, and participation in religious or community activities. For example, it can record food intake with photos and use AI to estimate nutrients; use excretion sensors to automatically record toilet usage or incontinence; use sleep sensors to measure sleep duration or number of awakenings; use activity meters to obtain steps or exercise intensity; use applications to record hobbies or frequency of outings; allow personnel to evaluate household chores performed or independence in shopping and cooking; and use voice recognition AI to analyze the frequency of communication with friends or family. Specifically, this Life Information Acquisition Department collects raw data asynchronously transmitted from multiple heterogeneous sensor groups and portable terminals. The Life Information Acquisition Unit inputs the collected raw data into the normalization processing module. This module maps image data, time-series waveform data, and natural language text data into different feature spaces. The Life Information Acquisition Unit inputs the image data of eating into a convolutional neural network. This convolutional neural network segments the regions of food from the input image tensor. It outputs the estimated weight and nutrient composition of each food item as multi-dimensional vectors. The Life Information Acquisition Unit inputs time-series data representing changes in electrostatic capacitance obtained from excretion sensors into a recurrent neural network. This recurrent neural network outputs the probability of excretion events as a score. The Life Information Acquisition Unit uses a speech recognition module to convert spoken dialogue into text data. This speech recognition module inputs the converted text data into a large-scale language model. This large-scale language model outputs the emotional polarity and communication quality of the dialogue as numerical vectors. This system stores these multi-dimensional feature vectors along with timestamps in a unified database. Through this structure, this system quantifies the collection of life information that relies on human subjective evaluation. This system significantly improves the accuracy of data collection. This system significantly reduces the workload of caregivers in record-keeping. This technology is applied to the daily monitoring of elderly care facilities or the objective evaluation of the pace of life in home care.

[0101] The Long-Term Memory Management Department systematically accumulates all this information in a time-series manner, meticulously recording and managing long-term changes in each caregiver's health status or living conditions, variations in health indicators accompanying specific events (hospitalization, falls, changes in family composition, home renovations, changes in nursing services, disasters, infectious disease outbreaks, etc.), the effects of nursing interventions, seasonal variations or changes in lifestyle, changes in social isolation or economic status, etc. For example, it meticulously records the past 5 years' blood pressure fluctuations, changes in cognitive function scores, changes in walking ability or ADL, functional changes before and after rehabilitation interventions, the correlation between fall history and living environment, changes in medication adherence, seasonal variations in sleep duration or activity levels, changes in health indicators accompanying changes in family composition or nursing services, deterioration of health status during disasters, and changes in lifestyle during infectious disease outbreaks. Specifically, this Long-Term Memory Management Department uses a distributed graph database to construct multidimensional time-series data. The department sets the caregiver's identifier as the root node. It generates a graph structure that combines various health indicators and life events using edges. The department synchronizes data with different sampling rates along the timeline. This long-term memory management system applies spline interpolation algorithms to missing values. It segments past records using fixed-length time windows. The segmented time-series data is then input into an autoregressive moving average model and a long short-term memory (LSM) network. The LSM network extracts hidden state vectors from the input time-series data. It outputs feature tensors representing seasonal variations or long-term trends. The system uses the occurrence time of a specific event as a trigger to extract data intervals before and after it. It calculates the statistical significance difference between the extracted data intervals. This system uses graph-structured time-series data to expedite the inference of causal relationships. It improves the retrieval speed of complex correlations between events that are difficult to obtain in existing relational databases. Through this structure, the system quantitatively measures the long-term effects of nursing interventions. This technology can be applied to systems predicting the progression of chronic diseases or assessing the effectiveness of rehabilitation.

[0102] The generative AI processing unit comprehensively analyzes all the above data. Natural language processing AI extracts risk factors or processes from medical and nursing records; time series analysis AI detects changes or abnormal patterns in vital signs, ADL (Activities of Daily Living), cognitive function, and life information; image recognition AI extracts risk factors from images of food content or behavioral patterns and the living environment; causal inference AI analyzes the relationship between living environment or care history, medication status, and health status; and generative model AI predicts future risks such as fall risk, dementia progression risk, reduced functioning risk, dehydration or infection risk, social isolation risk, and economic hardship risk. For example, AI automatically extracts the tendency for walking ability or cognitive function to decline, changes in medication adherence, seasonal changes in physical condition, the correlation between fall history and living environment, the association between sleep deprivation and cognitive decline, the association between social isolation and depressive tendencies, and the association between deteriorating economic conditions and reduced nutritional status, and learns individual-specific patterns. As a specific example of a prompt text that the generative AI can read, one could be: "Based on the following caregiver data, predict the risk of falls and the risk of dementia progression, and generate a life support and preventative care plan to reduce these risks. Caregiver data: Age 83, female; Past medical history: osteoporosis, hypertension, history of falls; ADL: Walking ability 2 (using a cane); Toileting independence: partially assisted; Dementia assessment: MMSE 18; Vital signs: Blood pressure 155 / 95, heart rate 102, Sleep duration 4 hours, missed medication twice a week; Living environment: many steps, living alone; Care history: home care three times a week; Life information: reduced food intake, decreased activity level; all data from the past year are appended in time series." Specifically, this generative AI processing unit employs an integrated learning architecture that combines multiple specialized neural networks. This generative AI processing unit includes a natural language processing module, a time series analysis module, an image recognition module, and a causal inference module. This natural language processing module installs a large-scale language model based on Transformer. It tokenizes textual data from medical records to generate input tensors. It uses a self-attention mechanism to compute contextual dependencies. Its output is structured data extracting entities representing risk factors. The time series analysis module inputs time series data of vital signs into a one-dimensional convolutional neural network. This one-dimensional convolutional neural network outputs the probability of abnormal patterns as a score. The causal inference module uses a Bayesian network to compute the conditional probability between environmental data and health status data. The generative AI processing unit inputs feature vectors from these modules into a consolidation layer. This consolidation layer inputs the integrated multidimensional feature tensor into the generative model. This generative model is constructed using a generative adversarial network or a diffusion model. The generative model outputs a probability distribution representing the progression of future health status.This generative model outputs the risk of falls or the risk of dementia progression as a continuous score from 0 to 1. The system compares the output risk score to a predetermined threshold. If the threshold is exceeded, the system triggers subsequent planning proposal processing. Through this structure, the system identifies subtle correlations in high-dimensional spaces that are beyond the reach of human intuition. This system significantly improves the accuracy of risk prediction compared to existing methods. This technology can be applied to personalized risk assessment in preventative medicine or to optimal resource allocation systems in care facilities.

[0103] Based on the analysis results of the generative AI processing department, the Preventive Care Plan Proposal Department automatically generates plans such as eliminating steps or installing handrails in the living environment, introducing walking training or balance exercises, brain training or dialogue programs to inhibit the progression of dementia, introducing medication support devices, strengthening care through nursing staff, adjusting the rhythm of life to improve sleep, reassessing food content or promoting water intake, recommending hobby activities or social participation, strengthening collaboration with family or community, proposing economic support policies, promoting community participation to prevent social isolation, and strengthening the safety confirmation system during disasters. It also takes into account the individualized optimization of the caregiver's living environment or hobbies, family composition, economic status, religious constraints, social background, etc. For example, we can strengthen remote care and emergency reporting systems for elderly people living alone; introduce dementia prevention programs when there is a tendency for cognitive decline; strengthen home renovation or walking training when there is a high risk of falls; introduce medication support devices when there are frequent cases of forgetting to take medication; propose sleep improvement programs when there are sleep disorders; make use of nutritional supplements or eating support services when food intake is low; recommend community activities or volunteer participation when there is a tendency for social isolation; propose the use of public support systems when there is economic hardship; and strengthen safety confirmation systems or evacuation support policies during disasters. We can combine and propose a variety of preventive care and life support policies. As a specific example of a prompt text that the generative AI can read, one could be: "Based on the following health, living, and environmental data of the caregiver, generate a specific preventive care and life support plan in Japanese to reduce the risk of falls and the risk of dementia progression. Data: Age 80, male; Past medical history: diabetes, hypertension, history of falls; ADL: walking ability 3 (using a walker); toileting independence: independent; dementia assessment: MMSE 20; vital signs: blood pressure 150 / 90, heart rate 95, sleep duration 5 hours, missed medication once a week; living environment: few steps, cohabitation; care history: day care twice a week; life information: slightly reduced activity level, interested in activities; all data from the past 2 years are attached in time series." Specifically, this preventive care plan proposal department has an inference engine equipped with a reinforcement learning algorithm. The preventive care plan proposal department defines the state space from the risk score output by the generative AI processing department and the caregiver's profile vector. The preventive care plan proposal department defines the action space from the list of available care services or life support policies. This preventive care plan proposal department conducts simulations based on Markov decision processes. The department calculates the expected return for each combination of support strategies. This expected return is evaluated using an objective function that optimizes the trade-off between risk reduction effectiveness and cost. The department selects the combination of support strategies that maximizes the expected return as the optimal strategy. The selected optimal strategy is then input into a large-scale language model.This large-scale language model converts the parameter set of the optimal strategy into text data in natural language. The model outputs individually optimized plans in a form easily understood by caregivers or families. The system sends the output plans to the display control unit of a healthcare professional's terminal. Through this architecture, the system instantly derives computationally optimal solutions from a vast array of nursing services. This system standardizes the plan-making process, which previously relied on the personal experience of nursing managers. The system dramatically reduces the computation time required for plan creation. This technology can be applied to automated nursing plan creation support systems in public long-term care insurance systems or to health improvement program proposal systems in private insurance companies.

[0104] The remote care collaboration department receives biometric data and life information in real time from the caregiver's home medical devices or wearable devices, care cameras, fall sensors, door opening and closing sensors, sleep sensors, wandering detection sensors, emergency notification buttons, GPS devices, etc., and sends them sequentially to the generative AI processing department. The generative AI analyzes the received data in real time, automatically sending alerts to caregivers or family members in case of anomalies such as a sudden increase in blood pressure or heart rate, fall sensor activation, respiratory arrest detected during sleep, repeated forgetting of medication, wandering or abnormal behavior, door opening and closing sensor activation at night, or pressing of the emergency notification button. It also automatically dispatches remote medical examinations or emergency responses when necessary. For example, if an elderly person living alone falls at night, family members or caregivers are immediately notified, and emergency transport or remote medical examinations can be arranged if needed. If wandering is detected, GPS tracking can be used to notify family members or the local care network. As a specific example of a prompt message that generative AI can read, one could say, "Analyze the following real-time biological data and life information. If an anomaly or increased risk is detected, generate in Japanese the content to be notified to caregivers or family members, along with recommended coping strategies. Data: Blood pressure 160 / 100, heart rate 110, fall sensor activation, sleep time 3 hours, two consecutive missed medications, wandering sensor activation, and append all data from the past week in time series." Specifically, this remote care collaboration unit employs a hybrid processing architecture integrating edge computing and cloud computing technologies. This remote care collaboration unit asynchronously receives raw data from various sensor devices using a message queue protocol. The received streaming data is temporarily buffered in an in-memory database. The buffered time-series data is input into a lightweight neural network for anomaly detection. This lightweight neural network has an autoencoder structure. This lightweight neural network compresses the input time-series data into a low-dimensional latent space. This lightweight neural network restores the compressed data to its original dimensions. This lightweight neural network calculates the reconstruction error between the input data and the reconstructed data. The remote monitoring collaboration unit determines an anomaly when the reconstruction error exceeds a dynamically set threshold. This unit immediately transfers the flag data indicating the anomaly and the corresponding sensor data to the generative AI processing unit. The generative AI processing unit analyzes the transferred data and outputs a score indicating the urgency level and text data indicating the recommended response. Based on the output urgency score, the system performs notification routing. In cases of high urgency, the system automatically executes an API call to the emergency medical services system. Through this architecture, the system minimizes the communication bandwidth consumption associated with constant monitoring. The system reduces the delay from anomaly occurrence to notification to milliseconds.This technology has been applied to systems for preventing the death of elderly people living alone or to patient monitoring systems in intensive care units in hospitals.

[0105] In this embodiment, the system involves collaborative efforts across various departments to comprehensively manage and analyze diverse medical, health, living, and environmental data of each patient. It places generative AI at its core, enabling highly practical and scalable nursing support systems such as individually optimized preventative care, automated generation of life support plans, real-time remote monitoring, automatic notification of anomalies, emergency response, handling of social isolation or economic hardship, and enhanced safety verification mechanisms during disasters. Specifically, this system is installed as a distributed processing system where multiple independent microservices coordinate their operations on a container orchestration platform. The system employs the gRPC protocol for data communication between processing departments. This system accelerates data serialization processing. This system features automatic scaling capabilities to dynamically add computing nodes to accommodate the increasing computational load associated with an increase in the number of patients. This system offloads the inference processing of generative AI to a dedicated cluster of tensor processing units. This system parallelizes the forward propagation computation of deep learning models. This system maximizes the overall processing throughput of the system. Through this architecture, the system transcends mere digitization of operations, realizing a new computer paradigm of data-driven predictive care intervention. The system computationally ensures optimal allocation of medical and care resources. It achieves the significant technological effect of reducing overall societal care costs. This technology can be applied to national-scale healthcare information infrastructure systems or elderly support platforms in smart cities.

[0106] (Implementation steps) Step 1: Collection of Diverse Patient Data. In this step, the Patient Information Acquisition Department automatically acquires health records, chronic disease history, fall history, fracture history, hospitalization history, medication content, allergy information, past medical conditions, surgical history, vaccination history, infectious disease history, rehabilitation records, nutritional status assessment, swallowing function assessment, and pressure ulcer history from a variety of information sources, including electronic medical records, nursing records, home care records, rehabilitation records, medication management applications, nutrition management systems, and pressure ulcer management systems. The Activities of Daily Living (ADL) Data Acquisition Department acquires data on walking ability and excretion from records kept by caregivers or sensors, motion capture devices, bed sensors, toilet sensors, door opening and closing sensors, GPS devices, etc. Independence levels include: eating habits, independence in bathing or dressing, independence in getting out of bed or standing up, independence in moving to the toilet, ability to climb stairs, outdoor mobility, independence in shopping or cooking, and independence in money management or medication management; The vital signs data acquisition department acquires real-time data on blood pressure, heart rate, body temperature, blood oxygen saturation, respiratory rate, weight, body fat percentage, blood glucose levels, electrocardiogram data, sleep apnea events, activity level, steps, calories burned, and resting heart rate from home medical devices or wearable devices, smartwatches, activity meters, scales, body composition analyzers, blood glucose meters, electrocardiographs, ECG machines, respiratory sensors, and sleep sensors. The data acquisition department obtains data such as maximum heart rate, heart rate during exercise, and nocturnal heart rate fluctuations; the cognitive function assessment data acquisition department obtains regular dementia assessment scores, AI-automated assessments from daily conversations or behavioral patterns, observation records from family members or individuals, and signs of cognitive decline; the living environment factors acquisition department obtains information such as the condition of steps or handrails in the residence, whether the person lives alone or with others, whether there are pets, the temperature or humidity of the residence, noise levels, the insulation performance of the residence, visits to nearby medical institutions, disaster risks, accessibility in the surrounding area, history of home renovations, family composition, frequency of communication with nearby residents, the local care system, and the availability of nursing services in the local government; the nursing history acquisition department... Record the utilization status of home care or day care, rehabilitation and medical history, past nursing accidents or near-miss events, history of changes in nursing plans, history of nursing staff turnover, satisfaction evaluation of nursing services, utilization status of nursing insurance services, changes in nursing identification classification, nursing payment performance, nursing consultation history, requests or complaints from family members, etc.; The medication status acquisition department obtains information from medication support equipment or medication record applications on whether medication was forgotten or taken incorrectly, medication adherence rate, medication alarm activation status, occurrence of medication side effects, medication change history, presence of medication allergies, medication costs, medication storage status, management of remaining medication, medication guidance history, etc.The Daily Living Information Acquisition Unit obtains information from sensors or applications and personnel records, including food intake or calories consumed, frequency of excretion or incontinence, sleep duration or number of awakenings, activity level or exercise habits, hobbies or social participation, daytime location or movement history, bathing frequency, outing frequency, household chores performed, independence in shopping or cooking, communication frequency with friends or family, television or radio viewing time, newspaper or book reading time, and participation in religious or community activities. Specifically, in this step, the system initiates a data pipeline for extracting, transforming, and loading data from heterogeneous hybrid data sources. The Patient Information Acquisition Unit issues an SQL query to the database of the electronic medical record system. The Patient Information Acquisition Unit extracts structured medical record data. The Daily Living Motion Data Acquisition Unit receives a time-series tensor of three-dimensional coordinates output from a motion capture device. The Daily Living Motion Data Acquisition Unit inputs the three-dimensional coordinate data into a skeletal estimation algorithm. The skeletal estimation algorithm outputs joint angles and movement speeds as feature vectors. The Vital Signs Data Acquisition Unit receives waveform data of biological signals from a wearable device via Bluetooth communication. The vital signs data acquisition unit applies Fast Fourier Transform to waveform data. This unit extracts a feature array in the frequency domain. The cognitive function evaluation data acquisition unit inputs acoustic and linguistic features extracted from speech data into a Support Vector Machine (SVM). The SVM outputs a continuous score representing the degree of cognitive decline. The environmental factors acquisition unit acquires spatial data from an external geographic information system (GIS). This unit converts the spatial data into Geohash code. This system maps these diverse data formats to a common data pattern. The system performs a preprocessing module to impute missing values ​​and remove outliers. Through this structure, the system automatically constructs a high-quality dataset for machine learning from noisy real-world environmental data. This system guarantees the accuracy of subsequent AI analysis. This technology can be applied to health management applications that utilize multimodal data, or to subject data collection systems in clinical trials.

[0107] Step 2: Accumulation of Time-Series Data through Long-Term Memory Management. In this step, the Long-Term Memory Management Department comprehensively accumulates all the aforementioned information in a time-series manner, meticulously recording and managing long-term changes in the health status or living conditions of each caregiver, variations in health indicators accompanying specific events (hospitalization, falls, changes in family composition, home renovations, changes in nursing services, disasters, infectious disease outbreaks, etc.), the effects of nursing interventions, seasonal variations or changes in lifestyle, changes in social isolation or economic status, etc. For example, detailed records are kept of blood pressure fluctuations over the past 5 years, changes in cognitive function scores, changes in walking ability or ADL, functional changes before and after rehabilitation interventions, the correlation between fall history and living environment, changes in medication adherence rates, seasonal variations in sleep duration or activity levels, changes in health indicators accompanying changes in family composition or nursing services, deterioration of health status during disasters, and changes in lifestyle during infectious disease outbreaks, etc. Specifically, in this step, the Long-Term Memory Management Department performs the process of writing the collected normalized data into a distributed time-series database. The Long-Term Memory Management Department assigns high-precision timestamps to each data point. The Long-Term Memory Management Department sets partition keys corresponding to the types of data. This long-term memory management unit distributes data across multiple storage nodes. It applies data compression algorithms when writing data. It minimizes storage capacity consumption. As a background process, it periodically executes data aggregation tasks. These aggregation tasks downsample high-resolution data (second-level) to hourly or daily statistical values ​​(mean, variance, maximum, minimum). The downsampled data is then moved to a long-term storage layer. It indexes flags representing specific events. This system accelerates event-driven data retrieval. Through this architecture, the system maintains query response times for petabyte-scale time-series data in the millisecond range. It ensures data continuity over long periods. It significantly improves data read / write performance. This technology is applicable to data management systems for large-scale sensor networks or high-frequency trading data accumulation systems in financial markets.

[0108] Step 3: Comprehensive Analysis and Risk Prediction through Generative AI. In this step, the generative AI processing unit comprehensively analyzes all the data mentioned above. Natural Language Processing AI extracts risk factors or events from medical and nursing records; Time Series Analysis AI detects changes or abnormal patterns in vital signs, ADL (Activities of Daily Living), cognitive function, and daily life information; Image Recognition AI extracts risk factors from images of food intake or behavioral patterns and the living environment; Causal Inference AI analyzes the correlation between living environment or nursing history, medication status, and health status; and Generative Model AI predicts future risks such as fall risk, dementia progression risk, reduced daily living function risk, dehydration or infection risk, social isolation risk, and economic hardship risk. For example, AI automatically extracts the tendency for walking ability or cognitive function to decline, changes in medication adherence, seasonal changes in physical condition, the correlation between fall history and living environment, the association between sleep deprivation and cognitive decline, the association between social isolation and depressive tendencies, and the association between deteriorating economic conditions and reduced nutritional status, learning individual-specific patterns. As a specific example of a prompt text for the generative AI to read, one could be: "Based on the following caregiver data, predict the risk of falls and the risk of dementia progression, and generate a life support and preventative care plan to reduce these risks. Caregiver data: Age 83, female; Past medical history: osteoporosis, hypertension, history of falls; ADL: Walking ability 2 (using a cane); Toileting independence: partially assisted; Dementia assessment: MMSE 18; Vital signs: Blood pressure 155 / 95, heart rate 102, Sleep duration 4 hours, missed medication twice a week; Living environment: many steps, living alone; Care history: home care three times a week; Life information: reduced food intake, decreased activity level; all data from the past year are appended in time series." Specifically, in this step, the generative AI processing unit reads accumulated time-series data and static configuration file data in batch processing. The generative AI processing unit then inputs the read data into a multimodal neural network. This multimodal neural network has independent encoder layers corresponding to each modality of text, image, and time-series waveforms. This natural language processing AI converts text data into word embedding vectors. This time-series analysis AI normalizes vital sign data, generating tensors sliced ​​out using a sliding window approach. This image recognition AI converts image data into a multidimensional array of pixel values. This generative AI processing unit inputs the feature vectors output from each encoder layer into a cross-attention mechanism. This cross-attention mechanism calculates the strength of the correlations between different modalities as weight coefficients. This cross-attention mechanism outputs a unified context vector. This causal inference AI uses a structural causal model to eliminate the influence of confounding factors from the context vector. This generative model AI uses a decoder layer to generate a future risk probability distribution from the context vector. This generative model AI outputs a prediction score and its confidence interval for each risk item.This system sets an alarm flag in memory when the predicted score exceeds a predetermined safety threshold. Through this structure, the system mathematically detects complex risk factors that might be missed in a single data source. This system significantly reduces the false positive rate of predictions. This technology can be applied to fault prediction systems for complex mechanical systems or to integrated risk management systems in financial institutions.

[0109] Step 4: Automatic generation of individually optimized preventive care and life support plans. In this step, the preventive care plan proposal department automatically generates plans based on the analysis results of the generative AI processing department, including: eliminating steps or installing handrails in the living environment; introducing walking training or balance exercises; brain training or dialogue programs to inhibit dementia progression; introducing medication support devices; strengthening care through nursing staff; adjusting the rhythm of life to improve sleep; reassessing food intake or promoting hydration; recommending hobbies or social participation; strengthening collaboration with family or community; proposing economic support policies; promoting community participation to prevent social isolation; and strengthening the safety confirmation system during disasters. The plan also considers the individualized optimization of the caregiver's living environment or hobbies, family composition, economic status, religious constraints, and social background. As a specific example of a prompt text that the AI ​​can read, one could say, "Based on the following health, living, and environmental data of the caregiver, generate a specific preventive care and life support plan in Japanese to reduce the risk of falls and the risk of dementia progression. Data: Age 80, male; Past medical history: diabetes, hypertension, history of falls; ADL: walking ability 3 (using a walker); Occupational independence: independent; Dementia assessment: MMSE 20; Vital signs: blood pressure 150 / 90, heart rate 95, sleep duration 5 hours, missed medication once a week; Living environment: few steps, cohabitation; Care history: day care twice a week; Life information: slightly reduced activity level, interested in activities; all data from the past 2 years are attached in time series." Specifically, in this step, the preventive care plan proposal department receives the risk prediction score and context vector output in step 3 as input data. The preventive care plan proposal department then starts a constrained optimization solver. This constrained optimization solver formulates the caregiver's economic situation, family support capabilities, and local service resources as hard constraints. It sets maximizing risk reduction as the objective function. The solver uses mixed-integer linear programming to compute the optimal combination vector of support strategies. The preventive care plan proposal department inputs the computed combination vector into the prompt generation module. This module embeds the combination vector into a natural language template. It generates structured prompt text. The preventive care plan proposal department sends the generated prompt text to the inference endpoint of a large-scale language model. The large-scale language model performs an autoregressive token generation process. It outputs a natural language plan text containing specific action plans, schedules, and precautions. The system converts the output plan text into PDF or HTML report data. Through this structure, the system automatically generates plans that balance logical consistency and human readability. This system dramatically improves the productivity of plan creation.This technology can be applied to personalized course generation systems in the education field or optimal delivery route proposal systems in the logistics field.

[0110] Step 5: Real-time Remote Monitoring and Automatic Notification and Emergency Response in Case of Anomalies. In this step, the remote monitoring collaboration unit receives biometric data or life information in real time from the patient's home medical devices or wearable devices, monitoring cameras, fall sensors, door opening / closing sensors, sleep sensors, wandering detection sensors, emergency notification buttons, GPS devices, etc., and sends it to the generative AI processing unit sequentially. The generative AI analyzes the received data in real time. For example, in cases of sudden increases in blood pressure or heart rate, fall sensor activation, respiratory arrest detected during sleep, repeated forgetting of medication, detection of wandering or abnormal behavior, door opening / closing sensor activation at night, or emergency notification button presses, it automatically sends alarm notifications to caregivers or family members and automatically dispatches remote examinations or emergency responses when necessary. As a specific example of a prompt message for the AI ​​to read, one could be: "Analyze the following real-time biological data and life information. If an anomaly or increased risk is detected, generate a message in Japanese that should be sent to caregivers or family members, along with recommended coping strategies. Data: Blood pressure 160 / 100, heart rate 110, fall sensor activation, sleep duration 3 hours, two consecutive missed medication doses, wandering sensor activation, and all data from the past week appended in time series." Specifically, in this step, the remote care collaboration unit establishes a two-way communication session between the edge device and the cloud server. The remote care collaboration unit continuously receives a stream of biological data using the WebSocket protocol. The remote care collaboration unit stores the received stream data in a circular buffer in memory. The remote care collaboration unit applies a moving average filter and a Kalman filter to the data in the circular buffer. The remote care collaboration unit generates smoothed data to remove sensor noise. The remote care collaboration unit inputs the smoothed data into an anomaly detection algorithm. This anomaly detection algorithm is constructed using either an isolated forest or a single-class support vector machine. This anomaly detection algorithm outputs an anomaly score as the degree to which the input data deviates from the clustering of normal data. The remote care collaboration unit initiates an interruption process when the anomaly score exceeds a dynamic threshold. The unit generates a JSON-formatted payload containing a code representing the type of anomaly, the time of occurrence, and a snapshot of the corresponding sensor data. This payload is then sent to a push notification server. The push notification server immediately distributes alerts to the portable terminals of registered caregivers. Upon receiving a pressed emergency notification button, the system activates an automated voice response system to dial emergency services. Through this architecture, the system constructs a fully automated anomaly detection and response loop that does not rely on human monitoring. This system reduces the response delay to fatal events to physical limits. This technology can be applied to remote monitoring systems for autonomous vehicles or security monitoring systems for critical infrastructure.

[0111] (Specific examples) For example, when applying the system of the present invention to elderly women living alone as caregivers, the caregiver information acquisition unit automatically obtains information such as hypertension, osteoporosis, past fall history, medication content, allergy information, vaccination history, rehabilitation records, swallowing function evaluation, and history of pressure ulcers from electronic medical records or nursing records. The daily living activities data acquisition unit obtains information such as walking ability (using a cane), toileting independence (partial care), eating status (independent), bathing or dressing independence (caregiver), and getting out of bed or standing independence (independent) from personnel records or sensors. The data acquisition department collects data on the following: independence in moving to the toilet (independence), ability to climb stairs (inability), outdoor mobility (once a week), independence in shopping or cooking (caregiver), and independence in money management or medication management (family management). The vital signs data acquisition department collects real-time data on blood pressure (155 / 95), heart rate (102), body temperature, blood oxygen saturation, weight, and body fat percentage from a home blood pressure monitor or wearable device; real-time data on sleep duration (4 hours) and number of awakenings (2 times) from a sleep sensor; and real-time data on steps (less than 1000 steps per day) and calories burned from an activity meter. The cognitive function assessment data acquisition department collects MMSE score 18, HDS-R score 20, response delay in daily conversation, hesitancy, and observation records from family members or other personnel. The living environment factors acquisition department collects data on factors such as the number of steps in the residence, the presence of handrails, living alone, no pets, temperature or humidity of the residence, noise, insulation performance of the residence, difficulty accessing nearby medical institutions, high disaster risk, and weak local caregiving system. The Nursing History Acquisition Department records home care visits (3 times per week), no use of day care services, history of rehabilitation visits, past fall accidents or near-miss events, history of changes to the nursing plan, staff turnover history, service satisfaction evaluations, changes in nursing assessment categories, nursing payment performance, requests or complaints from family members, etc. The Medication Status Acquisition Department obtains information from medication support equipment, including missed medication visits (twice per week), medication adherence rate (80%), medication alarm activation status, no medication side effects, medication change history, no medication allergies, medication costs, medication storage conditions, medication remaining quantity management, medication instruction history, etc. The Life Information Acquisition Department obtains information from... Data collected from sensors or applications and personnel records includes: food intake (low intake), frequency of bowel movements (twice a day), presence or absence of incontinence (once a week), sleep duration (4 hours), activity level (decreasing tendency), hobbies (reading only), social participation (almost none), bathing frequency (twice a week), frequency of going out (once a week), household chores (not possible), independence in shopping or cooking (in care), frequency of communication with friends or family (once a week), television viewing time (2 hours a day), newspaper reading time (30 minutes a day), and participation in religious or community activities (none). Specifically, this system executes a customized data collection profile for each specific caregiver profile.The Patient Information Acquisition Unit sends HTTP requests to the API endpoint of the electronic medical record system. This unit parses the medical record data returned in JSON format. It extracts the ICD-10 codes for hypertension or osteoporosis. The Daily Living Motion Data Acquisition Unit acquires point cloud data from a depth camera installed indoors. This unit inputs the point cloud data into a 3D convolutional neural network. The 3D convolutional neural network outputs asymmetry in walking posture or stride deviation as a numerical tensor. The Vital Signs Data Acquisition Unit acquires heart rate RR interval data in milliseconds from a smartwatch. It calculates the ratio of high-frequency to low-frequency components as indicators of autonomic nervous activity from the RR interval data. The Cognitive Function Evaluation Data Acquisition Unit acquires speech waveforms of daily conversations via a smart speaker. It extracts feature vectors representing pause lengths or vocabulary diversity from the speech waveforms. The Environmental Factors Acquisition Unit acquires environmental data from indoor temperature and humidity sensors. It also acquires external meteorological data from the Japan Meteorological Agency's open data API. This system associates this multidimensional data with the unique ID of the person being cared for and stores it in a relational database and a time-series database. Through this structure, the system reconstructs the state of the person being cared for as a digital twin in virtual space using multi-angle sensor fusion technology. This system improves the comprehensiveness and accuracy of the data. This technology can be applied to personalized healthcare monitoring systems or resident status estimation systems in smart homes.

[0112] The Long-Term Memory Management Department comprehensively accumulates all this information over a time series, meticulously recording the past three years' blood pressure fluctuations, the tendency for cognitive function scores to decline, changes in walking ability or ADL (Activities of Daily Living), functional changes before and after rehabilitation interventions, the correlation between fall history and living environment, changes in medication adherence, seasonal variations in sleep duration or activity levels, changes in health indicators accompanying changes in family composition or care services, deterioration of health status during disasters, and changes in lifestyle during infectious disease outbreaks. Specifically, the Long-Term Memory Management Department encodes the collected multidimensional data into node features of a graph neural network. It establishes edges between nodes based on temporal adjacency and semantic relevance. The department formats the past three years' blood pressure fluctuation data into a time series tensor. It quantifies the tendency for cognitive function scores to decline as the slope parameter of a linear regression model. The department generates a context vector combining the time of the fall event and its immediate preceding living environment data (illuminance, temperature, humidity) and vital sign data (blood pressure fluctuations, sleep deprivation). This context vector is then input into a clustering algorithm. This clustering algorithm extracts composite patterns that are the root causes of falls. The long-term memory management system calculates the cross-correlation function between time-series changes in medication adherence rates and changes in cognitive function scores. This long-term memory management system determines the causal relationship between the two with accompanying time delays. This long-term memory management system assigns these analysis results as metadata to the original time-series data. This long-term memory management system persists the data to a distributed file system. Through this structure, the system constructs not merely a list of data, but a semantically structured knowledge base. This system significantly improves the learning efficiency and inference accuracy of subsequent predictive models. This technology can be applied to advanced data mining systems for electronic medical records or long-term analysis systems for customer behavior history.

[0113] The generative AI processing unit comprehensively analyzes all this data. Natural language processing AI extracts risk factors or processes from medical and nursing records; time series analysis AI detects changes or abnormal patterns in vital signs, ADL (Activities of Daily Living), cognitive function, and life information; image recognition AI extracts risk factors from images of eating content or behavioral patterns and the living environment; causal inference AI analyzes the correlation between living environment or care history, medication status, and health status; and generative model AI predicts future risks such as fall risk, dementia progression risk, reduced functioning risk, dehydration or infection risk, social isolation risk, and economic hardship risk. For example, the AI ​​automatically extracts the tendency for declining walking ability or cognitive function, changes in medication adherence, seasonal changes in physical condition, correlation between fall history and living environment, correlation between sleep deprivation and cognitive decline, correlation between social isolation and depressive tendencies, and correlation between deteriorating economic conditions and reduced nutritional status, learning individual-specific patterns. Specifically, this generative AI processing unit reads the feature tensors of the caregivers from the constructed knowledge base in batches of size 1. This generative AI processing unit feeds the feature tensors into the input layer of the deep learning model. This natural language processing AI employs the BERT architecture. It extracts keywords such as "shaking" and "decreased willingness" from the text data of nursing records and converts them into multi-dimensional embedding vectors. The time series analysis AI uses an LSTM network. It receives a time series array of vital sign data and ADL data from the past three years as input. It calculates the probability of future walking ability decline as the output value of a sigmoid function. The image recognition AI uses a ResNet architecture. It scores the degree of nutrient deficiency from the image tensor of eating. The causal inference AI uses a Bayesian network to model the probabilistic dependencies between sleep deprivation, missed medication, and fall risk as a graph structure. The generative model AI receives a ensemble vector combining feature vectors from the outputs of these AIs as input. It uses a variational autoencoder to map the ensemble vector to a latent space. It generates multiple scenarios of future health states from the latent space. It uses a Softmax function to calculate the probability of each scenario. The system outputs the scenario with the highest probability of occurrence. This system uses this architecture to mathematically identify hidden risk factors from a vast array of variables that are beyond the reach of human experts. It optimizes the timing of preventative interventions. This technology can be applied to prognostic prediction systems for complex diseases or predictive maintenance systems for machinery.

[0114] Based on the analysis results of the generative AI processing department, the Preventive Care Plan Proposal Department automatically generates plans for the living environment, such as eliminating steps or installing handrails, introducing walking training or balance exercises, brain training or dialogue programs to inhibit the progression of dementia, introducing medication support devices, strengthening care through nursing staff, adjusting lifestyle rhythms to improve sleep, reassessing dietary content or promoting hydration, recommending hobbies or social participation, strengthening collaboration with family or community, proposing economic support policies, promoting community participation to prevent social isolation, and strengthening safety confirmation systems during disasters. It also considers individual optimization plans based on the caregiver's living environment or hobbies, family composition, economic status, religious constraints, social background, etc. For example, for elderly people living alone, the plan proposes various preventative care and life support measures, including strengthening remote care and emergency reporting systems; introducing dementia prevention programs when there is a tendency for cognitive decline; enhancing home renovations or walking training when there is a high risk of falls; introducing medication support devices when there is frequent medication forgetfulness; proposing sleep improvement programs when there are sleep disorders; utilizing nutritional supplements or eating support services when food intake is low; recommending community activities or volunteer participation when there is a tendency for social isolation; proposing the utilization of public support systems when there is economic hardship; and strengthening safety confirmation systems or evacuation support measures during disasters. Specifically, the preventative care plan proposal department receives vector data of predicted scenarios and risk scores output from the generative AI processing department. The department then activates a hybrid system combining a rule-based inference engine and a deep reinforcement learning agent. For items with risk scores exceeding a predetermined threshold, the inference engine extracts a predefined candidate list of coping strategies from the database. The deep reinforcement learning agent reads the extracted candidate list as the action space. This deep reinforcement learning agent reads the caregiver's profile vector (living alone, economic status, medical history, etc.) as its state space. Using the Q-learning algorithm, it calculates the action value function for combinations of coping strategies. The agent selects the set of coping strategies that maximize action value as the optimal strategy. The preventative care plan proposal department inputs the parameter set of the selected optimal strategy into a large-scale language model. This model converts the parameter set into a natural language article. The model outputs structured plan data containing specific action items such as "setting up remote monitoring sensors," "adding twice-weekly home rehabilitation visits," and "importing medication calendars." The system sends the output plan data to the care manager's terminal and initiates the approval workflow. Through this structure, the system computationally solves multi-objective optimization problems, automatically generating plans that maximize effectiveness under resource constraints. This system significantly reduces the time required for plan creation. This technology can be applied to automated systems for generating personalized financial investment portfolios or to optimization systems for cultivation plans in smart agriculture.

[0115] The remote care collaboration unit receives real-time biometric data or life information from home medical devices or wearable devices, care cameras, fall sensors, door opening / closing sensors, sleep sensors, wandering detection sensors, emergency notification buttons, GPS devices, etc., and sends it sequentially to the generative AI processing unit. The generative AI analyzes the received data in real time, automatically sending alarm notifications to caregivers or family members when anomalies occur, such as a sudden increase in blood pressure or heart rate, fall sensor activation, respiratory arrest detected during sleep, repeated forgetting of medication, detection of wandering or abnormal behavior, door opening / closing sensor activation at night, or emergency notification button activation. It also automatically dispatches remote medical examinations or emergency responses when necessary. Specifically, this remote care collaboration unit maintains a constant TCP / IP session with the edge gateway device located in the caregiver's home. This remote care collaboration unit uses the MQTT protocol to receive sensor data streams from the edge gateway device. This remote care collaboration unit writes the received stream data into an in-memory time-series database. This remote care collaboration unit then activates its stream processing engine. This stream processing engine applies a statistical anomaly detection algorithm to data windows updated in milliseconds. It calculates the first derivative of blood pressure values. It detects a sharp rise when the first derivative exceeds a set safety threshold. It calculates the norm of the acceleration tensor output from the fall sensor. It detects a fall when the norm exceeds a specified multiple of gravitational acceleration. Upon detecting an anomaly, the remote care collaboration unit generates an alarm payload containing a snapshot of the corresponding sensor data and basic information about the person being cared for. This alarm payload is then input into a generative AI processing unit. The generative AI processing unit analyzes the alarm payload and determines the urgency level. When the urgency level is at its highest, the system uses an automatic speech synthesis module to generate a notification voice to the emergency command center and executes an automatic phone call using the SIP protocol. Through this structure, the system completes the entire process from anomaly detection to initial response without human intervention. This system plays a crucial role in improving life-saving rates. This technology can be applied to automated monitoring systems for hazardous operations in factories or real-time accident detection systems for transportation infrastructure.

[0116] As a specific example of a prompt text that generative AI can read, it could be: "Based on the following caregiver data, predict the risk of falls and the risk of dementia progression, and generate a life support and preventative care plan to reduce these risks. Caregiver data: Age 83, female; Past medical history: osteoporosis, hypertension, history of falls; ADL: Walking ability 2 (using a cane); Toilet management independence: partially assisted; Dementia assessment: MMSE 18; Vital signs: Blood pressure 155 / 95, heart rate 102, Sleep duration 4 hours, missed medication twice a week; Living environment: many steps, living alone; Care history: home care three times a week; Life information: reduced food intake, decreased activity level; all data from the past year appended in time series." Specifically, this system uses a prompt generation module to convert the structured data extracted from the database into natural language text format. This prompt generation module maps multidimensional variables such as the caregiver's age, gender, medical history, ADL score, mean and variance of vital signs, and environmental characteristics according to a predefined JSON pattern. The module then inputs the mapped JSON data into a template engine. This template engine expands the variables into text strings, dynamically constructing context-aware prompts that are easily understood by a large-scale language model. The system then inputs the constructed prompts into a tokenizer. The tokenizer converts the text strings into a tag ID tensor, which is an array of integers. The system feeds this tag ID tensor into the input layer of a large-scale language model with a Transformer architecture. This large-scale language model uses a self-attention mechanism with billions of parameters to compute dependencies between input tags. The large-scale language model generates an output tag column containing probability values ​​for fall risk, predicted scores for dementia progression, and a list of specific interventions to mitigate them. The system inputs the output tag column into a decoder, which inversely converts it into natural language text data. The system uses regular expressions to extract numerical risk scores or action items from the text data and saves them as structured plan data in a database. This architecture enables seamless conversion between unstructured and structured data. The system optimizes a general language model for specific domain tasks and leverages it. This technology can be applied to automated response generation systems in customer support, or to automated summarization and risk extraction systems for legal documents.

[0117] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0118] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0119] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0120] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0121] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0122] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0123] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0124] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0125] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0126] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0127] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0128] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0129] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0130] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0131] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0132] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0133] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0134] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0135] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0136] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0137] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0138] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0139] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0140] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0141] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0142] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0143] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0144] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0145] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0146] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0147] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0148] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0149] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the special processing unit 290.

[0150] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0151] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0152] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0153] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0154] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0155] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0156] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0157] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0158] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0159] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0160] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0161] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0162] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0163] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0164] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0165] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0166] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0167] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0168] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the special processing unit 290.

[0169] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0170] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0171] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0172] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0173] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0174] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0175] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0176] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The robot's emotions can be determined by the emotion-specific model 59. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0177] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0178] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0179] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0180] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0181] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0182] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, nearby sentiment values ​​are similar to each other. Figure 10 Examples show how emotions such as "peace of mind," "stability," and "reassurance" can result in similar emotional values.

[0183] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0184] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0185] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0186] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0187] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0188] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0189] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0190] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0191] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0192] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0193] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0194] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0195] In addition, the following notes further explain the above explanation.

[0196] (Note 1) A system comprising a patient information acquisition unit, a daily living activities data acquisition unit, a vital signs data acquisition unit, a cognitive function evaluation data acquisition unit, a living environment factors acquisition unit, a nursing history acquisition unit, a medication status acquisition unit, a daily living information acquisition unit, a long-term memory management unit, a generative AI processing unit, a preventive care plan proposal unit, and a remote care collaboration unit, characterized in that... The patient information acquisition unit automatically acquires each patient's health record, chronic disease history, fall history, fracture history, hospitalization history, medication content, allergy information, etc. from electronic medical records or nursing records; The daily living activities data acquisition unit acquires information such as walking ability, toileting independence, eating status, and independence in bathing or changing clothes from the records of caregivers or sensors. The vital signs data acquisition unit acquires blood pressure, heart rate, body temperature, blood oxygen saturation, respiratory rate, weight, etc. in real time from home medical machines or wearable devices; The cognitive function evaluation data acquisition unit acquires regular dementia evaluation scores or AI-automated evaluations of daily conversations and behavioral patterns, as well as observation records from family members or personnel. The living environment factors acquisition department acquires information such as the condition of steps or handrails in the residence, whether the person lives alone or with a partner, the temperature or humidity of the residence, noise levels, and visits to medical institutions. The nursing history acquisition department records the utilization status of home care or day care, rehabilitation and medical history, past nursing accidents or near-miss events, etc. The medication status acquisition unit obtains information from the medication support device or medication record application regarding whether medication was forgotten or taken incorrectly, and the medication adherence rate. The life information acquisition unit obtains information such as food intake or calories consumed, number of times of excretion or incontinence, sleep time or number of times of waking up during the night, activity level or exercise habits from sensors or applications and personnel records; The long-term memory management department comprehensively accumulates all this information in a time-series manner; The generative AI processing unit comprehensively analyzes this data and quantifies factors such as fall risk, dementia progression risk, risk of reduced daily living function, and risk of dehydration or infection. Based on the analysis results of the generative AI processing unit, the preventive care plan proposal department automatically generates individualized and optimized preventive care and life support plans, including improvements to the living environment, walking training, dementia progression inhibition programs, introduction of medication support equipment, enhanced care by nursing staff, sleep improvement, reassessment of food intake, and promotion of water intake. The remote care collaboration unit receives real-time biometric data from the patient's home medical devices or wearable devices, automatically sends alerts to caregivers or family members in case of abnormalities, and automatically schedules remote examinations or emergency responses when necessary; and, All of these processes are run by generative AI.

[0197] (Note 2) As described in Appendix 1, the system is characterized in that the generative AI processing unit analyzes the time-series changes of past walking ability or cognitive function scores, vital sign data, medication adherence rate, living environment factors, nursing history, life information, etc., accumulated in the long-term memory management unit, extracts individual-specific health and lifestyle patterns such as seasonal changes or life events, the effects of nursing interventions, changes in medication adherence rate, and shifts in sleep or activity levels, detects early signs of future fall risk or dementia progression risk, and risk of reduced functioning, and dynamically optimizes the timing or content of interventions and makes proposals in the preventive care plan proposal unit.

[0198] (Note 3) As described in Appendix 1, the system is characterized in that the remote care collaboration unit sequentially sends real-time biological data or life information obtained from the patient's home medical device or wearable device to the generative AI processing unit. When the generative AI processing unit detects abnormal values ​​or an increase in risk, it automatically sends an alarm notification to the caregiver or family member. Furthermore, it determines whether remote diagnosis or emergency response is feasible based on the patient's place of residence, care system, and mobility difficulties. When necessary, it automatically schedules remote diagnosis or emergency response and notifies the patient, family member, and caregiver.

Claims

1. A system comprising a patient medical record acquisition unit, a biological data acquisition unit, a genetic information acquisition unit, an environmental factor acquisition unit, a long-term memory management unit, a generative AI processing unit, a preventive medical plan proposal unit, and a remote monitoring and collaboration unit, characterized in that, The medical record acquisition unit retrieves each patient's diagnostic history, treatment history, and past medical history from the database; the bio-data acquisition unit retrieves real-time bio-information such as heart rate, blood pressure, blood sugar, body temperature, and respiratory rate from wearable devices or medical machines; the genetic information acquisition unit retrieves the patient's genetic markers or disease risk gene information; the environmental factors acquisition unit retrieves environmental factor data such as the patient's lifestyle habits, smoking history, drinking history, exercise habits, living environment, and occupational environment; the long-term memory management unit accumulates and manages all of each patient's medical records, bio-data, genetic information, and environmental factor information from the past to the present in a time-series manner; the generative AI processing unit comprehensively analyzes all the data obtained from the acquisition and management units, and predicts each patient's health risk through multivariate analysis and pattern recognition, detecting an increase in risk before symptoms appear. The preventive medical plan proposal department, based on the analysis results of the generative AI processing department, generates optimal preventive medical plans for nutrition, exercise, sleep, lifestyle improvement, and medical intervention according to the patient's genetic predisposition, lifestyle, environmental exposure, and past health patterns, and proposes these plans to medical practitioners and patients. The remote monitoring and collaboration department acquires biological data or diagnostic data from the patient remotely in real time, automatically notifies medical practitioners when abnormal values ​​or increased risks occur, and can conduct remote diagnosis when necessary. All of these processes are performed by generative AI.

2. The system as described in claim 1, characterized in that, The generative AI processing unit analyzes the time-series changes of past biological data or medical records accumulated in each patient's long-term memory management unit, extracts individual-specific health patterns caused by seasonal changes, life events, treatment interventions, etc., detects early signs of future symptom deterioration or disease onset, and dynamically optimizes individual intervention timing or content and proposes them in the preventive medical plan proposal unit.

3. The system as described in claim 1, characterized in that, The remote monitoring and collaboration unit will sequentially send real-time biometric data obtained from the patient's home medical device or wearable device to the generative AI processing unit. When the generative AI processing unit detects anomalies or an increase in risk, it will automatically send an alert to the medical practitioner. It will also determine whether remote diagnosis is feasible based on the patient's place of residence or difficulty in moving, and automatically schedule remote diagnosis when necessary, and notify both the patient and the medical practitioner.

Images