Accompanying diagnosis management system based on artificial intelligence technology

Abstract

The invention discloses an accompanying diagnosis management system based on an artificial intelligence technology, and belongs to the technical field of medical accompanying diagnosis. The system comprises a user management and health archive module, an intelligent reservation module, a health suggestion and diagnosis and treatment assistance module, an accompanying diagnosis and emotion support module, a data analysis and feedback module and a safety and privacy protection module. The user management and health archive module realizes archive management by using biological recognition, an NLP technology and a block chain; the intelligent appointment module recommends departments and optimizes appointment reminding based on an ICD-11 standard; the health suggestion and diagnosis and treatment auxiliary module is combined with medical knowledge retrieval and interpretability AI to provide diagnosis and treatment suggestions; the accompanying diagnosis and emotion support module evaluates emotion and intervenes through multi-modal data acquisition; the data analysis and feedback module performs deep analysis and provides decision support; the security and privacy protection module uses various encryption and detection technologies to guarantee data security. The system effectively solves the problem of traditional doctor-seeing accompanying diagnosis, and achieves the functions of accurate registration, personalized diagnosis and treatment, emotion care, data safety management and the like.

Description

Technical Field

[0001] This invention relates to the field of medical escort technology, and in particular to an escort management system based on artificial intelligence technology. Background Technology

[0002] With increasing health awareness and more complex medical needs, the importance of medical escort services is becoming increasingly prominent. In the traditional medical process, patients often face numerous difficulties. For example, patients are unfamiliar with the layout of hospital departments and the consultation process, making it difficult to choose the appropriate department when registering, delaying their consultation time and affecting the efficiency of diagnosis and treatment. This is especially true for patients with multiple illnesses or complex conditions, where determining the initial consultation department is even more challenging.

[0003] Meanwhile, patients' emotional state during medical visits is easily overlooked. Physical discomfort from illness and anxiety about the condition can easily lead to negative emotions such as anxiety and fear, which may affect treatment effectiveness and the recovery process. Currently, most escort services only provide basic guidance and lack effective assessment and intervention methods for patients' emotions.

[0004] In terms of health management, existing patient health record management methods have many shortcomings. Health data in hospital systems is mostly unstructured, making integration and processing difficult. It cannot reflect patients' health status in a real-time and comprehensive manner, hindering doctors from making accurate diagnoses and personalized treatment plans. Furthermore, data from different medical institutions is often independent, forming information silos, which is detrimental to the long-term tracking and comprehensive management of patient health information.

[0005] Furthermore, medical data security and privacy protection are also major concerns. Patients' personal information and health records contain sensitive content, and any leakage could have serious consequences for them. However, some medical systems have vulnerabilities in data access control and encryption protection, failing to meet increasingly stringent data protection regulations in the medical industry and thus making it difficult to guarantee patient data security. In conclusion, developing an AI-based escort management system is of significant practical importance and can effectively solve many of the problems in the traditional medical treatment and escort processes. Summary of the Invention

[0006] The main objective of this invention is to provide a medical companion management system based on artificial intelligence technology, which can effectively solve the problems in the background technology.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An artificial intelligence-based escort management system, the system comprising: User Management and Health Records Module: Used to create individual health records for patients and update these records in real time based on treatment results; The intelligent appointment module recommends suitable outpatient departments based on individual health records and current patient symptoms, assisting patients in making appointments and setting reminders. Health advice and treatment assistance module: This module generates personalized health assistance and treatment advice for patients based on their health records and common disease templates. The escort and emotional support module is used to provide guidance services to patients through voice interaction, help patients find their destination, assess patients' emotional state through emotion analysis technology, and comfort patients based on the assessment results. Data analysis and feedback module: used to perform in-depth analysis of patients' health records, identify potential health risks, generate personalized health reports, and recommend appropriate medical or examination suggestions; Security and Privacy Protection Module: This module is used to protect patients' personal information and health records, ensuring that only authorized personnel can access this data. It also sets different access permissions for different users and conducts regular privacy protection and data security compliance checks to ensure compliance with the data protection requirements of the healthcare industry.

[0008] Preferably, the user management and health record module includes: a registration / login subsystem for biometrics and two-factor authentication; an NLP-based data cleaning engine for processing unstructured data from the hospital's HIS / LIS system; and a blockchain network for automatically updating EHR data via smart contracts, wherein data storage complies with HIPAA compliance requirements.

[0009] Preferably, the intelligent appointment module achieves department recommendation through the following steps: constructing an ICD-11 standard symptom-department association matrix, using a multilayer perceptron classification model to process user-input symptom features; dynamically optimizing appointment time window allocation based on the Q-learning algorithm; and adaptively triggering multi-channel reminders via SMS / WeChat / voice through a behavior analysis model.

[0010] Preferably, the health advice and treatment assistance module includes: a medical knowledge retrieval system built on the RAG architecture, which is connected to the UpToDate clinical guidelines in real time; an interpretable AI component that generates treatment advice with SHAP value analysis; and then, based on the prediction results, clinicians review the advice according to their own judgment.

[0011] Preferably, the companionship and emotional support module achieves emotional intervention in the following ways: edge computing devices collect speech spectrum, facial micro-expression and heart rate variability data in real time; an emotion state classification model based on LSTM network outputs anxiety level scores; progressive reassurance scripts are dynamically generated according to the CBT therapy template library; and a crisis intervention protocol is automatically triggered when the suicide risk score is detected to exceed the threshold.

[0012] Preferably, the security protection module implements: a fine-grained data access control strategy based on ABE encryption; federated learning model training using homomorphic encryption; an abnormal behavior detection model constructed according to the MITREATT&CK framework; and a penetration testing and emergency response drill mechanism executed every quarter.

[0013] Preferably, the data analysis module includes: a digital twin modeling unit that synchronizes time-series data of patient physiological parameters; a dynamic threshold early warning mechanism that analyzes the deviation of health indicators through a sliding window; and an automated A / B testing framework that compares the effectiveness indicators of different intervention programs.

[0014] A computer-readable storage medium storing program instructions for executing the aforementioned AI-based escort management system.

[0015] Compared with the prior art, the present invention has the following beneficial effects: Precise Registration and Intelligent Reminders: The intelligent appointment module constructs an ICD-11 standard symptom-department association matrix and uses a multilayer perceptron classification model to recommend suitable outpatient departments with an accuracy rate of ≥98%, assisting patients in quickly and accurately registering. Simultaneously, it optimizes appointment time window allocation based on the Q-learning algorithm and provides reminders through multiple channels, such as SMS, WeChat, and voice, flexibly reminding patients based on their appointment time, preferences, and historical feedback, greatly improving the punctuality and convenience of patient visits.

[0016] Personalized Health Management: The user management and health record module utilizes biometrics, two-factor authentication, and blockchain technology to establish and update health records in real time, complying with HIPAA requirements. The health advice and treatment assistance module leverages the RAG architecture to retrieve medical knowledge and access clinical guidelines. Combined with interpretable AI components, it generates treatment recommendations with SHAP value analysis. After review by doctors, these recommendations provide patients with personalized health management plans, improving the accuracy and scientific rigor of diagnosis and treatment.

[0017] Emotional Support and Crisis Intervention: The companionship and emotional support module collects multimodal data through edge computing devices and accurately assesses patients' anxiety levels using an LSTM network. Personalized reassurance messages are generated based on a CBT therapy template library to alleviate patients' negative emotions. When the suicide risk score exceeds a threshold, a crisis intervention protocol is automatically triggered to ensure the patient's mental health and safety.

[0018] Data security assurance: The security and privacy protection module uses ABE encryption to achieve fine-grained data access control, homomorphic encryption to support federated learning model training, and builds an abnormal behavior detection model based on the MITREATT&CK framework. Penetration tests and emergency response drills are conducted regularly to comprehensively protect patient data security and ensure that the system meets the data protection requirements of the medical industry.

[0019] In-depth data analysis and decision support: The data analysis and feedback module predicts patients' physiological parameters through digital twin modeling, promptly detects abnormal health indicators using a dynamic threshold early warning mechanism, and compares the effectiveness of different intervention programs using an automated A / B testing framework. This provides strong data support for doctors to adjust treatment plans, optimize escort services, and make hospital management decisions, thereby promoting the continuous improvement of medical service quality. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This diagram illustrates the portion of the smart contract code that updates the blockchain EHR in this invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0022] like Figure 1 As shown, a medical companion management system based on artificial intelligence technology specifically includes: User Management and Health Records Module: Used to create individual health records for patients and update these records in real time based on treatment results; The intelligent appointment module recommends suitable outpatient departments based on individual health records and current patient symptoms, assisting patients in making appointments and setting reminders. Health advice and treatment assistance module: This module generates personalized health assistance and treatment advice for patients based on their health records and common disease templates. The escort and emotional support module is used to provide guidance services to patients through voice interaction, help patients find their destination, assess patients' emotional state through emotion analysis technology, and comfort patients based on the assessment results. Data analysis and feedback module: used to perform in-depth analysis of patients' health records, identify potential health risks, generate personalized health reports, and recommend appropriate medical or examination suggestions; Security and Privacy Protection Module: This module is used to protect patients' personal information and health records, ensuring that only authorized personnel can access this data. It also sets different access permissions for different users and conducts regular privacy protection and data security compliance checks to ensure compliance with the data protection requirements of the healthcare industry.

[0023] Furthermore, the user management and health record module includes: A) Biometric and Two-Factor Authentication Registration / Login Subsystem: This subsystem employs a multimodal fusion algorithm, comprehensively utilizing facial recognition and speech recognition technologies to improve the accuracy and security of user authentication. The formula for the multimodal fusion algorithm is:

[0024] in, , These two parameters were obtained through extensive experimental optimization, indicating that in the comprehensive score, the weight of the face recognition score is 0.6 and the weight of the speech recognition score is 0.4. This represents the face matching score, which is the score of the input face image. The degree of matching with the pre-stored facial template; The speech recognition score represents the input speech signal. The degree of matching with pre-stored voice templates. When At that time, the user is considered to have passed verification.

[0025] B) An NLP-based data cleaning engine for processing unstructured data from hospital HIS / LIS systems: Specifically, the clinical text processing workflow is as follows: 1) Extract entities using the BioBERT model: The formula for entity extraction is:

[0026] in, This represents the probability of entity e appearing given the input sentence s; It is a weight matrix. These are the hidden states labeled [CLS] in the output of the BioBERT model. This is the bias term. The output is converted into a probability distribution using the softmax function.

[0027] BioBERT is a pre-trained language model specifically designed for the biomedical field. Pre-trained on large-scale biomedical text data, it is better able to understand biomedical terminology and semantics. Using the BioBERT model, various entities can be extracted from clinical texts, such as disease names, drug names, and symptoms.

[0028] 2) Rule-based outlier correction: Outliers frequently appear in clinical data, such as inconsistent blood pressure units. A rule engine is a rule-based reasoning system that processes data according to predefined rules. For example, regarding blood pressure unit conversion, if 1 mmHg = 0.133 kPa, when blood pressure data is detected in mmHg units, it is converted to kPa units.

[0029] C) A blockchain network that enables automatic updates of EHR data through smart contracts, wherein data storage complies with HIPAA compliance requirements; Adopting such Figure 2 The smart contract code shown updates the electronic health record (EHR), explained in detail below: The function "function updateEHR(bytes32patientID,stringmemorynewRecord)public" defines a public function "updateEHR" that accepts two parameters: "patientID" represents the patient's unique identifier, and "newRecord" represents the new health record to be added.

[0030] “require(ABE.checkAccess(msg.sender,patientID));”: Before updating a record, the caller needs to be checked for permission to access the patient record using the Attribute-Based Encryption (ABE) scheme.

[0031] "ABE.checkAccess" is a function that checks access permissions, and "msg.sender" represents the address where the function is called.

[0032] “EHR[patientID].push(newRecord)”: If the caller has access permissions, the new health record is added to the patient’s EHR list.

[0033] “emitNewEntry(block.timestamp)”: Triggers an event “NewEntry” and records the current blockchain timestamp for subsequent auditing and tracking.

[0034] Furthermore, the intelligent appointment module recommends departments through the following steps: First, we construct the ICD-11 standard symptom-department association matrix, and then use a multilayer perceptron classification model to process the user-input symptom features. Specifically, feature vector construction: feature vector The definition is as follows:

[0035] in, It is a numerical value obtained by encoding the patient's symptoms according to the International Classification of Diseases, 11th Revision (ICD-11), which can accurately represent the patient's symptom type; The degree of pain a patient experiences is typically quantified using numbers from 1 to 10. This indicates the duration of the patient's symptoms, such as the number of days or hours.

[0036] Multilayer Perceptron Output Layer Calculation: The Multilayer Perceptron (MLP) is a commonly used neural network model for classifying and predicting feature vectors. The output layer calculation of an MLP is as follows:

[0037] in, d represents the prediction score of the j-th department; d is the number of neurons in the hidden layer. These are the weights from the hidden layer to the output layer; It is an activation function (such as ReLU, Sigmoid, etc.); It is the weight vector from the input layer to the hidden layer; This is the bias term of the hidden layer. By comparing the predicted scores of all departments, the department with the highest score is selected as the recommended department.

[0038] Dynamically optimize appointment time window allocation based on Q-learning algorithm; Q-learning is a model-free reinforcement learning algorithm used to solve optimal decision problems. In the intelligent appointment module, Q-learning is used to optimize the allocation of appointment numbers. The Q-learning update rules are as follows:

[0039] in, α represents the Q-value of taking action a in state s, where Q represents the expected cumulative reward of taking action a in that state; α is the learning rate, which controls the step size of each update; r is the immediate reward, which is the reward obtained immediately after taking action a; γ is the discount factor, used to balance the importance of immediate reward and future reward. It is the next state that is transitioned to after taking action 'a'; Indicates the next state The optimal Q-value for taking the action is determined. State s is defined as (department, time period), representing the current appointment status; action a is assigning an appointment number. By continuously interacting with the environment and updating the Q-value, the optimal appointment number allocation strategy is ultimately found.

[0040] Based on the specific allocation strategy, multiple channels such as SMS / WeChat / voice notifications are triggered to inform patients of appointment information; Allocation strategy formulation: The process is divided according to appointment time: For patients whose appointment time is approaching (e.g., 1 day in advance, 2 hours in advance, etc.), multiple reminder methods are used to ensure that patients are aware of the appointment. For example, a text message reminder is sent 1 day in advance, and a voice reminder is sent 2 hours in advance.

[0041] Based on patient preferences: The system allows patients to choose their preferred notification method during registration, such as whether they prefer WeChat or SMS notifications. If the patient does not set a preference, the default multi-channel notification strategy will be used.

[0042] Based on patient feedback history: If a patient has previously given negative feedback on a certain reminder method (such as a text message being mistakenly identified as spam), then the subsequent reminder method will be adjusted.

[0043] Multiple channels triggering alerts: SMS reminders: The system integrates the API of SMS service providers. When the reminder conditions are met, the system generates an SMS message with the appointment information (such as appointment time, department, doctor, etc.) according to the preset SMS template and calls the API to send it to the patient.

[0044] WeChat notification: If the patient has linked their WeChat Official Account, the system will push the appointment information to the patient through the WeChat Official Account's message template interface. The message content can include appointment details, navigation links, etc., for the patient's convenience in viewing and operating the app.

[0045] Voice reminder: Using speech synthesis technology, appointment information is converted into a voice file, which is then used to call the patient via a voice call platform to play the voice reminder content.

[0046] Furthermore, the health advice and treatment support modules include: A medical knowledge retrieval system built on the RAG architecture is connected to UpToDate clinical guidelines in real time. The RAG (Retrieval-Augmented Generation) retrieval system is used to retrieve information related to patient problems from a document database, providing doctors with auxiliary diagnostic information. The document relevance scoring formula is as follows:

[0047] in, This represents the relevance score between query q and document d; It is a classic retrieval algorithm based on term frequency and inverse document frequency, used to calculate the relevance of a query to a document; It is a pre-trained language model based on the clinical field, used to capture the semantic relevance between queries and documents; It is a weight parameter, obtained through tuning on a medical question-answering dataset, representing... The weight of the score in the overall score. The document with the highest relevance score is selected as the search result by ranking all documents according to their relevance scores.

[0048] Interpretable AI components generate diagnostic and treatment recommendations with SHAP value analysis; In the medical field, model interpretability is crucial, as doctors need to understand the rationale behind the model's decisions. SHAP (SHapley Additive Explanations) values ​​are a method for interpreting the predictions of machine learning models. SHAP values ​​are calculated as follows:

[0049] in, Indicates the influence of feature i on the model prediction results, The contribution of M; M is the set of all features; S is a subset of features; Indicates the number of elements in the feature subset S; This represents the model's prediction result under the feature subset S. By calculating the SHAP value of each feature, we can understand the degree of influence of each feature on the model's prediction result, thus providing doctors with interpretable decision-making basis.

[0050] Then, based on the prediction results, clinicians review them according to their own judgment.

[0051] Furthermore, the companionship and emotional support module achieves emotional intervention through the following methods: 1) Edge computing devices collect real-time data on speech spectrum, facial micro-expressions, and heart rate variability; Data acquisition equipment: Voice spectrum acquisition: The edge computing device is equipped with a high-sensitivity microphone, which can clearly acquire the patient's voice signal in different environments. The microphone has noise reduction capabilities to reduce the interference of environmental noise on the voice signal.

[0052] Facial micro-expression capture: Employing a high-definition camera, it boasts rapid capture and high-resolution imaging capabilities, enabling real-time capture of changes in the patient's facial expressions. The camera's frame rate and resolution have been optimized to ensure accurate recognition of even the smallest facial expressions.

[0053] Heart rate variability data acquisition: Heart rate variability data is acquired in real time via Bluetooth connection between wearable devices (such as smart bracelets and smartwatches) and edge computing devices. The wearable devices utilize high-precision sensors to accurately measure heart rate and heart rate variability.

[0054] Data transmission and processing: The collected speech spectrum, facial micro-expression, and heart rate variability data undergo preliminary processing on edge computing devices, such as data cleaning and feature extraction.

[0055] The processed data is transmitted in real time to the system's core server via wireless communication technologies (such as Wi-Fi, Bluetooth, 4G / 5G, etc.) for further analysis and processing.

[0056] 2) An emotion state classification model based on LSTM network outputs an anxiety level score; Specifically, the emotion recognition model employs a multimodal feature fusion method, comprehensively utilizing information from multiple modalities such as audio, facial expressions, and heart rate variability (HRV) to improve the accuracy of emotion recognition. The specific formula is as follows:

[0057] in, Represents the audio feature vector. Represents facial expression feature vectors. Represents the feature vector of heart rate variability; It is a Long Short-Term Memory (LSTM) network used to process sequential data and can capture long-term dependencies in the sequence. Multimodal features are fused using an LSTM network to obtain a comprehensive feature vector h.

[0058] The anxiety level of patients is classified based on the fused feature vector h. The anxiety level classification formula is as follows:

[0059] in, This represents the probability that the patient is at the k-th anxiety level given the feature vector h. It is the weight vector of the k-th anxiety level; the output is converted into a probability distribution by the softmax function, and the anxiety level with the highest probability is selected as the final classification result.

[0060] 3) Dynamically generate progressive reassurance scripts based on the CBT therapy template library; Construction of a CBT therapy template library: Collect and organize common psychological problem scenarios, such as anxiety, depression, and stress, and develop corresponding CBT therapy templates for each scenario.

[0061] The template includes several parts such as opening remarks, guiding questions, cognitive restructuring techniques, and behavioral suggestions, with multiple ways to express each part available.

[0062] Dynamically generated scripts: The system selects appropriate templates from the CBT therapy template library based on the patient's emotional state (assessed through the emotional support module) and specific problems.

[0063] By combining the patient's personal information (such as name, gender, age, etc.) and historical interaction records, the template content is personalized to generate progressive reassurance messages.

[0064] During interactions with patients, the content of the conversation is dynamically adjusted based on their feedback, gradually guiding them to reconstruct their cognition and change their behavior.

[0065] 4) When the suicide risk score is detected to exceed the threshold, the crisis intervention protocol is automatically triggered.

[0066] To promptly detect emotional crises in patients, a risk assessment of their emotional state is necessary. The risk assessment formula is as follows:

[0067] Among them, the negative emotion score in speech indicates the degree of negative emotion in the patient's speech obtained through speech analysis; the frowning frequency indicates the frequency of frowning on the patient's face, reflecting the patient's level of emotional tension; and heart rate variability (SDNN) is an indicator of heart rate variability, reflecting the patient's autonomic nervous system function. A risk assessment value for the patient is obtained by weighted summation of these three indicators. When the risk assessment value exceeds a certain threshold, a crisis intervention mechanism is triggered to provide timely psychological support and assistance to the patient.

[0068] Furthermore, the security protection module is implemented as follows: Fine-grained data access control strategy based on ABE encryption; Attribute-based encryption (ABE) is a novel encryption scheme that allows data to be encrypted and decrypted based on user attributes. The access policy is defined as an LSSS matrix. Where M is a matrix, It is a mapping from matrix row indices to attributes. Decryption requires the following conditions to be met:

[0069] in, It is the set of legal attributes, that is, the set of attributes that the user possesses; For reconstruction coefficients. Only when the user's attribute set Data can only be successfully decrypted when the above conditions are met. This encryption scheme enables fine-grained access control, ensuring data security.

[0070] Homomorphic encryption is used to train the federated learning model; Homomorphic encryption is a special encryption technique that allows specific computations to be performed on ciphertext without decrypting the data. In this system, the CKKS scheme is used for homomorphic encryption computations. The CKKS scheme supports addition and multiplication operations, as detailed below:

[0071]

[0072] in, and There are two ciphertexts. and It is the corresponding plaintext; This indicates a decryption operation. Homomorphic encryption computation allows for data computation and analysis while protecting data privacy.

[0073] An abnormal behavior detection model built according to the MITREATT&CK framework; Frame mapping: A thorough study of the MITREATT&CK framework is conducted, mapping its attack techniques and tactics to the system's business scenarios and security requirements. For example, identifying potential attack surfaces within the system, such as user login, data transmission, and API calls.

[0074] For each attack surface, determine the corresponding ATT&CK technology and tactics, and establish corresponding detection rules.

[0075] Model building: Collect normal behavior data and known abnormal behavior data of the system as training datasets.

[0076] An abnormal behavior detection model is built based on a training dataset using machine learning and deep learning algorithms (such as decision trees and neural networks). The model's input consists of various behavioral characteristics of the system (such as operation time, operation frequency, and data flow direction), and its output is a judgment result indicating whether the behavior is abnormal.

[0077] Model training and optimization: The model is trained using a training dataset, and its performance is evaluated using methods such as cross-validation.

[0078] Regularly collect new abnormal behavior data to update and optimize the model, thereby improving its detection accuracy and generalization ability. Quarterly penetration testing and incident response drills: Penetration testing: Test Plan Development: A detailed penetration testing plan is developed at the beginning of each quarter, clearly defining the test objectives, scope, methods, and schedule. Test objectives include aspects such as system network security, application security, and data security.

[0079] Test Execution: A professional penetration testing team conducts a comprehensive penetration test on the system using various testing techniques (such as vulnerability scanning, port scanning, password cracking, etc.). Vulnerabilities and security risks discovered during the test are recorded.

[0080] Reporting and Remediation: After the test, a detailed penetration test report is generated, including a description of the vulnerability, its risk level, and remediation recommendations. The system development and operations teams promptly remediate the vulnerabilities based on the report's content and verify the findings.

[0081] Emergency response drill: Exercise scenario design: Based on penetration test results and common security incident types, various emergency response exercise scenarios are designed, such as network attacks, data breaches, and system failures.

[0082] Exercise execution: Organize relevant personnel to conduct emergency response drills according to the exercise plan, simulate the actual occurrence of a safety incident, and test the emergency response team's reaction speed and handling capabilities.

[0083] Summary and Improvement: After the exercise, the process was summarized and evaluated, the problems and shortcomings were analyzed, improvement measures were formulated, and the emergency response plan was improved.

[0084] Furthermore, the data analysis module includes: The digital twin modeling unit synchronizes the patient's physiological parameters with time-series data; Digital twins are a technology that combines physical entities with virtual models to predict and optimize the state and behavior of physical entities. In this system, the ARIMA (Autoregressive Integrated Moving Average) model is used to predict physiological parameters. The prediction formula for the ARIMA model is as follows:

[0085] in, represents the predicted value at time t; p and q are the orders of the autoregressive term and the moving average term, respectively; It is the autoregressive coefficient. It is the moving average coefficient; yes The actual value at that moment; yes Error term at time step. Optimal parameters p, d (difference order), and q are selected using the AIC (Akaike Information Criterion) criterion. The AIC criterion is an information criterion used for model selection that comprehensively considers the model's goodness of fit and complexity, selecting the model with the smallest AIC value as the optimal model.

[0086] The dynamic threshold early warning mechanism analyzes the deviation of health indicators through a sliding window; To promptly detect abnormal changes in patients' physiological parameters, a sliding window outlier detection method is employed. The specific formula is as follows:

[0087] in, Represents the physiological parameter value at time t; and These are the mean and standard deviation of physiological parameters for the first 30 days, respectively. (Calculated...) The value can be used to determine whether the current physiological parameter is an outlier. When... When the system detects that the current physiological parameter values ​​are abnormal compared to historical data, it triggers an alarm, reminding doctors to pay close attention to the patient's health status.

[0088] An automated A / B testing framework was used to compare the effectiveness metrics of different intervention programs. Setting up the testing framework: An automated A / B testing framework was built based on the system's business logic and requirements. The framework includes modules for test case management, traffic allocation, data collection, and analysis.

[0089] Define different intervention strategies, such as different treatment recommendations, different reminder methods, and different reassurance phrases.

[0090] Traffic allocation: User traffic is allocated to different intervention plan groups according to preset rules (such as random allocation, allocation based on user characteristics, etc.). This ensures that users in each plan group have similar characteristics and behavioral patterns, guaranteeing the comparability of test results.

[0091] Data collection and analysis: During the testing process, the system automatically collects relevant data for each treatment group, such as effectiveness indicators like user participation, satisfaction, and treatment results.

[0092] Use data analysis tools (such as SQL, Python data analysis libraries, etc.) to perform statistical analysis on the collected data and compare the effectiveness indicators of different intervention programs.

[0093] Results evaluation and decision-making: Based on the data analysis results, evaluate the advantages and disadvantages of different intervention programs. Select the program with the best performance in effectiveness indicators as the final application program, and feed the test results back to relevant business departments to support business decision-making.

[0094] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based accompanying diagnosis management system, characterized by, The system comprises: User management and health record module: for establishing patient personal health records and updating user health records in real time according to diagnosis and treatment results; Intelligent appointment module: for recommending appropriate outpatient registration departments according to personal health records and current patient symptoms, assisting patients in making appointments, and setting reminders; Health advice and diagnosis assistance module: for generating personalized health assistance and diagnosis suggestions for patients based on their health records and common disease templates; Accompanying diagnosis and emotional support module: for providing guided services for patients through voice interaction, helping patients find destination locations, evaluating patients' emotional states through emotional analysis technology, and soothing patients according to the evaluation results; Data analysis and feedback module: for in-depth analysis of patients' health records, identification of possible health risks, generation of personalized health reports, and recommendation of appropriate medical advice or examination suggestions; Security and privacy protection module: for protecting patients' personal information and health records, ensuring that only authorized personnel can access these data, setting different access permissions for different users, and conducting regular privacy protection and data security compliance checks to ensure compliance with medical industry data protection requirements. 2.The AI-based accompanying diagnosis management system of claim 1, wherein: The user management and health record module includes: a biometric identification and two-factor authentication registration / login subsystem; an NLP-based data cleaning engine for processing unstructured data from hospital HIS / LIS systems; a blockchain network that implements automatic EHR data updates through smart contracts, with data storage conforming to HIPAA compliance requirements. 3.The AI-based accompanying diagnosis management system of claim 1, wherein: The intelligent appointment module implements department recommendation through the following steps: constructing an ICD-11 standard symptom-department association matrix, and using a multi-layer perception classification model to process user input symptom features; Based on the Q-learning algorithm, dynamically optimize the appointment time window allocation; through the behavior analysis model, adaptively trigger multi-channel reminders such as SMS, WeChat, and voice. 4.The AI-based accompanying diagnosis management system of claim 1, wherein: The health advice and diagnosis assistance module includes: a medical knowledge retrieval system based on RAG architecture, which accesses UpToDate clinical guidelines in real time; an explainable AI component that generates diagnosis suggestions with SHAP value analysis; then, according to the prediction results, the clinical doctors review according to their own judgment. 5.The AI-based accompanying diagnosis management system according to claim 1, characterized by: The accompanying diagnosis and emotional support module implements emotional intervention through the following methods: edge computing devices collect voice spectrum, facial micro-expression, and heart rate variability data in real time; an emotion state classification model based on LSTM network outputs anxiety level scores; dynamically generate progressive soothing speech based on the CBT therapy template library; when the suicide risk score exceeds the threshold, automatically trigger the crisis intervention protocol. 6.The AI-based accompanying diagnosis management system according to claim 1, characterized by: The security protection module implements: fine-grained data access control strategy based on ABE encryption; federated learning model training using homomorphic encryption; abnormal behavior detection model based on MITREATT&CK framework; quarterly penetration testing and emergency response exercise mechanism. 7.The AI-based accompanying diagnosis management system of claim 1, wherein: The data analysis module comprises: a digital twin modeling unit synchronizing patient physiological parameter time series data; a dynamic threshold early warning mechanism analyzing health index deviation through a sliding window; and an automatic A / B test framework comparing effectiveness indicators of different intervention schemes. 8.A computer readable storage medium storing program instructions for executing the accompanying diagnosis management system based on artificial intelligence technology according to any one of claims 1-7.

Images