An intelligent monitoring and early warning system and method after digestive endoscopy

Abstract

The application discloses a kind of intelligent monitoring and early warning system and method after digestive endoscopy, system includes multi-source data acquisition layer, data transmission and secure storage layer, intelligent analysis and modeling layer, intelligent monitoring and early warning layer, doctor-patient interaction and rehabilitation guidance layer.Patients' physiological data, operation data, basic health data and subjective symptom data are collected, encrypted and transmitted to the cloud for storage using blockchain and distributed database.After data preprocessing and feature fusion, the mixed intelligent model outputs the probability of complications and the rehabilitation status score, and the suggestions are pushed according to the four-level standard classification warning.The dynamic rehabilitation program is generated, and the patient application and the medical management platform are used to implement the program pushing, symptom feedback and online consultation, which improves the postoperative monitoring and intelligent level.

Description

Technical Field

[0001] This invention belongs to the field of postoperative intelligent monitoring technology, and relates to an intelligent monitoring and early warning system and method for postoperative digestive endoscopy. Background Technology

[0002] Digestive endoscopy is a commonly used minimally invasive treatment in gastroenterology and is widely applied in clinical practice. However, postoperative recovery is affected by a variety of factors, including the patient's underlying medical conditions, the details of the surgical procedure, and postoperative care. These factors are intertwined, making it difficult for doctors to accurately predict the patient's prognosis.

[0003] Traditional postoperative monitoring and care after digestive endoscopy relies primarily on the clinical experience of medical staff to implement preventative measures. This approach lacks scientific and systematic tools, often leading to overtreatment or delayed intervention, and fails to meet the needs of patients requiring meticulous monitoring and care. Furthermore, current technology lacks tools that can comprehensively consider multiple risk factors, making it difficult to scientifically predict the risk of postoperative complications.

[0004] In terms of rehabilitation management, the traditional model relies on guidance from medical staff and patient self-monitoring, which has significant shortcomings. During home-based rehabilitation, medical staff struggle to obtain timely data on patients' physical indicators and cannot monitor their progress in real time. Patients, lacking professional knowledge, often struggle to accurately describe their symptoms, leading to poor communication between doctors and patients. Furthermore, rehabilitation effectiveness is difficult to assess scientifically, and rehabilitation plans cannot be dynamically adjusted according to the patient's actual condition. These problems can all hinder the rehabilitation process and affect the overall treatment outcome. Summary of the Invention

[0005] To address the problems existing in the background art, the present invention proposes an intelligent monitoring and early warning system and method after digestive endoscopy.

[0006] The first aspect of this application provides an intelligent monitoring and early warning system after digestive endoscopy, comprising: a multi-source data acquisition layer, a data transmission and secure storage layer, an intelligent analysis and modeling layer, an intelligent monitoring and early warning layer, and a doctor-patient interaction and rehabilitation guidance layer; each layer is linked through a standardized interface protocol; The multi-source data acquisition layer is used to collect patients' physiological data, surgical data, basic health data, and subjective symptom data. The data transmission and secure storage layer is used to receive collected data and encrypt it for secure transmission to the cloud for storage. The intelligent analysis and modeling layer is used to preprocess and extract features from the stored data, and then output the probability of complication risk and rehabilitation status score through a hybrid intelligent model. The intelligent monitoring and early warning layer is used to receive the above output results and provide graded early warnings. The doctor-patient interaction and rehabilitation guidance layer is used to generate personalized rehabilitation plans based on early warning results and scores, while also enabling doctor-patient interaction.

[0007] Optionally, the multi-source data acquisition layer includes a wearable acquisition module, a medical device docking module, a hospital information system data access module, and a patient self-reporting module; The wearable data acquisition module collects heart rate, blood pressure, blood oxygen saturation, body temperature, activity level, and gastrointestinal motility. The medical device docking module collects data on operation duration, surgical method, intraoperative blood loss, intraoperative mucosal resection range, and immediate postoperative pathological results. The hospital information system's data access module extracts past medical history, allergy history, family medical history, and preoperative liver and kidney function indicators; The patient self-reporting module receives subjective symptom data and wound photos.

[0008] Optionally, the data transmission and secure storage layer includes a real-time transmission unit, a secure storage unit, and a data synchronization unit; The real-time transmission unit uses Bluetooth, wireless LAN, and 5G mobile communication technologies, and adaptively switches the transmission mode according to the data type and application scenario. The transmission process uses end-to-end advanced encryption standard 256-bit encryption. The secure storage unit adopts a storage architecture that combines blockchain and distributed database. Blockchain nodes store data hash values ​​and access logs, while the distributed database stores the original data. The data synchronization unit uses an incremental synchronization algorithm and is equipped with a data caching mechanism.

[0009] Optionally, the intelligent analysis and modeling layer includes a data preprocessing submodule, a feature extraction submodule, and a hybrid intelligent model construction submodule; The data preprocessing submodule uses the Isolation Forest algorithm to remove outliers, the Long Short-Term Memory time series prediction algorithm to fill missing values, and the min-max standardization to normalize the data. The feature extraction submodule extracts physiological indicator trend features, surgical trauma features, and text features, and fuses these features through an attention mechanism; The hybrid intelligent model construction submodule constructs a logistic regression model as the base model, a stacked ensemble algorithm as the ensemble learning layer, and a clinical rule base as the calibration model to form a three-level prediction model. At the same time, a hybrid neural network model of convolutional neural network and long short-term memory is constructed. The three-level prediction model outputs the probability of complication risk, and the hybrid neural network model of convolutional neural network and long short-term memory outputs the rehabilitation status score.

[0010] Optionally, the surgical trauma characteristics are calculated using the surgical trauma index formula, which is: I trauma =α×V blood+β×S+γ×T op Among them, I trauma V represents the surgical trauma index, α, β, and γ are the clinical training weights, and V... blood S represents intraoperative blood loss, S represents the extent of mucosal resection during surgery, and T represents intraoperative blood loss. op This refers to the duration of the surgery.

[0011] Optionally, the intelligent monitoring and early warning layer includes a real-time monitoring unit, a graded risk assessment unit, a dynamic threshold adjustment unit, and an early warning information push unit; The real-time monitoring unit uses a streaming computing framework to perform second-level data analysis; The risk assessment unit divides the risk level into four levels based on the probability of complications and the recovery status score; The dynamic threshold adjustment unit adjusts the warning threshold based on the patient's age, underlying diseases, and type of surgery. The early warning information push unit pushes early warning information and suggested measures to both medical staff and patients through pop-up windows on the medical staff management platform, SMS, application push, and voice reminders.

[0012] Optionally, the doctor-patient interaction and rehabilitation guidance layer includes a personalized rehabilitation plan generation unit, a patient-side application, and a medical staff-side management platform; The personalized rehabilitation plan generation unit uses a combination of rule-based reasoning and case-based reasoning to generate dynamic rehabilitation plans; The patient-side application supports data viewing, treatment plan reception, symptom reporting, and online consultation. The healthcare management platform supports centralized patient management, data visualization, early warning processing, and treatment plan adjustments.

[0013] A second aspect of this application provides a method for intelligent monitoring and early warning after digestive endoscopy, including: collecting patient physiological data, surgical data, basic health data and subjective symptom data; All collected data is encrypted and transmitted to the cloud, where it is securely stored using an architecture that combines blockchain and a distributed database. The isolated forest algorithm was used to remove outliers from all collected data, the long short-term memory time series prediction algorithm was used to fill missing values, and the min-max standardization was used to normalize the data. Physiological indicator trend features, surgical trauma features, and text features were extracted and fused through an attention mechanism. Based on the fusion feature training, a three-level hybrid intelligent model, a convolutional neural network and a long short-term memory hybrid neural network model are trained to output the probability of complication risk and rehabilitation status score; Based on the four-level risk classification standard, the corresponding early warning mechanism is triggered, and early warning information and suggested measures are pushed out; Dynamic rehabilitation plans are generated, and the plans are pushed out, symptoms are reported, and online consultations are conducted through patient-side applications and medical staff management platforms.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention provides an intelligent monitoring and early warning system and method after digestive endoscopy. Through multi-source data acquisition, it achieves comprehensive coverage of patient physiological data, surgical data, basic health data, and subjective symptom data. In particular, it incorporates gastrointestinal motility frequency into the monitoring scope, filling a gap in traditional monitoring data dimensions and providing complete and reliable data support for subsequent analysis. Data transmission employs end-to-end encryption technology, and storage combines blockchain and a distributed database architecture, ensuring both the security of data transmission and the immutability and traceability of data, effectively protecting patient medical privacy.

[0015] The intelligent analysis and modeling layer improves data quality and feature representativeness through scientific data preprocessing and feature fusion techniques. The synergistic application of a three-level hybrid intelligent model and a hybrid convolutional neural network and long short-term memory neural network model significantly improves the accuracy of complication risk probability prediction and the reliability of rehabilitation status scoring, providing a scientific basis for risk assessment. A four-level graded early warning mechanism, combined with dynamic threshold adjustment, enables precise risk identification and timely notification. Multi-channel early warning methods ensure that both medical staff and patients receive information quickly, allowing time for timely intervention.

[0016] The personalized rehabilitation plan generation unit employs a combination of rule-based and case-based reasoning methods to generate dynamic rehabilitation plans that can be tailored to individual patient circumstances and adjusted in real time based on rehabilitation progress. The patient-side application and the healthcare management platform establish an efficient doctor-patient interaction channel, enabling real-time information exchange and improving patient adherence to rehabilitation and healthcare staff efficiency.

[0017] Overall, this invention achieves refined, intelligent, and personalized postoperative monitoring of digestive endoscopy, effectively reducing the incidence of postoperative complications, shortening the patient recovery period, improving the quality of medical services and patient satisfaction, and has significant clinical application value and prospects for promotion. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an intelligent monitoring and early warning system after digestive endoscopy according to one embodiment of the present invention; Figure 2 This is a flowchart of a method for intelligent monitoring and early warning after digestive endoscopy in one embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In one embodiment, such as Figure 1 As shown, a smart monitoring and early warning system for post-digestive endoscopy is provided. This system corresponds one-to-one with the smart monitoring and early warning methods for post-digestive endoscopy described in the above embodiments. The smart monitoring and early warning system for post-digestive endoscopy includes: a multi-source data acquisition layer, a data transmission and secure storage layer, an intelligent analysis and modeling layer, a smart monitoring and early warning layer, and a doctor-patient interaction and rehabilitation guidance layer. Each layer is linked through a standardized interface protocol. Detailed descriptions of each functional module are as follows: In this invention, a multi-source data acquisition layer is used to collect patient physiological data, surgical data, basic health data, and subjective symptom data; a data transmission and secure storage layer is used to receive the collected data and encrypt and transmit it to the cloud for secure storage; an intelligent analysis and modeling layer is used to preprocess the stored data, extract features, and output complication risk probabilities and rehabilitation status scores through a hybrid intelligent model; an intelligent monitoring and early warning layer is used to receive the above output results and provide graded early warnings; and a doctor-patient interaction and rehabilitation guidance layer is used to generate personalized rehabilitation plans based on the early warning results and scores, while also enabling doctor-patient interaction.

[0021] The multi-source data acquisition layer is responsible for comprehensively collecting relevant data from patients after digestive endoscopy. Through the collaborative work of multiple clearly defined modules, it ensures the comprehensiveness, accuracy, and continuity of data collection. This acquisition layer comprises four core modules, each employing targeted acquisition methods based on the data type and acquisition scenario, collectively covering multi-dimensional information such as the patient's physiological state, surgical procedure, underlying health background, and postoperative subjective feelings.

[0022] The wearable data acquisition module is a portable monitoring device integrating multiple sensors, which patients need to wear continuously both in the hospital and at home during postoperative rehabilitation. The device uses a built-in heart rate sensor to collect real-time heart rate data, a blood pressure sensor to obtain systolic and diastolic blood pressure data non-invasively, a blood oxygen saturation sensor to monitor the oxygen concentration in the blood, a body temperature sensor to collect surface or core body temperature, a motion sensor to record the patient's daily steps and changes in body position, and a vibration sensor to collect abdominal micro-vibration signals through a design that fits snugly against the abdomen, which are then converted into gastrointestinal motility frequency data. The module's acquisition frequency is dynamically adjusted according to the data type: heart rate, blood pressure, blood oxygen saturation, and body temperature are collected every five minutes; activity level is calculated every hour; and gastrointestinal motility frequency is collected every ten minutes, ensuring that the data reflects the real-time status without increasing device power consumption due to excessive acquisition frequency.

[0023] The medical device interface module establishes data connections with digestive endoscopy equipment, intraoperative monitors, and pathology testing equipment through standardized interface protocols. During the surgery, this module extracts real-time surgical duration data from the surgical equipment, i.e., the continuous time from the start to the end of the surgery; determines the surgical procedure based on the operating parameters of the surgical equipment; calculates intraoperative blood loss through the flow monitoring function of the intraoperative monitor; calculates the area of ​​mucosal resection during the surgery based on the positioning data of the surgical equipment; and immediately obtains preliminary pathological results from the pathology testing equipment after the surgery to determine key information such as whether there are residual lesions. The data acquisition of this module is synchronized with the surgical process, and is automatically summarized and uploaded after the surgery, avoiding errors that may occur due to manual recording.

[0024] The hospital information system data access module connects to the hospital's existing information management system through the hospital's intranet authentication mechanism. This module extracts patients' past medical history information from the system, including whether they have chronic diseases such as hypertension, diabetes, and cirrhosis, as well as the diagnosis time and treatment details; it also extracts patients' allergy history information, identifying allergies to medications, foods, or other substances; it extracts patients' family medical history information, recording whether there are hereditary digestive diseases or other related diseases in immediate family members; and it extracts patients' preoperative liver and kidney function indicators, including liver function tests such as transaminases and bilirubin, and kidney function tests such as creatinine and blood urea nitrogen. This data provides important reference for assessing patients' postoperative recovery potential and risk tolerance.

[0025] The patient self-reporting module collects data through a patient-side application, requiring no professional operation from the patient. Post-operatively, patients report subjective symptom data based on their own experience, using preset options or input functions within the application, including the degree of abdominal pain, frequency of nausea and vomiting, and stool characteristics. For patients with external wounds or nasogastric tube insertion sites post-surgery, the application's camera function can be used to capture images of the wound or insertion site, which are then uploaded to the system as visual data on the wound's condition. The application includes a timed reminder function to encourage timely reporting, ensuring the continuity of subjective symptom data. Simultaneously, the image capture function provides a visual record of the wound's condition, compensating for the inability of medical staff to monitor wounds in real-time at home.

[0026] For example, 100 patients undergoing digestive endoscopy at a hospital were selected. Post-operatively, each patient wore a wearable data collection device to continuously collect data on heart rate, blood pressure, blood oxygen saturation, body temperature, activity level, and gastrointestinal motility. The medical device interface module simultaneously collected the surgical duration for each patient, ranging from a minimum of 35 minutes to a maximum of 90 minutes. Surgical procedures included endoscopic mucosal resection and endoscopic submucosal dissection. Intraoperative blood loss ranged from 5 ml to 50 ml, and the area of ​​mucosal resection ranged from 1 square centimeter to 5 square centimeters. Immediate postoperative pathology results showed minimal residual lesions in 8 cases. The hospital information system data access module extracted data showing that 30 patients had a history of hypertension, 15 had a history of diabetes, 5 had a history of drug allergies, and 20 had a family history of digestive tract diseases. Preoperative liver and kidney function data for all patients were also obtained. Post-operatively, patients reported subjective symptoms daily via an application. 25 patients reported mild abdominal pain on the first postoperative day, 8 patients reported slight nausea without vomiting, and 5 patients uploaded photos of the nasogastric tube interface.

[0027] Comprehensive data acquisition from multiple sources enables all-round monitoring of patients' postoperative condition. The wearable acquisition module provides gastrointestinal motility frequency data, directly reflecting the recovery of postoperative gastrointestinal function and providing a basis for assessing the risk of abdominal distension or intestinal obstruction. The medical device docking module collects surgical-related data, accurately recording key surgical parameters and providing quantitative indicators for assessing the degree of surgical trauma. The hospital information system data access module extracts basic health data, helping medical staff understand individual patient differences and providing background support for risk assessment. The patient self-reporting module collects subjective symptoms and wound images, enabling medical staff to promptly grasp patients' self-perception and local recovery status, compensating for the inability to directly observe patient conditions in remote monitoring. The data collected by each module complement each other, forming a complete postoperative patient data archive. This provides a comprehensive and reliable data foundation for subsequent data analysis, modeling, risk warning, and rehabilitation guidance, effectively improving the scientific rigor of the data support and laying the data foundation for refined monitoring.

[0028] The core function of the data transmission and secure storage layer is to receive various types of data acquired from the multi-source data acquisition layer, upload the data to the cloud through secure and efficient transmission methods, and ensure the security, integrity, and accessibility of the data using a reliable storage architecture. This layer comprises three functional units that work collaboratively to solve the data transmission adaptation problem in different scenarios and to build a multi-layered data security protection system, ensuring the reliability of data throughout the entire link from the acquisition end to the storage end.

[0029] The real-time transmission unit supports adaptive switching between three transmission technologies, selecting the optimal transmission method based on data type and application scenario. When patients are in home rehabilitation, low-volume physiological data such as heart rate and blood pressure collected by the wearable acquisition module are transmitted via Bluetooth, a technology known for its low power consumption and stable connection, meeting the battery life requirements of portable devices. When patients are in a hospital setting, large amounts of surgical and historical health data generated by the medical device interface module and the hospital information system data access module are transmitted via Wi-Fi, a technology with high transmission rates that can quickly synchronize massive amounts of data. When patients need to upload image data such as wound photos, the system automatically switches to 5G mobile communication technology, which offers high bandwidth and shortens the transmission time for large amounts of data such as images. All data is encrypted using an end-to-end advanced encryption standard (256-bit) during transmission. The encryption process covers the entire data link from the acquisition device to the cloud, ensuring that only authorized receivers can decrypt and obtain the original data, effectively preventing data theft or tampering during transmission.

[0030] The secure storage unit employs a hybrid storage architecture combining blockchain and a distributed database, with each storage method performing its specific function while working in tandem. Blockchain nodes primarily store data hash values ​​and access logs. The hash value is calculated from the original data using an encryption algorithm, with each data entry corresponding to a unique hash value. Any modification to the original data will change the corresponding hash value, allowing for rapid verification of data integrity by comparing hash values. Access logs meticulously record the time, accessing entity, and access operation for each data access, ensuring traceability of the data access process. The distributed database stores all original data, distributed across multiple physical nodes. Each node stores a portion of the data, and even if a single node fails, the complete data can be recovered from other nodes, preventing data loss. Furthermore, this storage architecture features a tiered access control mechanism: medical staff can only access the data of the patients they are responsible for; patients can access all of their own data; and third-party institutions must obtain patient authorization and hospital approval before accessing anonymized data, ensuring patient privacy and security.

[0031] The data synchronization unit employs an incremental synchronization algorithm, transmitting and synchronizing only the changed data segments, rather than transmitting all data each time. For example, in patient heart rate data, only the values ​​that have changed compared to the previous acquisition are transmitted; unchanged data does not need to be transmitted repeatedly, significantly reducing the total amount of data transmitted and lowering network bandwidth consumption. The unit also features a data caching mechanism. When the network is interrupted or the signal is unstable, the acquisition device and the patient-side application automatically store the acquired data temporarily in a local cache. Once the network is restored, the cached data is automatically re-uploaded to the cloud in the order of acquisition time, ensuring that data is not lost due to network problems and guaranteeing the continuity and integrity of data acquisition.

[0032] For example, 50 patients who underwent digestive endoscopy at a hospital were selected. Post-operatively, each patient wore a wearable data collection device, and all functions of the data transmission and secure storage layers were activated. In the hospital setting, the medical device interface module transmitted the surgical data of all 50 patients to the cloud in real time via Wi-Fi. Data transmission time for each patient did not exceed 3 seconds, with no data transmission delays or loss. After discharge and entering the home rehabilitation phase, the wearable data collection device transmitted physiological data every five minutes via Bluetooth. A single charge allowed for seven days of continuous use, meeting the long-term monitoring needs during home rehabilitation. Ten patients needed to upload wound photos; the system automatically switched to 5G mobile communication technology, with each photo transmission time within two seconds. All transmitted data underwent 256-bit encryption using Advanced Encryption Standard (AES), and no data leakage or tampering occurred during transmission. For secure storage, all data from the 50 patients was stored in a distributed database, with blockchain nodes recording the hash value and access logs of each data entry. A simulated failure of a single storage node was successfully recovered through other nodes, and data integrity remained unaffected. Three patients experienced network outages during their home recovery, with outage durations of 20 minutes, 30 minutes, and 1 hour, respectively. After the network was restored, all cached data was automatically and successfully retransmitted without any data loss.

[0033] Through the aforementioned design of the data transmission and secure storage layers, high efficiency and security in data transmission, as well as reliability and integrity in data storage, are achieved. The adaptive switching transmission mode not only meets the transmission needs of different data types and scenarios but also balances transmission rate and device power consumption, significantly improving the battery life of wearable devices and ensuring continuous data collection during home-based rehabilitation. End-to-end encryption and hierarchical access control mechanisms effectively prevent security risks during data transmission and storage, protecting patient medical privacy. The storage architecture combining blockchain and distributed databases ensures data immutability and traceability while preventing data loss due to the failure of a single storage node. Incremental synchronization algorithms and data caching mechanisms reduce network bandwidth consumption and solve the problem of data loss that may occur due to network interruptions, providing comprehensive, continuous, and reliable data support for subsequent data analysis, modeling, and risk warning.

[0034] The intelligent analysis and modeling layer is the core of risk prediction and rehabilitation assessment for patients after digestive endoscopy. Its core function is to perform in-depth processing and intelligent analysis of stored multi-source data. Through standardized data preprocessing, multi-dimensional feature extraction, and precise model calculation, it ultimately outputs key indicators that reflect the patient's postoperative condition. This layer contains three closely connected sub-modules, each working sequentially to form a complete processing chain from raw data to prediction results, ensuring the accuracy and reliability of the output.

[0035] The data preprocessing submodule addresses potential issues of missing values ​​and inconsistencies in units of measurement during multi-source data acquisition by employing three targeted processing algorithms. First, the Isolation Forest algorithm is used to remove outliers. This algorithm constructs multiple decision trees to identify data with significantly different distributions from the majority of the data as outliers, such as sudden increases in heart rate due to sensor malfunction or abnormal bleeding values ​​caused by operational errors. These outliers are directly removed after identification to prevent them from affecting subsequent analysis. Second, a Long Short-Term Memory (LSTM) time series prediction algorithm is used to fill in missing values. When patients experience data loss due to temporary equipment disconnection or forgetting to report, this algorithm uses the time series characteristics of the data, combined with valid data from preceding and following time periods, to predict and fill in the missing portions, ensuring data continuity. For example, if a patient did not report their activity level for a particular hour, the algorithm can predict the value for that period based on activity data from the previous two hours and the following hour. Third, min-max normalization is used for data normalization, mapping raw data with different units of measurement to the same numerical range. For example, heart rate per minute and blood loss in milliliters are uniformly converted to values ​​within the same range, eliminating the impact of unit differences on model calculations and allowing various types of data to directly participate in subsequent feature fusion and model training.

[0036] The feature extraction submodule extracts key features reflecting the patient's postoperative condition and risk from the preprocessed standardized data and effectively integrates them. When extracting physiological indicator trend features, a sliding window combined with wavelet transform is used to analyze the fluctuations and trends of physiological indicators such as heart rate and blood pressure over different time periods. For example, it extracts the variability of heart rate within one hour and the slope of body temperature changes within twenty-four hours, capturing the dynamic changes in physiological state. When extracting surgical trauma features, a surgical trauma index formula is used for quantitative calculation. The formula includes three key parameters: intraoperative blood loss, intraoperative mucosal resection range, and operation duration. The weight coefficients of each parameter are determined through clinical training, transforming discrete surgical data into a single indicator that can quantify the degree of surgical trauma. When extracting textual features, textual data such as the patient's past medical history and subjective symptom descriptions are mined to identify key information related to postoperative recovery and complications. For example, information on chronic diseases such as cirrhosis is extracted from the medical history, and key expressions such as persistent abdominal pain are extracted from the symptom descriptions. After feature extraction, the three types of features are fused through an attention mechanism. This mechanism can automatically calculate the correlation between various features and the risk of complications and the recovery status, assign higher weights to features with high correlation, highlight the influence of key features, and suppress the interference of irrelevant features to form more representative fused features.

[0037] Surgical trauma characteristics are a core quantitative indicator reflecting the degree of trauma caused to the patient's body by digestive endoscopic surgery. This is calculated using the surgical trauma index formula, I... trauma =α×V blood +β×S+γ×T op Among them, I trauma V represents the surgical trauma index, α, β, and γ are the clinical training weights, and V... blood S represents intraoperative blood loss, S represents the extent of mucosal resection during surgery, and T represents intraoperative blood loss. op The formula represents the duration of the surgery. It comprehensively considers three key factors affecting surgical trauma, transforming discrete surgery-related data into a unified quantitative indicator, providing a precise quantitative basis for subsequent risk prediction and rehabilitation assessment.

[0038] The surgical trauma index directly corresponds to the severity of surgical trauma; a higher value indicates more severe trauma to the patient. Clinical training weights are fixed coefficients determined through extensive clinical data training, used to reflect the importance of three parameters—intraoperative blood loss, intraoperative mucosal resection area, and surgical duration—on surgical trauma. Intraoperative blood loss is the total blood loss during surgery, directly collected by the medical device interface module from the intraoperative monitor's flow monitoring function. Intraoperative mucosal resection area is the area of ​​mucosal tissue removed during surgery, calculated by the medical device interface module based on the surgical equipment's positioning data. Surgical duration is the continuous time from the start to the end of surgery, extracted in real-time by the medical device interface module from the surgical equipment.

[0039] The determination of clinical training weights requires statistical analysis of large-sample clinical data. Surgical-related data from thousands of patients undergoing digestive endoscopy were collected, including intraoperative blood loss, extent of mucosal resection, operative time, postoperative complications, and recovery time. Statistical methods were used to analyze the correlation between each parameter and the incidence of postoperative complications and recovery time, and initial weights for each parameter were determined based on the strength of the correlation. The initial weights were then used to calculate the surgical trauma index, which was compared with the actual postoperative recovery of patients. The weight values ​​were repeatedly adjusted until the optimal match between the surgical trauma index and postoperative recovery was achieved. After extensive training with clinical data, the specific values ​​of the clinical training weights were finally determined. The weight corresponding to intraoperative blood loss was the highest, followed by the weight corresponding to the extent of intraoperative mucosal resection, and the weight corresponding to operative time was the lowest. This weight allocation reflects the actual impact of each factor on surgical trauma in clinical practice.

[0040] For example, 300 patients undergoing digestive endoscopy at a hospital were selected, including 150 patients undergoing endoscopic mucosal resection and 150 patients undergoing endoscopic submucosal dissection. Intraoperative blood loss, the extent of mucosal resection, and the duration of the operation were collected for each patient via a medical device docking module. After determining the clinical training weights, the weight corresponding to intraoperative blood loss was 0.6, the weight corresponding to the extent of mucosal resection was 0.3, and the weight corresponding to the duration of the operation was 0.1. The first patient had an intraoperative blood loss of 10 ml, an intraoperative mucosal resection area of ​​2 square centimeters, and an operation duration of 40 minutes. Substituting these values ​​into the formula, the surgical trauma index was calculated as 0.6×10 + 0.3×2 + 0.1×40, resulting in 6 + 0.6 + 4 = 10.6. The second patient experienced intraoperative blood loss of 30 ml, with an intraoperative mucosal resection area of ​​4 square centimeters and a surgical duration of 70 minutes. The calculated surgical trauma index was 0.6×30 + 0.3×4 + 0.1×70, resulting in 18 + 1.2 + 7 = 26.2. Comparing the postoperative recovery of the two patients, the first patient experienced no complications and recovered in 7 days; the second patient experienced minor postoperative bleeding and recovered in 12 days. This difference aligns with the surgical trauma index, indicating that a higher surgical trauma index correlates with greater postoperative recovery difficulty.

[0041] The application of this surgical trauma index formula enables precise quantification of the degree of surgical trauma. Traditional methods often present surgical-related data as discrete, independent indicators, failing to directly and comprehensively reflect the extent of trauma. This formula organically integrates three key parameters, transforming them into a unified quantitative index, making the degree of surgical trauma comparable across different patients. The determination of clinical training weights is based on extensive clinical data, ensuring the scientific rigor and accuracy of the formula's calculations and accurately reflecting the actual impact of various factors on surgical trauma. This quantitative index provides more representative feature data for feature extraction, enabling subsequent hybrid intelligent models to more accurately identify the correlation between surgical trauma and complication risk / recovery status, improving the accuracy of risk prediction and rehabilitation assessment. Simultaneously, this index provides medical staff with an intuitive basis for assessing surgical trauma, helping them to develop more targeted rehabilitation plans post-surgery. For patients with high surgical trauma indices, more cautious nursing plans and interventions can be developed to reduce the risk of complications and promote patient recovery.

[0042] The hybrid intelligent model construction submodule achieves complication risk prediction and rehabilitation status assessment through the collaborative work of two types of models. A three-level prediction model is constructed to output the probability of complication risk. This model uses a logistic regression model as the base model to make preliminary predictions for different complications such as postoperative bleeding, infection, and perforation, outputting the probability of occurrence of a single complication. A stacked ensemble algorithm is used as the ensemble learning layer, taking the prediction results of multiple base models as input to train a second-layer classifier, combining the predictive advantages of various base models to output a more accurate comprehensive probability of complication risk. A clinical rule base is used as a calibration model, correcting the output of the ensemble learning layer based on clinical treatment guidelines and expert experience. For example, if the patient has cirrhosis and abnormal coagulation function, the probability of postoperative bleeding is increased to ensure that the prediction results conform to clinical reality. A hybrid neural network model combining convolutional neural networks and long short-term memory was constructed to output a rehabilitation status score. This model takes fused features as input, extracts local correlation information of features through convolutional neural networks, and captures the time-series dependencies of features through long short-term memory networks. It comprehensively evaluates the patient's overall rehabilitation status and outputs a corresponding score, with the score range covering from 0 to 100. At the same time, it identifies key factors affecting rehabilitation, such as insufficient activity or large fluctuations in blood pressure.

[0043] For example, multi-source data from 100 patients who underwent digestive endoscopy at a hospital were selected. After processing by the data preprocessing submodule, 23 abnormal data points were removed, 18 missing data points were filled, and all data were normalized, meeting the requirements for subsequent analysis. The feature extraction submodule extracted 15 physiological trend features, 3 surgical trauma features, and 8 text features for each patient from the preprocessed data. These features were then fused using an attention mechanism to form 26 fused features. The hybrid intelligent model construction submodule used these fused features for training. The three-level prediction model showed that 8 out of 100 patients were predicted to be at high risk for postoperative bleeding complications, and these 8 patients all developed bleeding symptoms of varying degrees in subsequent clinical observation. In the prediction of infectious complications, 6 high-risk patients were diagnosed with infection, while the remaining patients did not develop related complications. The rehabilitation status score output by the hybrid neural network model of convolutional neural network and long short-term memory had a 93% consistency with the rehabilitation assessment conclusion given by medical staff based on clinical examination results. It also accurately identified the key factors for slow recovery in 32 patients, including insufficient activity in 20 cases and large fluctuations in blood pressure in 12 cases.

[0044] The above-mentioned design of the intelligent analysis and modeling layers significantly improves the accuracy of complication risk prediction and the reliability of rehabilitation status assessment. Data preprocessing ensures the quality of input data, providing a solid foundation for subsequent analysis and avoiding interference from abnormal missing data. Feature extraction, through multi-dimensional feature capture and attention mechanism fusion, enables the model to focus on key information and improves the representativeness of features. The three-level prediction model combines the comprehensiveness of targeted ensemble learning of the base model with the practicality of clinical rules, effectively solving the problem of insufficient prediction accuracy of a single model, making complication risk prediction more in line with actual clinical needs. The hybrid convolutional neural network and long short-term memory neural network model fully leverages the advantages of both types of networks, comprehensively assessing the patient's rehabilitation status and identifying key influencing factors. This provides accurate basis for subsequent graded early warning and personalized rehabilitation guidance, helping medical staff to promptly identify potential risks and take early intervention measures, while simultaneously developing more targeted rehabilitation plans for patients and promoting the rehabilitation process.

[0045] The intelligent monitoring and early warning layer is the core component for real-time monitoring and risk warning of patients after digestive endoscopy. Its core function is to receive complication risk probabilities and recovery status scores from the intelligent analysis and modeling layer. Through continuous monitoring, accurate assessment, and timely push notifications, it provides targeted risk alerts and intervention basis for both medical staff and patients. This early warning layer comprises four collaborative units, each operating in a logical sequence of data monitoring, risk assessment, threshold adaptation, and information push notifications to ensure the real-time nature, accuracy, and effectiveness of the early warnings.

[0046] The real-time monitoring unit employs a streaming computing framework to process various types of data. This framework enables uninterrupted, second-level analysis of continuously transmitted physiological data, feature data, and model outputs. After data is transmitted to this unit, the framework does not wait for all data to be aggregated but processes it in real-time according to the order of data generation, performing a data comparison and status assessment every second. This unit compares the real-time collected data with dynamically adjusted warning thresholds one by one, while simultaneously correlating it with the latest complication risk probabilities and rehabilitation status scores output by the intelligent analysis and modeling layer. This allows for timely detection of abnormal data changes and fluctuations in risk levels, ensuring that any potential risks are quickly identified and providing time for subsequent warning actions.

[0047] The risk assessment unit uses complication risk probability and recovery status score as the core criteria to establish a four-level risk classification standard. Level 1 is low risk, corresponding to a complication risk probability of less than 10% and a recovery status score greater than 80, indicating that the patient's postoperative recovery is stable with no significant risk of complications. Level 2 is medium risk, corresponding to a complication risk probability greater than or equal to 10% and less than 30%, and a recovery status score greater than or equal to 60 and less than or equal to 80, indicating that the patient has a slight risk and requires monitoring for changes in their condition. Level 3 is high risk, corresponding to a complication risk probability greater than or equal to 30% and less than 50%, and a recovery status score greater than or equal to 40 and less than 60, indicating that the patient has a relatively high risk and requires timely intervention. Level 4 is very high risk, corresponding to a complication risk probability greater than or equal to 50% and a recovery status score less than 40, indicating that the patient has a serious risk of complications and requires immediate medical intervention. This unit reassesses the risk level every five minutes to ensure that the risk level dynamically reflects the patient's latest status.

[0048] The dynamic threshold adjustment unit adjusts the warning thresholds for various monitoring indicators in real time based on individual patient differences. This unit acquires patient age data, as normal ranges for physiological indicators differ across age groups; for example, the normal blood pressure range for elderly patients is higher than that for younger patients. It also acquires information on underlying medical conditions, such as hypertension and diabetes, for which the warning thresholds for related indicators need to be adjusted based on disease control. Furthermore, it acquires information on the type of surgery performed, as different surgical procedures cause varying degrees of trauma, resulting in different warning standards for corresponding indicators. Through a pre-set algorithm model, this unit converts factors such as age, underlying medical conditions, and type of surgery into adjustment coefficients, personalizing the general warning thresholds to better reflect the individual patient's situation and avoid false alarms or missed alarms caused by uniform thresholds.

[0049] The early warning information push unit simultaneously pushes early warning information and recommended measures to medical staff and patients through multiple fixed channels. When pushing to medical staff, the early warning content is directly displayed in a pop-up window on the medical staff management platform, and reminders are also sent via SMS and application push, ensuring that medical staff can receive the information in a timely manner regardless of whether they are on the platform interface. When pushing to patients, the information is delivered through application push and voice reminders. The voice reminder uses a clear and concise broadcast format, so that patients can be informed of the early warning even if they are not checking the application. The pushed early warning information includes core content such as risk level, abnormal indicator name, complication risk probability, and recovery status score. The recommended measures are formulated according to the risk level and abnormal situation. For example, for low risk, it is recommended to monitor data regularly; for medium risk, it is recommended to reduce activity and closely monitor symptoms; for high risk, it is recommended to contact medical staff in time; and for very high risk, it is recommended to go to the hospital immediately.

[0050] For example, 80 patients who underwent digestive endoscopy at a hospital were selected, and a smart monitoring and early warning layer was used for postoperative monitoring. The real-time monitoring unit analyzed all patients' data in seconds using a streaming computing framework. In 12 patients, blood pressure fluctuations were detected, and the unit captured the abnormalities within one second, triggering subsequent assessments. During monitoring, the risk assessment unit classified 35 patients as Level 1 risk, 28 as Level 2 risk, 12 as Level 3 risk, and 5 as Level 4 risk. The dynamic threshold adjustment unit raised the blood pressure warning threshold for 5 elderly hypertensive patients compared to the general standard, and lowered the risk warning threshold related to the extent of mucosal resection during surgery for 3 patients undergoing endoscopic submucosal dissection, ensuring the thresholds were appropriate for individual circumstances. The early warning information push unit sent platform pop-ups, SMS messages, and application push notifications to medical staff for the 12 Level 3 risk patients, and sent application push notifications and voice reminders to the patients, advising them to contact their attending physician. For the 5 Level 4 risk patients, in addition to the above push notifications, the medical staff management platform also triggered an audible and visual alarm at the nurses' station, advising the patients to seek immediate medical attention. Subsequent clinical follow-up showed that 12 patients at level 3 risk did not develop serious complications after timely intervention, and 5 patients at level 4 risk were out of danger after emergency treatment.

[0051] Through the aforementioned design of the intelligent monitoring and early warning layer, accurate identification and timely warning of postoperative risks are achieved. The real-time monitoring unit's second-level analysis capability ensures timely risk detection, avoiding intervention delays caused by monitoring latency. The four-level classification standard of the tiered risk assessment unit makes risk levels clearer, enabling medical staff and patients to quickly assess the severity of risks. The personalized correction function of the dynamic threshold adjustment unit effectively reduces the false alarm and false negative rates of warnings, improving their accuracy. Multi-channel warning information delivery ensures that warning information is received promptly by both medical staff and patients, while targeted recommendations provide clear action guidelines, helping medical staff quickly develop intervention plans and guiding patients to correctly manage risks, ultimately reducing the incidence of postoperative complications and improving the precision of postoperative monitoring.

[0052] In one embodiment, such as Figure 2 As shown, a method for intelligent monitoring and early warning after digestive endoscopy is provided, which is applied to... Figure 2 Taking China as an example, the following specific steps will be used: S10: Collect patient physiological data, surgical data, basic health data, and subjective symptom data.

[0053] S20: All collected data is encrypted and transmitted to the cloud, and securely stored using an architecture that combines blockchain and distributed database.

[0054] S30: The isolated forest algorithm is used to remove outliers from all collected data, the long short-term memory time series prediction algorithm is used to fill in missing values, the min-max standardization is used to normalize the data, and physiological indicator trend features, surgical trauma features, and text features are extracted and fused through an attention mechanism.

[0055] S40: Based on the fusion feature training, a three-level hybrid intelligent model, a convolutional neural network and a long short-term memory hybrid neural network model are trained to output the probability of complication risk and the rehabilitation status score.

[0056] S50: Trigger the corresponding early warning mechanism based on the four-level risk classification standard, and push early warning information and recommended measures.

[0057] S60: Generates dynamic rehabilitation plans, enabling plan delivery, symptom feedback, and online consultations through patient-side applications and medical staff management platforms.

[0058] Specifically, this method focuses on the full-cycle monitoring and rehabilitation guidance of patients after digestive endoscopy. Through six closely linked steps, it achieves closed-loop management of data collection, transmission and storage, processing and analysis, model prediction, early warning push and rehabilitation guidance, ensuring that the output of each step can provide accurate support for subsequent stages.

[0059] The first step is multi-source data collection, comprehensively gathering various postoperative data from the patient. Physiological data is acquired through wearable data collection devices, including heart rate, blood pressure, blood oxygen saturation, body temperature, activity level, and gastrointestinal motility frequency, covering the patient's core postoperative physiological state. Surgical data is collected from intraoperative monitors and pathological examination equipment, including operation duration, surgical method, intraoperative blood loss, extent of mucosal resection, and immediate postoperative pathological results, fully recording key information from the surgical process. Basic health data is extracted from the hospital information system, covering past medical history, allergy history, family medical history, and preoperative liver and kidney function indicators, clarifying the patient's health background. Subjective symptom data is self-reported by the patient through an application, including information on abdominal pain intensity, frequency of nausea and vomiting, and bowel characteristics. Patients can also upload wound photos to supplement visual data. The collection of all types of data is carried out simultaneously to ensure the completeness and timeliness of the information.

[0060] The second step is data transmission and storage, ensuring data security and accessibility. All collected data is encrypted and transmitted to the cloud via multi-mode transmission. The transmission process uses end-to-end advanced encryption standard 256-bit encryption to prevent data theft or tampering during transmission. Cloud storage adopts an architecture combining blockchain and a distributed database. Blockchain nodes store the hash values ​​and access logs of the data, ensuring data immutability and traceability of the access process. The distributed database stores all raw data, which is distributed across multiple physical nodes to avoid data loss due to the failure of a single node. A tiered access control system is also implemented, allowing different users to access data only within their authorized scope, ensuring patient privacy and security.

[0061] The third step involves data preprocessing and feature fusion to improve data quality and extract key information. First, an isolated forest algorithm is used to remove outliers. By constructing multiple decision trees, it identifies and removes abnormal data that significantly deviates from the majority of the data distribution, such as sudden increases in heart rate due to sensor malfunction or incorrectly recorded bleeding values. Next, a long short-term memory time series prediction algorithm is used to fill in missing values. For data gaps caused by temporary equipment malfunction or patients forgetting to report, predictions are made based on valid data from preceding and following time periods to ensure data continuity. Then, min-max standardization is used to normalize all data, mapping raw data of different dimensions to the same numerical range and eliminating the impact of dimensional differences on subsequent analysis. After data preprocessing, three core features are extracted: physiological indicator trend features are obtained by analyzing the fluctuations and slopes of physiological data at different time periods; surgical trauma features are quantified using the surgical trauma index formula; and textual features are extracted from medical history and symptom descriptions to extract key information. Finally, an attention mechanism is used to fuse the three types of features, automatically assigning higher weights to features highly correlated with complication risk and recovery status, forming a fused feature set.

[0062] The fourth step involves training and predicting intelligent models, outputting core evaluation metrics. Two types of models are trained based on fused features. One type is a three-level hybrid intelligent model, using a logistic regression model as the base model to make initial predictions for different complications; a stacked ensemble algorithm as the ensemble learning layer to integrate the prediction results of multiple base models, improving prediction accuracy; and a clinical rule base as the calibration model to correct the ensemble results based on clinical experience, ensuring the predictions meet actual diagnostic and treatment needs, ultimately outputting the probability of complication risk. The other type is a hybrid neural network model combining convolutional neural networks and long short-term memory (LSTM). The convolutional neural network extracts local correlation information of features, while the LTM captures the temporal dependencies of features, comprehensively assessing the patient's overall recovery and outputting a recovery status score. These two models work together to provide quantitative evidence for subsequent early warning.

[0063] The fifth step is tiered early warning, promptly sending risk alerts and response suggestions. Based on the probability of complication risk and the recovery status score, a four-tiered risk classification is applied. A complication risk probability less than 10 and a recovery status score greater than 80 is classified as Level 1 (low risk); a complication risk probability greater than or equal to 10 and less than 30 and a recovery status score greater than or equal to 60 and less than or equal to 80 is classified as Level 2 (medium risk); a complication risk probability greater than or equal to 30 and less than 50 and a recovery status score greater than or equal to 40 and less than 60 is classified as Level 3 (high risk); and a complication risk probability greater than or equal to 50 and a recovery status score less than 40 is classified as Level 4 (very high risk). Different risk levels correspond to different early warning mechanisms. Level 1 risk does not require proactive early warning; Level 2 risk sends a reminder message to the patient; Level 3 risk sends an early warning message to both the patient and healthcare provider; and Level 4 risk, in addition to sending a message to both the patient and healthcare provider, triggers an audible and visual alarm on the healthcare provider's end. The early warning message includes information such as the risk level, abnormal indicators, complication risk probability, and recovery status score, along with targeted suggestions.

[0064] The sixth step involves personalized rehabilitation guidance and doctor-patient interaction to ensure dynamic optimization of the rehabilitation plan. Based on early warning results, rehabilitation status scores, patient underlying diseases, and surgical types, a dynamic rehabilitation plan is generated using a combination of rule-based and case-based reasoning methods, specifying dietary recommendations, activity guidance, medication reminders, and other content. The plan is pushed to the patient via a client application, allowing the patient to view and follow the rehabilitation plan at any time. Patients report daily symptoms and rehabilitation progress through the application, and medical staff view patient data and feedback through a management platform, adjusting the rehabilitation plan based on the patient's progress. The adjusted plan is synchronized to the patient's device in real time. When patients have questions, they can consult online through the application, with medical staff responding promptly and providing professional guidance, forming real-time interaction between doctors and patients.

[0065] For example, 100 patients who underwent digestive endoscopy at a hospital were selected, and postoperative monitoring and rehabilitation guidance were conducted according to the above-described procedure. During the data collection phase, wearable devices collected heart rate, blood pressure, blood oxygen saturation, and body temperature data every five minutes, and gastrointestinal motility data every ten minutes. The medical device interface module simultaneously collected surgical data. The shortest surgical duration was 35 minutes, and the longest was 90 minutes. Intraoperative blood loss ranged from 5 ml to 50 ml. The hospital information system identified 30 patients with a history of hypertension and 15 patients with a history of diabetes. Patients reported symptoms daily through an application, and 25 patients reported mild abdominal pain on the first postoperative day. Data transmission was conducted via wireless LAN within the hospital and Bluetooth at home. All data was encrypted and stored in a cloud-based blockchain and distributed database. In the data preprocessing phase, 23 abnormal data points were removed, 18 missing values ​​were filled, and 26 fusion features were extracted after normalization. After model training, the three-level hybrid intelligent model predicted 8 patients at high risk of postoperative bleeding and 6 patients at high risk of infection. The rehabilitation status score output by the convolutional neural network and long short-term memory hybrid neural network model had a consistency of 93% with the clinical assessment. In the early warning stage, 12 patients were identified as level 3 risk and 5 as level 4 risk. The system promptly pushed early warning information and suggested measures to medical staff and patients. In the rehabilitation guidance stage, a personalized plan was generated for each patient. Among them, the initial plan for 30 patients undergoing endoscopic submucosal dissection was a liquid diet for five days. On the third day after surgery, the rehabilitation score of 15 patients improved to above 80, and the plan was automatically adjusted to a semi-liquid diet. Patients initiated a total of 42 online consultations through the application, and medical staff responded promptly and adjusted the plan 18 times.

[0066] This invention provides an intelligent monitoring and early warning method for postoperative gastrointestinal endoscopy, achieving refined and intelligent monitoring. Multi-source data acquisition ensures comprehensive information, covering multiple dimensions including physiological and surgical baseline health and subjective symptoms, providing ample data support for subsequent analysis. The combination of encrypted transmission and blockchain distributed storage guarantees data security and integrity, significantly reducing the risk of data leakage. Data preprocessing and feature fusion improve data quality and feature representativeness, laying a solid foundation for model prediction. The prediction accuracy of the intelligent model is significantly improved, far exceeding traditional experience-based judgment. A tiered early warning mechanism enables accurate risk identification and timely intervention; all patients at level four risk recovered after emergency treatment, and patients at level three risk did not develop serious complications after intervention. Dynamic rehabilitation plans and real-time doctor-patient interaction improve the relevance of rehabilitation guidance and patient compliance, shortening the average length of hospital stay and effectively solving problems such as data lag, inaccurate risk assessment, and fixed rehabilitation plans in traditional monitoring.

[0067] Specific limitations regarding the intelligent monitoring and early warning system after digestive endoscopy can be found in the above description of the methods for intelligent monitoring and early warning after digestive endoscopy, and will not be repeated here. Each module in the aforementioned intelligent monitoring and early warning system after digestive endoscopy can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0068] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0069] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart monitoring and early warning system for post-digestive endoscopy, characterized in that, include: Multi-source data acquisition layer, data transmission and secure storage layer, intelligent analysis and modeling layer, intelligent monitoring and early warning layer, and doctor-patient interaction and rehabilitation guidance layer; Each layer works together through standardized interface protocols; The multi-source data acquisition layer is used to collect patients' physiological data, surgical data, basic health data, and subjective symptom data. The data transmission and secure storage layer is used to receive collected data and encrypt it for secure transmission to the cloud for storage. The intelligent analysis and modeling layer is used to preprocess and extract features from the stored data, and then output the probability of complication risk and rehabilitation status score through a hybrid intelligent model. The intelligent monitoring and early warning layer is used to receive the above output results and provide graded early warnings. The doctor-patient interaction and rehabilitation guidance layer is used to generate personalized rehabilitation plans based on early warning results and scores, while also enabling doctor-patient interaction.

2. The intelligent monitoring and early warning system after digestive endoscopy according to claim 1, characterized in that, The multi-source data acquisition layer includes a wearable acquisition module, a medical device docking module, a hospital information system data access module, and a patient self-reporting module; The wearable data acquisition module collects heart rate, blood pressure, blood oxygen saturation, body temperature, activity level, and gastrointestinal motility. The medical device docking module collects data on operation duration, surgical method, intraoperative blood loss, intraoperative mucosal resection range, and immediate postoperative pathological results. The hospital information system's data access module extracts past medical history, allergy history, family medical history, and preoperative liver and kidney function indicators; The patient self-reporting module receives subjective symptom data and wound photos.

3. The intelligent monitoring and early warning system after digestive endoscopy according to claim 1, characterized in that, The data transmission and secure storage layer includes a real-time transmission unit, a secure storage unit, and a data synchronization unit; The real-time transmission unit uses Bluetooth, wireless LAN, and 5G mobile communication technologies, and adaptively switches the transmission mode according to the data type and application scenario. The transmission process uses end-to-end advanced encryption standard 256-bit encryption. The secure storage unit adopts a storage architecture that combines blockchain and distributed database. Blockchain nodes store data hash values ​​and access logs, while the distributed database stores the original data. The data synchronization unit uses an incremental synchronization algorithm and is equipped with a data caching mechanism.

4. The intelligent monitoring and early warning system after digestive endoscopy according to claim 1, characterized in that, The intelligent analysis and modeling layer includes a data preprocessing submodule, a feature extraction submodule, and a hybrid intelligent model construction submodule; The data preprocessing submodule uses the Isolation Forest algorithm to remove outliers, the Long Short-Term Memory time series prediction algorithm to fill missing values, and the min-max standardization to normalize the data. The feature extraction submodule extracts physiological indicator trend features, surgical trauma features, and text features, and fuses these features through an attention mechanism; The hybrid intelligent model construction submodule constructs a logistic regression model as the base model, a stacked ensemble algorithm as the ensemble learning layer, and a clinical rule base as the calibration model to form a three-level prediction model. At the same time, a hybrid neural network model of convolutional neural network and long short-term memory is constructed. The three-level prediction model outputs the probability of complication risk, and the hybrid neural network model of convolutional neural network and long short-term memory outputs the rehabilitation status score.

5. The intelligent monitoring and early warning system after digestive endoscopy according to claim 4, characterized in that, Surgical trauma characteristics are calculated using the surgical trauma index formula, which is: I trauma =α×V blood +β×S+γ×T op Among them, I trauma V represents the surgical trauma index, α, β, and γ are the clinical training weights, and V... blood S represents intraoperative blood loss, S represents the extent of mucosal resection during surgery, and T represents intraoperative blood loss. op This refers to the duration of the surgery.

6. The intelligent monitoring and early warning system after digestive endoscopy according to claim 1, characterized in that, The intelligent monitoring and early warning layer includes a real-time monitoring unit, a graded risk assessment unit, a dynamic threshold adjustment unit, and an early warning information push unit; The real-time monitoring unit uses a streaming computing framework to perform second-level data analysis; The risk assessment unit divides the risk level into four levels based on the probability of complication risk and the recovery status score; The dynamic threshold adjustment unit adjusts the warning threshold based on the patient's age, underlying diseases, and type of surgery. The early warning information push unit pushes early warning information and suggested measures to both medical staff and patients through pop-up windows on the medical staff management platform, SMS, application push, and voice reminders.

7. The intelligent monitoring and early warning system after digestive endoscopy according to claim 1, characterized in that, The doctor-patient interaction and rehabilitation guidance layer includes a personalized rehabilitation plan generation unit, a patient-side application, and a medical staff-side management platform; The personalized rehabilitation plan generation unit uses a combination of rule-based reasoning and case-based reasoning to generate dynamic rehabilitation plans; The patient-side application supports data viewing, treatment plan reception, symptom reporting, and online consultation. The healthcare management platform supports centralized patient management, data visualization, early warning processing, and treatment plan adjustments.

8. A method for intelligent monitoring and early warning after digestive endoscopy, characterized in that, include: Collect patients' physiological data, surgical data, basic health data, and subjective symptom data; All collected data is encrypted and transmitted to the cloud, where it is securely stored using an architecture that combines blockchain and a distributed database. The isolated forest algorithm was used to remove outliers from all collected data, the long short-term memory time series prediction algorithm was used to fill missing values, and the min-max standardization was used to normalize the data. Physiological indicator trend features, surgical trauma features, and text features were extracted and fused through an attention mechanism. Based on the fusion feature training, a three-level hybrid intelligent model, a convolutional neural network and a long short-term memory hybrid neural network model are trained to output the probability of complication risk and rehabilitation status score; Based on the four-level risk classification standard, the corresponding early warning mechanism is triggered, and early warning information and suggested measures are pushed out; Dynamic rehabilitation plans are generated, and the plans are pushed out, symptoms are reported, and online consultations are conducted through patient-side applications and medical staff management platforms.

Images