Intelligent rehabilitation method, device and medium for patients receiving left ventricular assist device

Abstract

The present invention discloses an intelligent rehabilitation method, device and medium for patients receiving a left ventricular assist device. The method proposed in the present invention is based on an agent scheduling system and smart wearable devices to obtain the vital sign data of patients with a left ventricular assist device in real time, and through artificial intelligence algorithms such as deep learning, medical staff are reminded to make regular return visits, so that patients can receive rehabilitation treatment in a timely manner, reduce the occurrence of abnormal ventricular assist pump circulation, and reduce the postoperative risks of patients.

Description

Technical Field

[0001] The present invention relates to the field of intelligent rehabilitation technology, and in particular to an intelligent rehabilitation method, device and medium for receiving a left ventricular assist device. Background Art

[0002] Among patients with heart failure, those who need to undergo ventricular assist device surgery are prone to psychological distress such as anxiety if they do not receive proper postoperative care and follow-up visits, which not only affects their quality of life, but also causes a series of complications, and even leads to death in severe cases. In view of the above problems, an intelligent rehabilitation method, device and medium for patients receiving left ventricular assist device are proposed. Summary of the invention

[0003] The purpose of the present invention is to solve the defects of the above-mentioned prior art and provide an intelligent rehabilitation method, device and medium for patients receiving left ventricular assist device.

[0004] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0005] An intelligent rehabilitation method for a patient receiving a left ventricular assist device, comprising:

[0006] An intelligent rehabilitation method for a patient receiving a left ventricular assist device, comprising:

[0007] Obtain patient data through the hospital server and create a patient information table;

[0008] Establishing a rehabilitation plan based on the patient information form, including but not limited to exercise rehabilitation, sleep time, dietary guidance, anxiety level assessment and classification, and medication management;

[0009] Based on wearable devices, the patient's life physiological data is obtained, and according to the event marking method and in combination with the rehabilitation plan, life risk fragments and risk hazards are marked;

[0010] Obtain a preset number of patients, obtain the motion detection start cycle, patient motion detection frequency, patient motion type, patient motion time, and life risk segments of each patient, and perform real-time detection on the patients in combination with high-precision professional detection equipment in the hospital to obtain high-precision patient physiological data. At the same time, perform real-time detection on the patients in combination with the wearable device to obtain low-precision patient physiological data, supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement, compare and count the differences between the high-precision patient physiological data and the low-precision patient physiological data, and obtain a list of differences in the patient physiological data life segments;

[0011] According to the labeled life risk fragment data, combined with the LSTM deep learning algorithm, a patient life risk prediction model is trained;

[0012] Obtain patient information tables of a preset number of patients, and obtain life risk segments and risk hazards of the preset patients, and construct a patient risk similarity information table; construct a patient hemodynamic risk hazard similarity model based on the patient risk similarity information table and in combination with a similarity algorithm;

[0013] According to the patient's life risk prediction model, predict the various life risk segments in the patient's life, combine the patient's hemodynamic risk hazard similarity model to predict risk hazards, and obtain the possible risk hazards of the patient. According to the patient's physiological data life segment difference list, determine whether the patient's life risk segment is a differential risk segment, and perform prompt text for the differential risk segment. According to the preset key risk segments corresponding to the patient's possible risk hazards, determine whether the patient's life risk segment lacks a certain key risk segment, and perform prompt text. Based on the large model prompt word template, combine the Agent to complete the patient's life situation, determine the patient's actual risk hazards, and remind medical staff to conduct manual follow-up visits after multiple rounds of intelligent follow-up visits based on the large model. Preferably, the wearable device is used to obtain the patient's life physiological data, and according to the event marking method, combined with the rehabilitation plan, the life risk segments and risk hazards are marked, including:

[0014] The wearable device includes monitoring technology for physiological parameters such as blood oxygen saturation, blood pressure, respiration, body position and movement. The parameters that can be detected include but are not limited to electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level.

[0015] The wearable devices include wrist-worn, smart vest, and head-worn;

[0016] The patient's life physiological data includes: based on the detection parameters obtained by the wearable device, recording the corresponding relationship between the detection parameters and time, that is, the correspondence between the detection parameters every minute and every second of every day, so as to obtain the patient's life physiological data;

[0017] The event marking method is combined with the rehabilitation plan to mark life risk segments and potential risks, including: requiring the patient, family members or doctors to view the patient's life physiological data in the app, record the life physiological data by themselves, and combine the rehabilitation plan content to determine whether it is a life risk segment, and mark the time period when the life risk segment occurs. The life risk segment includes sports type risk segments and physiological value type risk segments;

[0018] The risks include: not taking medication regularly, dietary problems, drinking, smoking, swimming, postoperative complications (pneumonia, thrombosis, joint contracture), wound infection, fear of exercise, breathing difficulties, lack of sleep, fatigue, other discomfort, forgetting to charge the left ventricular assist device, and abnormal ventricular assist pump circulation;

[0019] The sports type risk segment includes: the family members or doctors detect the running type data of the patient in the app based on the patient's life physiological data, including the number of steps, respiratory rate, and activity, to determine whether the patient's running status in this time period is normal, thereby marking the life risk segment; the physiological value type risk segment includes: the family members or doctors detect the physiological value type data of the patient in the app based on the patient's life physiological data, including electrocardiogram, electromyogram, heart rate, heart rate variability, blood pressure, blood oxygen saturation, and body surface temperature, to determine whether the patient's physiological value status in this time period is normal, thereby marking the life risk segment;

[0020] The risk hazards also include: presetting key risk segments corresponding to each risk hazard, and the key risk segments include: the time point of occurrence of the key risk segments, the key movement type risk segments, the key physiological value type risk segments, the frequency of occurrence of related key risk segments, and key patient information.

[0021] Preferably, the method comprises obtaining a preset number of patients, obtaining the motion detection start cycle, patient motion detection frequency, patient motion type, patient motion time, and life risk segments of each patient, and performing real-time detection on the patient in combination with high-precision professional detection equipment in the hospital to obtain high-precision patient physiological data. At the same time, the method comprises performing real-time detection on the patient in combination with the wearable device to obtain low-precision patient physiological data, supplementing the high-precision patient physiological data and the low-precision patient physiological data by data enhancement, and comparing and counting the differences between the high-precision patient physiological data and the low-precision patient physiological data to obtain a list of differences in the patient physiological data life segments, which specifically includes:

[0022] The high-precision professional testing equipment in the hospital includes: monitoring blood pressure by listening to Korotkoff sounds based on manual blood pressure cuff, or using Doppler ultrasound to monitor blood pressure;

[0023] The high-precision professional testing equipment in the hospital also includes the use of high-precision equipment to detect heart rate, electrocardiogram, electromyogram, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level;

[0024] The wearable device is used to detect the patient in real time to obtain low-precision physiological data of the patient, including using the wearable device to detect the patient's electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level;

[0025] The method of supplementing high-precision patient physiological data and low-precision patient physiological data by data enhancement includes: determining, according to the life risk segments and corresponding risk hazards annotated by the patient, the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, activity amount and the time period of the life risk segment detected by the wearable device of the patient in the life risk segment, and judging, according to the activity amount detected by the wearable device, that the activity amount belongs to the movement detection start cycle, patient movement detection frequency, patient movement type and patient movement time, so as to determine the time period, movement detection start cycle, patient movement detection frequency, patient movement type and patient movement time of the life risk segment that the patient needs to simulate in the data enhancement;

[0026] The patient maintains a life segment simulation of the movement detection start cycle, patient movement detection frequency, patient movement type, and patient movement time in the time period of the life risk segment within the preset days of admission. During the life segment simulation process, the high-precision professional detection equipment and the wearable device are combined to collect and detect high-precision patient physiological data and low-precision patient physiological data, so as to supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement;

[0027] Conduct similarity judgment on the captured risk segments to determine whether the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level are consistent with the previously marked segments;

[0028] The method of supplementing high-precision patient physiological data and low-precision patient physiological data by data enhancement also includes:

[0029] Determine the similarity between the low-precision patient physiological data obtained by data enhancement and the low-precision patient physiological data detected by the patient's annotated life risk segment, and if the similarity is lower than a threshold, repeat the method of supplementing the high-precision patient physiological data and the low-precision patient physiological data by data enhancement;

[0030] The comparing and counting the differences between the high-precision patient physiological data and the low-precision patient physiological data to obtain a list of differences in the patient's physiological data life segments includes:

[0031] Determine whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold value. If so, the life risk segments and risk hazards corresponding to the difference are included in the patient physiological data life segment difference list;

[0032] The determining whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity amount detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold also includes: determining whether the abnormalities of the high-precision patient physiological data and the low-precision patient physiological data are consistent, and including the life risk segments and potential risks corresponding to the differences into the patient physiological data life segment difference list.

[0033] Preferably, the life risk existence prediction model of the patient is trained in combination with the LSTM deep learning algorithm according to the labeled life risk segment data, specifically including: obtaining the life physiological data of the patient with a preset number of life risk segment data and the life physiological data of the patient with ordinary life risk-free segment data, the patient information table data corresponding to the patient is used as the input parameter of the LSTM algorithm, and whether the life segment corresponds to the life risk segment is used as the output parameter to train the life risk existence prediction model of the patient;

[0034] The ordinary life risk-free segment data corresponds to the preset number of life risk segment data, and the ordinary life risk-free segment data is the ordinary patient life physiological data that is not marked as risky. Preferably, the patient information table of the preset number of patients is obtained, and the life risk segments and risk hazards of the preset patients are obtained, and the patient risk similarity information table is constructed, specifically including:

[0035] Obtain patient information tables, corresponding life risk segments, and risk hazards of a preset number of patients, obtain life risk segments and corresponding risk hazards that occur in the life of each patient, and fill each field in the patient information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, the physiological value type or exercise type corresponding to the life risk segment, and the risk hazard into the patient risk similarity information table;

[0036] According to the patient risk similarity information table, a patient hemodynamic risk hazard similarity model is constructed in combination with a similarity algorithm, specifically including: according to each field in the patient information table in the patient risk similarity information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or movement type corresponding to the life risk segment, a patient hemodynamic risk hazard similarity model is constructed using the bm25 similarity algorithm, the above-mentioned characteristic values ​​of the patient who needs to predict the risk hazard are input to perform similarity calculation with other users, the user with the highest similarity is obtained, the risk hazard of the user is judged, and the possible risk hazard of the patient who needs to predict the risk hazard is obtained.

[0037] Preferably, the method predicts various life risk segments existing in the patient's life based on the patient's life risk existence prediction model, and predicts risk hazards in combination with the patient's hemodynamic risk hazard similarity model to obtain possible risk hazards of the patient, specifically including: obtaining the patient's life data in combination with the wearable device, predicting various life risk segments existing in the patient's life through the patient's life risk existence prediction model, obtaining the patient information table of the patient, predicting risk hazards in combination with the patient's hemodynamic risk hazard similarity model according to each field in the patient information table of the patient, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or movement type corresponding to the life risk segment, to obtain possible risk hazards of the patient, and generating risk hazards and key risk segments corresponding to risk hazards, as well as corresponding time tables;

[0038] The method of judging whether the patient's life risk segment is a differential risk segment based on the patient's physiological data life segment difference list and texting the differential risk segment prompt specifically includes: obtaining the life risk segment within a recent preset time, and calling the corresponding tool through the preset similarity number prompt template, and calculating the number of similarities with the patient's physiological data life segment difference list. When the number of occurrences exceeds a threshold, the life risk segment and the difference segment list are transmitted to the preset re-wearing prompt template, and a request for the patient to re-wear the wearable device is initiated; the patient's life risk segment is calculated to have a similarity with each life risk segment in the patient's physiological data life segment difference list. If the similarity value is greater than the threshold of the life risk segment, the patient's life risk segment is judged to be a differential risk segment, and the situation of abnormal values ​​in the differential risk segment is obtained, and the abnormal value situation is inserted into the differential risk segment prompt template;

[0039] The method of determining whether the patient's life risk fragments lack a certain key risk fragment according to the preset key risk fragments corresponding to the potential risk hazards of the patient, and texting the prompt words, and completing the patient's life situation based on the large model prompt word template in combination with the Agent to determine the actual risk hazards of the patient, specifically includes: according to the table data where the patient's risk hazards are stored, the table data stores the preset key risk fragments corresponding to the patient's risk hazards, and based on the preset missing data prompt template, calling the agent large model to find and analyze whether the key risk fragments of the patient's risk hazards are missing;

[0040] Calculate the similarity between the patient's life risk segment and each life risk segment in the patient's physiological data life segment difference list; if the similarity value is greater than the threshold life risk segment, determine that the patient's life risk segment is a differential risk segment; obtain the abnormal value of the differential risk segment, and put the abnormal value into the differential risk segment prompt template;

[0041] Based on the agent tool and the preset completion data prompt template, the agent big model is called to complete the missing life risk fragments of the patient, and the risk hidden danger similarity value is recalculated based on the completed life risk fragment to obtain the risk hidden danger similarity value, and according to the preset risk hidden danger accuracy threshold, the risk hidden danger similarity value is less than the preset risk hidden danger accuracy threshold, and the agent big model finds suitable repetition time intervals for different risk hidden dangers according to the preset time interval prompt template, until the risk hidden danger similarity value is greater than the preset risk hidden danger accuracy threshold, thereby outputting the actual risk hidden danger of the patient.

[0042] Preferably, after multiple rounds of intelligent follow-up visits based on the large model, the medical staff is reminded to conduct manual follow-up visits, which specifically include: determining the risk hazards of the patient, calculating the initial follow-up score according to the constructed follow-up formula, and if the set follow-up threshold is reached, the agent will follow up the patient in the first round, communicate with the patient through a virtual phone, obtain the patient's daily habits and intervention level based on the communication content with the patient, and generate a risk report summary. If the patient does not connect to communicate with the agent, the corresponding virtual phone follow-up weight is accumulated, and as the number of follow-up visits increases, the patient's repeated risk hazards will accumulate the corresponding follow-up weight, and the corresponding intervention level weight will be adjusted according to the degree of intervention for the patient. , it detects that the number of differential risk segments or missing key risk segments reaches the set differential missing threshold, reminds the patient to wear it again and improve the accuracy of the high-precision patient physiological data and the low-precision patient physiological data, obtains the patient's current anxiety level and fills it into the risk report, and the follow-up score reaches the preset manual follow-up threshold, entering the manual stage, and by collating historical risk reports, a long-term risk report summary is formed and sent to the doctor to intervene in the patient. During the historical multiple rounds of follow-up, the agent large model intercepts the risk data prompt template according to the preset template, obtains the corresponding time of the life risk data segment with fewer risk hazards and the life risk data segment with more risk hazards, and confirms the time interval for confirming the risk hazards.

[0043] On the other hand, an embodiment of the present invention further includes a computer device, including a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the at least one program to execute the method described in the embodiment.

[0044] On the other hand, an embodiment of the present invention further includes a storage medium storing a program executable by a processor, wherein the program executable by the processor is used to execute the method described in the embodiment when executed by the processor.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The method proposed in the present invention is based on an agent scheduling system and smart wearable devices to obtain the vital signs data of patients with left ventricular assist devices in real time. Through artificial intelligence algorithms such as deep learning, it dynamically adjusts the risk detection time interval, communicates with patients in smart follow-up visits in a timely manner, reduces the probability of patients experiencing risk hazards, and promptly reminds medical staff to conduct regular manual follow-up visits, provide patients with rehabilitation treatment, reduce the occurrence of ventricular assist pump circulation abnormalities, and reduce patients' postoperative risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of the first embodiment of the present invention; DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1:

[0050] s1: Obtain patient data from the hospital server and create a patient information table, including:

[0051] The patient data is obtained through the hospital server and stored in the patient information table in the form of structured data. The patient information table includes basic patient information, preoperative heart failure treatment, hemodynamics, examination indicators, postoperative related indicators, mechanical pump power data of the ventricular assist device, and degree of heart failure.

[0052] The patient's basic information includes gender, age, body surface area, height, weight, body mass index, body temperature, blood oxygen saturation, cause of heart failure, previous history of heart surgery, history of other diseases, type of ventricular assist, operation time, postoperative recovery, left ventricular ejection fraction, rehabilitation progress, physical condition, muscle strength assessment, activity tolerance, anxiety level, type of heart disease, heart function classification, and life stress.

[0053] The preoperative heart failure treatment includes preoperative cardiopulmonary resuscitation, intravenous medication, and the use of preoperative short-term circulatory assist devices.

[0054] The hemodynamic conditions include systemic blood pressure, heart rate, rhythm, central venous pressure, pulmonary artery pressure, pulmonary capillary wedge pressure, cardiac output, and cardiac index.

[0055] The examination indicators include blood sodium concentration, blood potassium concentration, blood glucose concentration, hemoglobin, white blood cell count, hematocrit, platelet count, international normalized ratio, partial prothrombin time, alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase, creatine kinase-MB, glutamine, total bilirubin, albumin concentration, total protein concentration, blood urea nitrogen concentration, blood creatinine concentration, and C-reactive protein concentration.

[0056] The postoperative related indicators include mechanical ventilation time, intensive care time, ventricular assist support time, reasons for ventricular assist device offline, conditions after ventricular assist pump implantation, and postoperative complications.

[0057] The mechanical pump power data of the ventricular assist device includes the model, working parameters, real-time pump power output, etc. of the device, so as to be analyzed in detail in subsequent steps.

[0058] The degree of heart failure includes: mild heart failure, moderate heart failure and severe heart failure.

[0059] The method for obtaining the heart failure degree value includes adopting a deep learning method based on a gradient boosting decision tree (GBDT), and classifying the heart failure degree of the patient according to the patient information table collected in step 1, including mild heart failure, moderate heart failure and severe heart failure, specifically including:

[0060] Before model training, the patient's basic information is first preprocessed, including data normalization, missing value processing, and selection of key features to ensure the quality and consistency of the input data.

[0061] The gradient boosting decision tree method improves the accuracy of prediction by constructing multiple decision trees and integrating their prediction results. Each tree contributes to reducing the residual error of the previous tree. This sequential enhancement method makes the model more efficient in dealing with complex nonlinear relationships.

[0062] Input parameters include key health indicators such as the patient's gender, age, body surface area, height, weight, body mass index, body temperature, blood oxygen saturation, as well as clinical information such as the cause of heart failure and previous heart surgery history.

[0063] The trained GBDT model can effectively classify the severity of heart failure in different patients, and its classification accuracy and robustness are ensured through cross-validation and independent dataset testing.

[0064] Ultimately, this classification model can serve as an important basis for training the pump power and hemodynamic risk relationship model in subsequent steps, improving the accuracy and reliability of the model in clinical applications.

[0065] In this step, the heart failure degree classification results will serve as an important reference for the training of the pump power and hemodynamic risk relationship model in the subsequent steps, ensuring that the model can more accurately reflect the risk characteristics of patients with different degrees of heart failure.

[0066] The patient data of the hospital server is acquired from various devices in the hospital in real time and is processed in a structured manner to obtain structured data.

[0067] S2. Establish a rehabilitation plan based on the patient information form, which includes but is not limited to exercise rehabilitation, sleep time, dietary guidance, anxiety level assessment and classification, and medication management, specifically including;

[0068] S21: Exercise rehabilitation settings in the rehabilitation plan, including but not limited to exercise detection start cycle, exercise detection frequency, patient exercise type and patient exercise time, etc., specifically including:

[0069] S211: Setting the starting period of the exercise detection in the exercise rehabilitation specifically includes:

[0070] According to the patient information table, combined with the preset first rule, the motion detection start period is set, and the preset first rule includes: according to the patient information table, relevant patient information is obtained, such as: gender, age, postoperative recovery, left ventricular ejection fraction, degree of heart failure, etc. According to the patient's gender, age, postoperative recovery, left ventricular ejection fraction and other factors, the patient is divided into different rehabilitation categories. For example, patients with good postoperative recovery (such as patients without complications and fast postoperative recovery) are classified as "early recovery category", and patients with left ventricular ejection fraction ≤ 30% or age ≥ 65 years old are classified as "delayed recovery category". For patients in the "early recovery category", the motion detection start period is set to 20 to 40 days after surgery; for patients in the "delayed recovery category", the motion detection start period is set to 4 months after surgery. The determination of the starting period is based on the analysis of past rehabilitation data of similar patients and the guidance of medical research.

[0071] According to the first rule above, the database presets the motion detection start period that should be maintained for each type of patient;

[0072] S212: The patient movement detection frequency setting in the movement rehabilitation specifically includes:

[0073] According to the patient information table, combined with the preset second rule, the patient's exercise detection frequency is set. The preset second rule includes: according to the patient information table, obtaining relevant patient information, such as: rehabilitation progress, physical condition, degree of heart failure, etc. According to the patient's rehabilitation progress, physical condition and other factors, the patient is divided into different exercise frequency categories. For example, patients with faster rehabilitation progress and good physical condition are classified as "high-frequency exercise category", and patients with slow rehabilitation progress or poor physical condition are classified as "low-frequency exercise category". For patients in the "high-frequency exercise category", it is recommended that their exercise detection frequency is 5 to 7 times a week; for patients in the "low-frequency exercise category", it is recommended that their exercise detection frequency is 2 to 3 times a week. The determination of the exercise detection frequency is based on the analysis of rehabilitation data of similar patients in the past and the guidance of medical research. According to the above-mentioned second rule, the exercise detection frequency that should be maintained for each type of patient is preset in the database;

[0074] S213: The patient exercise type setting in the exercise rehabilitation specifically includes:

[0075] According to the patient information table, combined with the preset third rule, the patient's exercise type is set. The preset third rule is, for example:

[0076] Obtain patient information: Obtain the patient's specific information from the patient information form, including postoperative recovery status, muscle strength assessment, activity tolerance, etc.

[0077] Patients were classified according to the information: Patients with stronger muscle strength and higher activity tolerance were classified into the “high-intensity exercise category”.

[0078] Patients with average muscle strength and moderate activity tolerance are classified as "moderate-intensity exercise category". Patients with weak muscle strength and low activity tolerance are classified as "low-intensity exercise category".

[0079] Set the exercise type in stages: Stage 1: All patients start by sitting quietly by the bed and gradually transition to getting out of bed. Stage 2: According to the patient's progress, gradually increase the range and intensity of activities, such as transitioning from sitting up in bed to standing by the bed, stepping on the spot with the help of manual or equipment assistance, and finally transitioning to independent walking and more complex activities. Stage 3: Carry out targeted training according to patient classification. For example, patients in the "high-intensity exercise category" can perform aerobic exercise, strength training, etc., including cycling, running on a treadmill, walking, climbing stairs, etc.; patients in the "moderate-intensity exercise category" can perform light stretching, balance training, etc.; patients in the "low-intensity exercise category" continue to exercise in bed, sit by the bed, etc.

[0080] The setting of the exercise type should be personalized according to the specific situation of the patient and based on the analysis of rehabilitation data of similar patients in the past and the guidance of medical research. According to the third rule mentioned above, the database presets the exercise type to be performed for each type of patient and the phased plan of transition training to ensure the safety and effectiveness of the exercise plan.

[0081] S214: The patient exercise time setting in the exercise rehabilitation specifically includes:

[0082] According to the patient information table, combined with the preset fourth rule, the patient's exercise time is set. The preset fourth rule is, for example:

[0083] Obtaining patient information: Obtaining the patient's specific information from the patient information form, including rehabilitation progress, degree of heart failure, postoperative recovery, activity tolerance, etc.

[0084] Patients were classified according to the information: For patients with strong muscle strength and high activity tolerance, they were classified as "high-intensity continuous training" and the exercise duration was greater than 30 minutes.

[0085] For patients with average muscle strength and moderate activity tolerance, it is classified as "moderate-intensity continuous training" with an exercise duration of 5 to 30 minutes. For patients with weak muscle strength and low activity tolerance, it is classified as "low-intensity continuous training" with an exercise duration of about 5 minutes.

[0086] In the database, according to the fourth rule above, the patient exercise time that should be maintained for each type of patient is preset;

[0087] S22: Sleep time setting in the rehabilitation plan, the sleep time includes but is not limited to sleep duration and sleep frequency, etc., specifically including:

[0088] S221: Setting the sleep duration in the sleep time, specifically including:

[0089] According to the patient information form, combined with the preset first sleep rule, set the sleep duration range. The preset first sleep rule includes: according to the patient information form, obtain relevant patient information, such as: heart disease type, heart function grade, rehabilitation progress, age, gender, etc. For patients with mild heart disease, good heart function and in the early stage of rehabilitation, set their sleep duration to 7 to 8 hours per night; for patients with severe heart disease, high heart function grade or in the middle and late stages of rehabilitation, set their sleep duration to 8 to 9 hours per night. For elderly patients, considering their physiological characteristics, the sleep duration can be appropriately reduced. The determination of the sleep duration range is based on the analysis of past similar patients' sleep data and the guidance of medical research.

[0090] In the database, the recommended sleep duration range is preset for each type of patient according to the first sleep rule mentioned above;

[0091] S222: Setting the sleep frequency during the sleep time, specifically including:

[0092] According to the patient information table, combined with the preset second sleep rule, the patient's sleep frequency is set. The preset second sleep rule includes: according to the patient information table, obtain relevant patient information, such as rehabilitation progress, physical condition, age, etc., set the patient who is over 60 years old, the physical condition is poor, and the rehabilitation progress is in the early stage, set their sleep frequency to go to bed at 10 o'clock every night, get up at 7 o'clock in the morning, and maintain 9 hours of stable sleep every day. As the rehabilitation progress develops to the later stage, the physical condition becomes good, and the sleep frequency is gradually changed to go to bed at 11 o'clock in the evening, get up at 5 o'clock in the morning, and maintain 6 hours of stable sleep every day. Set the patient who is less than 60 years old, the physical condition is poor, and the rehabilitation progress is in the early stage, set their sleep frequency to go to bed at 10 o'clock every night, get up at 6 o'clock in the morning, and maintain 8 hours of stable sleep every day. As the rehabilitation progress develops to the later stage, the physical condition becomes good, and the sleep frequency is gradually changed to go to bed at 11 o'clock in the evening, get up at 7 o'clock in the morning, and maintain 8 hours of stable sleep every day.

[0093] The database presets the sleep frequency that the patient should maintain according to the second sleep rule mentioned above;

[0094] S23. The dietary guidance settings in the rehabilitation plan specifically include:

[0095] The patient conducts regular dietary monitoring, records the types, quantities, and cooking methods of daily food intake, etc., uploads the recorded information to the hospital database, and uses the data analysis tools in the hospital database to analyze the patient's dietary intake levels of trace elements and macronutrients to obtain dietary analysis results. Based on the recorded dietary analysis results, combined with the patient's physical condition and recovery progress, the doctor promptly adjusts the diet plan and dynamically updates it in the database. For example, for patients with high blood lipid levels, the patient's dietary intake levels of trace elements and macronutrients are analyzed through the data analysis tools in the hospital database, so as to adjust the patient's diet plan to reduce the intake of foods that increase blood lipid levels;

[0096] Based on the above dietary guidelines, the database presets a diet log template that patients should record regularly, and provides corresponding dietary adjustment suggestions.

[0097] S24: Assessment and classification of anxiety levels in the rehabilitation program, specifically including:

[0098] According to the patient information form, combined with the preset anxiety assessment rules, the anxiety level of the patient is assessed and classified. The preset anxiety assessment rules include, but are not limited to: according to the patient information form, relevant patient information is obtained, such as age, gender, past medical history, rehabilitation progress, life pressure, etc. According to the patient's age, gender, rehabilitation progress, life pressure and other factors, the patient is divided into different anxiety level categories. For example, patients who are younger, have a good rehabilitation progress, and have less life pressure are classified as "mild anxiety category", patients who are older, have a slower rehabilitation progress, and have greater life pressure are classified as "moderate anxiety category", and patients with severe psychological distress and obstructed rehabilitation process are classified as "severe anxiety category".

[0099] Methods for assessing the level of anxiety include: using professional psychological assessment tools, such as an anxiety self-assessment scale, to allow the patient to conduct a self-assessment based on his or her own situation; or conducting a face-to-face psychological interview, in which a professional psychologist or medical staff has in-depth communication with the patient to assess his or her anxiety state; or observing the patient's daily behavior and verbal expression to determine whether there are emotional manifestations such as tension, anxiety, and irritability; at the same time, paying attention to monitoring the patient's physiological indicators, such as heart rate, blood pressure, etc., to assist in assessing the level of anxiety.

[0100] In the database, the possible anxiety level range and corresponding guidance strategies of each type of patient are preset according to the above anxiety assessment rules;

[0101] S25. Medication management settings in the rehabilitation plan include but are not limited to drug types and dosages, and time of administration, etc., specifically including:

[0102] S251: The drug type and dosage settings in the medication management settings specifically include:

[0103] Based on the patient's diagnosis, severity of illness, comorbidities and other factors, combined with the drug use guide and the doctor's professional judgment, the appropriate drug type and dosage are set for the patient. Based on the above factors, the database presets the recommended dosage range for each drug in specific circumstances to provide a reference for doctors.

[0104] S252: The medication time management setting in the medication management setting specifically includes:

[0105] Combined with the patient's work and rest schedule, eating habits and the pharmacological properties of the drug, a reasonable drug taking time is set. Based on the above factors, the database presets the recommended taking time for each drug in a specific situation to provide medication guidance for doctors and patients.

[0106] S3. Based on wearable devices, obtain the patient's life physiological data, and mark the life risk segments and risk hazards according to the event marking method and the rehabilitation plan, including:

[0107] The wearable device includes monitoring technology for physiological parameters such as blood oxygen saturation, blood pressure, respiration, body position and movement. The parameters that can be detected include but are not limited to electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, activity level, etc.

[0108] The wearable devices include wrist-worn, smart vest, head-worn, etc.;

[0109] The patient's life physiological data includes: based on the detection parameters obtained by the wearable device, recording the corresponding relationship between the detection parameters and time, that is, the correspondence between the detection parameters and the detection parameters every minute and every second of every day, so as to obtain the patient's life physiological data, for example: the patient wears the wearable device for life, and the wearable device detects the patient's electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity detection parameter data, and records the detection parameter data every minute and every second of every day, so as to obtain the patient's life physiological data;

[0110] The event marking method is combined with the rehabilitation plan to mark life risk segments and potential risks, including: requiring the patient, family members or doctors to view the patient's life physiological data in the app, record the life physiological data by themselves, and combine the rehabilitation plan content to determine whether it is a life risk segment, and mark the time period when the life risk segment occurs. The life risk segment includes sports type risk segments and physiological value type risk segments;

[0111] The risks include: not taking medication regularly, dietary problems, drinking, smoking, swimming, postoperative complications (pneumonia, thrombosis, joint contracture), wound infection, fear of exercise, breathing difficulties, lack of sleep, fatigue, other discomforts, forgetting to charge the left ventricular assist device, abnormal ventricular assist pump circulation, etc.

[0112] By marking life risk segments and potential risks through event marking methods, we can obtain the corresponding life risk segments of patients with potential risks, and then clarify which potential risks the patient has for auxiliary judgment in the future;

[0113] The exercise type risk segments include: family members or doctors use the app to detect the running type data of the patient's life physiological data, including the number of steps, respiratory rate, activity level, etc., to determine whether the patient's running status in the time period is normal, thereby marking the life risk segments. For example, whether the activity level during normal activity time is too low can be used to determine whether there is a life risk segment with an exercise type risk segment;

[0114] The physiological value type risk segment includes: the family members or doctors use the app to detect the physiological value type data of the patient's life physiological data, including electrocardiogram, electromyogram, heart rate, heart rate variability, blood pressure, blood oxygen saturation, body surface temperature, etc., to determine whether the patient's physiological value condition in the time period is normal, thereby marking the life risk segment. For example, whether the heart rate and blood pressure are abnormal can be used to determine whether there is a physiological value type life risk segment;

[0115] The risk hazards also include: presetting key risk segments corresponding to each risk hazard, and the key risk segments include: the time point of occurrence of the key risk segments, the key sports type risk segments, the key physiological value type risk segments, the frequency of occurrence of related key risk segments, and key patient information; for example, the key physiological value type risk segment of "not taking medication regularly" is a life risk segment related to abnormal blood pressure and heart rate after not taking medication regularly, and the time point of occurrence of the key risk segment is the preset time after the medication time, for example:

[0116] Table 1

[0117]

[0118] Table 1 shows the risk of "no regular medication". The key physiological value types corresponding to "no regular medication" include but are not limited to heart rate and blood pressure, as well as the abnormal physiological values ​​and start and end times corresponding to the key physiological value types; the key exercise type risk fragments of the risk of "lack of sleep" are life risk fragments related to excessive sleep and exercise. The frequency of occurrence of the relevant key risk fragments is the occurrence of the relevant life risk fragments for more than 3 consecutive nights. The key patient information is that the patient's anxiety level is severe, for example:

[0119] Table 2

[0120]

[0121] Table 2 shows the risk of "lack of sleep", the key physiological value types corresponding to the "lack of sleep", including but not limited to the number of steps, and the abnormal physiological values, start and end time and anxiety level corresponding to the key physiological value types; each preset risk hazard corresponds to a key risk segment, which can be used to clarify which risk hazard the patient has in the future for auxiliary judgment;

[0122] S4: Obtain a preset number of patients, and according to the method of steps S2-S3, obtain the motion detection start cycle, patient motion detection frequency, patient motion type, patient motion time, and life risk segments of each patient, and perform real-time detection on the patient in combination with high-precision professional detection equipment in the hospital to obtain high-precision patient physiological data. At the same time, perform real-time detection on the patient in combination with the wearable device to obtain low-precision patient physiological data, supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement, compare and count the differences between the high-precision patient physiological data and the low-precision patient physiological data, and obtain a list of differences in patient physiological data life segments, specifically including:

[0123] The preset number of patients may be 100 patients;

[0124] The high-precision professional testing equipment in the hospital may include: listening to Korotkoff sounds based on manual blood pressure cuff monitoring, or using Doppler ultrasound to monitor blood pressure;

[0125] The high-precision professional testing equipment in the hospital also includes the use of high-precision equipment to test heart rate, electrocardiogram, electromyogram, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, activity level, etc.;

[0126] The wearable device is used to detect the patient in real time to obtain low-precision physiological data of the patient, including using the wearable device to detect the patient's electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level;

[0127] The difference between the high-precision patient physiological data and the low-precision patient physiological data is that the high-precision patient physiological data

[0128] The detection and collection is based on high-precision professional detection equipment in the hospital, and there are professional operation and supervision by medical staff, with high accuracy and stable performance. The low-precision patient physiological data is detected and collected based on wearable devices, which may have low detection accuracy or certain errors due to different detection accuracies of different wearable devices and different body movements of patients;

[0129] The method of supplementing high-precision patient physiological data and low-precision patient physiological data by data enhancement includes: determining, according to the life risk segments and corresponding risk hazards marked by the patient, the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, activity amount and the time period of the life risk segment detected by the wearable device of the patient in the life risk segment, and judging, according to the activity amount detected by the wearable device, that the activity amount belongs to the motion detection start cycle, patient motion detection frequency, patient motion type and patient motion time in step s2, so as to determine the time period, motion detection start cycle, patient motion detection frequency, patient motion type and patient motion time of the life risk segment that the patient needs to simulate in this data enhancement;

[0130] The patient maintains a life segment simulation of the movement detection start cycle, patient movement detection frequency, patient movement type, and patient movement time in the time period of the life risk segment within the preset days of admission. In the life segment simulation process, the high-precision professional detection equipment and the wearable device are combined to collect and detect high-precision patient physiological data and low-precision patient physiological data, so as to supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement;

[0131] Perform similarity judgment on the captured risk segments to determine whether the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level are consistent with the previously marked segments.

[0132] The method of supplementing high-precision patient physiological data and low-precision patient physiological data by data enhancement also includes:

[0133] Determine the similarity between the low-precision patient physiological data obtained by data enhancement and the low-precision patient physiological data detected by the patient's annotated life risk segment. If the similarity is lower than a threshold, repeat the method of supplementing the high-precision patient physiological data and the low-precision patient physiological data by data enhancement to achieve a more sufficient and accurate amount of data;

[0134] The comparing and counting the differences between the high-precision patient physiological data and the low-precision patient physiological data to obtain a list of differences in the patient's physiological data life segments includes:

[0135] Determine whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold value. If exceeded, the life risk segments and risk hazards corresponding to the difference are included in the patient physiological data life segment difference list. For example, the blood pressure detected by the high-precision patient physiological data is 80 mmHg, and the blood pressure detected by the low-precision patient physiological data is 60 mmHg, which exceeds the threshold value of 5 mmHg. Retrospectively detect the life risk segments and risk hazards detected by the low-precision patient physiological data, and add them to the patient physiological data life segment difference list;

[0136] The thresholds and detection methods of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level are different. For example, in blood pressure detection, continuous pumping of blood by LVAD usually causes the patient's blood pressure to drop. Therefore, MAP is often monitored clinically, by listening to Korotkoff sounds through a manual blood pressure cuff, or using Doppler ultrasound monitoring. Due to the limited maximum output provided by the LVAD, the patient's own poor heart function, insufficient self-regulation of peripheral vascular resistance, excessive pumping speed or arrhythmia leading to insufficient left ventricular filling, patients are prone to hypotension during exercise; and hypertension may affect the LVAD's ability to pump blood. During this period, the map should be closely monitored, and a Doppler blood pressure monitor should be used to improve monitoring accuracy. It is usually recommended to maintain MAP at 70-90 mmHg. If the high-precision patient physiological data and the low-precision patient physiological data differ too much in this indicator, subsequent warnings or follow-up visits may be incorrect due to inaccuracies;

[0137] The determining whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and amount of activity detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold value also includes: determining whether the abnormalities of the high-precision patient physiological data and the low-precision patient physiological data are consistent, for example, the electrocardiogram detection data, the high-precision patient physiological data indicates an abnormal heart rate, and the low-precision patient physiological data indicates that the heart rate is not abnormal, and the two are inconsistent; the life risk segments and risk hazards corresponding to the differences are counted into the patient physiological data life segment difference list;

[0138] The patient physiological data life segment difference list also includes: high-precision patient physiological data and low-precision patient physiological data corresponding to each life risk segment, for example:

[0139] Table 3

[0140]

[0141] The high-precision physiological data in Table 3 are detected by the preset number of patients through the high-precision professional detection equipment of the hospital, and the low-precision physiological data are detected by the preset number of patients through the wearable device. When the high-precision physiological data and the low-precision physiological data reach the set difference threshold, they are recorded in the patient's physiological data life segment difference list.

[0142] S5: Based on the life risk segment data marked in step S3, combined with the LSTM deep learning algorithm, a patient life risk existence prediction model is trained, specifically including:

[0143] According to the method of step S3, the life physiological data of patients with a preset number of life risk segments and the life physiological data of patients with ordinary life risk-free segments are obtained, the patient information table data corresponding to the patient is used as the input parameter of the LSTM algorithm, and whether the life segment corresponds to the life risk segment is used as the output parameter to train the patient life risk existence prediction model;

[0144] The ordinary life risk-free segment data is in a corresponding relationship with the preset number of life risk segment data, and the ordinary life risk-free segment data is the ordinary patient life physiological data that is not marked as risky;

[0145] Long short-term memory (LSTM) is a deep learning model used to process and predict problems in time series data, natural language processing, etc. Compared with traditional recurrent neural networks (RNNs), LSTM can better solve the problem of gradient disappearance or explosion, and can effectively capture long-distance dependencies.

[0146] The core of LSTM is the memory cell, which can store information and control the flow of information. LSTM contains three key gating units, namely the forget gate, input gate, and output gate, as well as a cell state for controlling the update of the memory cell state.

[0147] For example: obtain 10,000 pieces of life risk segment data annotated by S3, and 10,000 pieces of ordinary life risk-free segment data annotated by S3, and use the time length of each 5 minutes as a window of data segments, that is, each data segment includes 5 minutes of patient life physiological data, and use the 20,000 patient life physiological data and the patient information table data corresponding to the patient as the LSTM algorithm input parameters, and use whether it is a life risk segment as the output parameter to train the patient life risk existence prediction model;

[0148] In subsequent use, the wearable device can continuously detect the patient's life physiological data, and use 5 minutes as a window to predict the time window to determine whether the patient's life physiological data is a life risk segment;

[0149] Furthermore, in another embodiment of the present invention, the output parameters of the patient life risk existence prediction model may specifically be a key sports type risk segment or a key physiological value type risk segment;

[0150] S6: Obtain patient information tables of a preset number of patients according to the method of step s1, and obtain life risk segments and risk hazards of preset patients according to the method of step s3, and construct a patient risk similarity information table, specifically including:

[0151] S61: Obtain patient information tables, corresponding life risk segments, and risk hazards of a preset number of patients, for example, obtain patient information tables of 10,000 patients, and according to the method of step s3, obtain life risk segments and corresponding risk hazards that occur in the life of each patient, and fill each field in the patient information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, the physiological value type or exercise type corresponding to the life risk segment, and the risk hazard into the patient risk similarity information table;

[0152] S62: Based on the patient risk similarity information table, a patient hemodynamic risk hazard similarity model is constructed in combination with a similarity algorithm, specifically including: based on each field in the patient information table in the patient risk similarity information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or movement type corresponding to the life risk segment, a patient hemodynamic risk hazard similarity model is constructed using the bm25 similarity algorithm, the above-mentioned characteristic values ​​of the patient who needs to predict the risk hazard are input and the similarity is calculated with other users to obtain the user with the highest similarity, and the risk hazard of the user is judged, thereby obtaining the possible risk hazard of the patient who needs to predict the risk hazard.

[0153] BM25 (Best Match 25) is an algorithm for calculating text similarity. It is a statistics-based algorithm used to evaluate the relevance between queries and documents.

[0154] The BM25 algorithm calculates similarity based on the degree of keyword matching between the query and the document. The following are the core principles and steps of the BM25 algorithm:

[0155] 1. First, segment the query and document to obtain the query keywords and document keywords.

[0156] 2. Calculate the frequency of each query term in the document (term frequency) and the frequency of each query term in the entire corpus (corpus term frequency)

[0157] 3. Use the BM25 formula to calculate the score of each query word: score(qi, D) = idf(qi)*(tf(qi, D)*(k+1)) / (tf(qi, D)

[0158] score(qi,D)=idf(qi)*(tf(qi,D)*(k+1)) / (tf(qi,D)+k*(1-b+b*,D, / avgdl))

[0159] Where qi is the query term, D is the document, tf(gi, D) is the frequency of occurrence of the query term qi in document D, idf(gi) is the inverse document frequency of the query term qi, k and b are adjustable parameters, and avgdl is the average document length.

[0160] Similarity algorithms may also include but are not limited to: Euclidean distance, Pearson correlation coefficient, cosine similarity, etc.;

[0161] S7: After the patient is discharged from the hospital, the patient's life data is obtained, and through the methods of s1-s6, according to the patient's life risk prediction model, the various life risk segments in the patient's life are predicted, and the risk hazard similarity model of the patient's hemodynamic risk hazard is combined to predict the risk hazard, and the possible risk hazard of the patient is obtained. According to the difference list of the patient's physiological data life segments, it is judged whether the patient's life risk segment is a differential risk segment, and the prompt of the differential risk segment is textualized. According to the preset key risk segments corresponding to the possible risk hazards of the patient, it is judged whether the patient's life risk segment lacks a certain key risk segment, and the prompt is textualized. Based on the large model prompt word template, combined with the Agent, the patient's life situation is completed to determine the patient's actual risk hazard, and after multiple rounds of intelligent follow-up visits based on the large model, medical staff are reminded to conduct manual follow-up visits.

[0162] S7.1: After the patient is discharged from the hospital, the patient's life data is obtained, and the various life risk segments existing in the patient's life are predicted by the patient's life risk prediction model, and the risk hazards are predicted in combination with the patient's hemodynamic risk hazard similarity model to obtain the possible risk hazards of the patient, specifically including: according to the method of step s3, in combination with the wearable device, the patient's life data is obtained, and the various life risk segments existing in the life are predicted by the patient's life risk prediction model, and the patient information table of the patient is obtained according to the method of step s1, and according to each field in the patient information table of the patient, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or movement type corresponding to the life risk segment, the risk hazards are predicted in combination with the patient's hemodynamic risk hazard similarity model to obtain the possible risk hazards of the patient, and risk hazards and key risk segments corresponding to risk hazards, as well as corresponding time tables, are generated, such as Table 1 and Table 2;

[0163] S7.2: Determine whether the patient's life risk segment is a differential risk segment based on the patient's physiological data life segment difference list, and convert the differential risk segment into a prompt text. Determine whether the patient's life risk segment lacks a certain key risk segment based on the preset risk potential hazards corresponding to the patient's key risk segments, and convert the prompt text. Based on the large model prompt word template, the Agent is combined to complete the patient's life situation, specifically including:

[0164] S7.21: According to S3, life risk segments within a recent preset time are obtained. Through the preset similarity number prompt template, the agent model calls the corresponding tool to calculate the number of similarities with the patient's physiological data life segment difference list. When the number of occurrences exceeds the threshold, the life risk segment and the difference segment list are sent to the preset re-wearing prompt template, and a request for the patient to re-wear the wearable device is initiated, which specifically includes:

[0165] According to S3, life risk segments within a recent preset time are obtained. Through the preset similarity number prompt template, the agent model calls the corresponding tool to calculate the number of similarities between the life risk segments in the difference segment list. When the number of occurrences exceeds the threshold, the life risk segments and the difference segment list are sent to the preset re-wearing prompt template, and a request for the patient to re-wear the wearable device is initiated. Specifically, after the patient is discharged from the hospital, the life risk segments of the patient within a recent preset time are obtained through S3. Through the preset similarity number prompt template,

[0166] The preset similarity number prompt template can be "

[0167] Input data: a list of differences in the patient's physiological data in the recent month, as shown in Table 3;

[0168] The various tools preset by Agent include: Euclidean similarity algorithm Agent tool, etc.

[0169] You are an intelligent analysis robot. You need to rely on the above information to determine whether to execute the tool preset by the Agent to calculate the number of similarities with the life risk segment in the difference segment list. The agent model calls the corresponding tool according to the preset similarity number prompt template to calculate the number of similarities with the life risk segment in the difference list of the patient's physiological data life segment. For example, the corresponding tool is called by the Euclidean similarity algorithm in the similarity number method. When the similarity exceeds the threshold, it is determined that the similarity number increases once. When the number exceeds the threshold, the life risk segment and the difference segment list are recorded and sent to the preset re-wearing prompt template. The re-wearing prompt template, for example:

[0170] “The number of similarities between the life risk segments in the recent preset time and the life risk segments in the difference segments list: For example:

[0171] Table 4

[0172]

[0173] Table IV shows the number of similarities between the actual low-precision physiological data in the life risk segments within the preset time and the low-precision physiological data in the difference segment list;

[0174] The various tools preset by the Agent include: data analysis Agent tool, the wearer's device Agent tool, etc.;

[0175] You are an intelligent interactive robot. You need to rely on the above information to determine whether to execute the tool preset by the Agent. According to the similar number of multiple physiological data questions, construct a request question for the patient to wear the device again, so as to call the wearer's wearable device Agent tool and ask the patient through voice or text message. The inquiry request question can be "When you use the wearable device to monitor blood pressure, do you find significant differences with the high-precision physiological data of the hospital? After our detection, your wearable device may have signal reception problems recently. Please wear the wearable device again." After the patient wears it again, the wearable device can normally obtain low-precision physiological data, thereby enhancing the detection accuracy of the patient and facilitating the next step of accurately predicting the patient's risk hazards;

[0176] The Agent is an intelligent tool in LangChain, acting as an intermediate layer to support and enhance the user's interactive experience with the system by calling different processing tools;

[0177] The Agent model is a language intelligence model used to process and generate natural language text. The Agent model works by pre-training on a large amount of text data, learning the patterns and structures of the language, and then using this knowledge to perform various language tasks. The Agent model includes but is not limited to GPT, Kimi, and Wenxinyiyan, etc.

[0178] The Euclidean similarity algorithm Agent tool calculates the number of similarities between the actual low-precision physiological data in the life risk segment within a preset time and the low-precision physiological data in the difference segment list;

[0179] The data analysis agent tool is a mathematical tool for calculating the number of occurrences of different physiological value types;

[0180] The wearer's worn device Agent tool has an interface api that can be used by the Agent to call its interactive interface tool.

[0181] S7.22: judging whether the patient's life risk segment is a differential risk segment according to the patient's physiological data life segment difference list, and converting the differential risk segment into a prompt text; judging whether the patient's life risk segment lacks a certain key risk segment according to the preset risk potential hazards corresponding to the patient's key risk segments, and converting the prompt text; based on the large model prompt word template, combining the Agent to complete the patient's life situation;

[0182] The determining whether the patient's life risk segment is a differential risk segment based on the patient's physiological data life segment difference list, and texting the prompt words of the differential risk segment, specifically includes: calculating the similarity between the patient's life risk segment and each life risk segment in the patient's physiological data life segment difference list, if the similarity value is greater than a threshold life risk segment, determining that the patient's life risk segment is a differential risk segment, obtaining the situation where the differential risk segment has an abnormal value, and fitting the abnormal value situation into the differential risk segment prompt template;

[0183] The differential risk segment prompt template may be: "The current differential risk segment, as shown in Table 4, has an abnormal value of xx, and it is suspected that the abnormal value is inaccurate". After the prompt word of the differential risk segment is textualized, the differential risk segment prompt text is obtained, for example: "The current differential risk segment has an abnormal value of blood pressure greater than 90, and it is suspected that the abnormal value is inaccurate", where the differential value xx may also be blood pressure, heart rate, exercise volume, etc.;

[0184] By template-izing the above-mentioned differential risk fragment prompt, a reliable prompt sub-template can be provided for subsequent agent questions, thereby improving the accuracy of agent questions;

[0185] The method of determining whether the patient's life risk segment lacks a certain key risk segment according to the preset risk hidden dangers corresponding to the key risk segment of the patient, and converting the prompt words into text, specifically includes:

[0186] Through S7.1, the table data storing the patient's risk hidden dangers is obtained. The table data stores the preset key risk segments corresponding to the patient's risk hidden dangers. Based on the preset missing data prompt template, the agent big model is called to search and analyze whether the key risk segments of the patient's risk hidden dangers are missing;

[0187] Preset missing data prompt template, which can be "

[0188] Risk hazards and corresponding key risk segments: Table x (e.g. Table 1 in the embodiment)

[0189] You are an intelligent analysis robot. You need to rely on the above information to search and analyze the risk potential table data to determine whether the risk segment data is missing. Therefore, the agent model is called to determine whether the risk potential table data is missing. The agent model finds the missing data and generates a missing risk data prompt template.

[0190] The missing risk data prompt template may be "The heart rate from 19:00:00 to 22:00:00 on 2024-04-23 is missing, and the blood pressure from 19:00:00 to 22:00:00 on 2024-04-24 is missing." By templating the missing risk segment prompt, a reliable prompt sub-template can be provided for subsequent Agent questions, thereby improving the accuracy of Agent questions;

[0191] The large model-based prompt word template is combined with the Agent to complete the patient's life situation to determine the actual risk hazards of the patient, specifically including:

[0192] Based on the agent tool and the preset completion data prompt template, the agent big model is called to complete the missing life risk segments of the patient, and the life risk segments after completion are recalculated through step S7.1 to obtain the risk hazard similarity value, and according to the preset risk hazard accuracy threshold, the risk hazard similarity value is less than the preset risk hazard accuracy threshold, so that the agent big model finds suitable repetition time intervals for different risk hazards according to the preset time interval prompt template, until the risk hazard similarity value is greater than the preset risk hazard accuracy threshold, thereby outputting the actual risk hazard of the patient, including:

[0193] Based on the agent tool and the preset completion data prompt template, the agent model is called to complete the missing life risk fragments of the patient, including:

[0194] The preset completion data prompt template may be "The differential risk segments that exist are: as shown in Table 4

[0195] The missing key risk fragments are as shown in Table 5

[0196] Existing life risk episodes include: For example:

[0197] Table 6

[0198]

[0199] Table 6 shows the life risk segment data generated by the patient wearing the wearable device in daily life

[0200] The various tools preset by Agent include: blood pressure monitor Agent tool, pedometer Agent tool, indoor infrared thermometer Agent tool, camera Agent tool, sleep monitoring device Agent tool, flexible ECG recorder Agent tool, etc.

[0201] The current patient information is: For example: Patient name: Zhang San; Age: 64; Gender: Male; Current recovery stage: Mid-term recovery

[0202] The suspected risk factors of the patient are: not taking medication regularly;

[0203] You are an intelligent interactive robot. You need to rely on the above information to determine whether to execute the tools preset by the Agent to determine the patient's actual life risks.

[0204] The combination of the Agent to complete the patient's life situation and determine the actual risk hazards of the patient specifically includes: the Agent executes the corresponding patient life situation completion action according to the instructions in the prompt word template, thereby determining the actual risk hazards of the patient. For example: if the template indicates that the patient may not take medication regularly, and the data for the corresponding time period is missing, the Agent will call the blood pressure monitor to obtain recent blood pressure data to find the data for the corresponding time period, complete the life risk segment data for the corresponding time period, and complete the complete life risk data;

[0205] Based on the completed life risk fragment, the step S7.1 is recalculated to obtain the risk hazard similarity value, and according to the preset risk hazard accuracy threshold, the risk hazard similarity value is less than the preset risk hazard accuracy threshold, so that the agent large model finds suitable repetition time intervals for different risk hazards according to the preset time interval prompt template, so as to output and determine the actual risk hazard of the patient, including: based on the completed life risk fragment, the step S7.1 is recalculated to obtain the risk hazard similarity value, and according to the preset risk hazard accuracy threshold, the risk hazard similarity value is less than the preset risk hazard accuracy threshold, so that the agent large model finds suitable repetition time intervals for different risk hazards according to the preset time interval prompt template and the agent tool,

[0206] The preset time interval prompt template can be "

[0207] Risks: xxx

[0208] Next execution time: Get the time interval for saving the xxx risk hidden danger in the hospital database, and start repeated execution according to the time interval. If the xxx risk hidden danger has no time interval saved in the database, initialize the xxx risk hidden danger time interval, as shown in Table 1 and Table 2. You are an intelligent analysis robot. You need to rely on the above information and set a suitable time interval for the next execution according to the risk hidden danger. Until the similarity value of the risk hidden danger is greater than the preset risk hidden danger accuracy threshold, the actual risk hidden danger of the patient is determined by output. Through multiple rounds of scheduling and setting the corresponding time interval for re-detection according to the risk hidden danger category, the problem of inaccurate risk hidden danger judgment can be effectively compensated, thereby improving the accuracy of risk hidden danger detection;

[0209] The blood pressure monitor Agent tool: a tool used to measure the patient's daily blood pressure, and has an API interface for the Agent to call;

[0210] The pedometer Agent tool: a tool used to calculate the daily steps of the patient, and there is an API interface for the Agent to call;

[0211] The indoor infrared thermometer Agent tool: a tool used to measure the patient's daily body temperature data, and there is an API interface for the Agent large model to call;

[0212] The camera Agent tool is used to identify the patient's daily life, assist in calculating the number of daily steps and daily activity, etc., and has an API interface for the Agent model to call;

[0213] The sleep monitoring device Agent tool: an Agent tool used to calculate the patient's sleep duration, sleep frequency and sleep quality, and has an API interface for the Agent large model to call;

[0214] The flexible ECG recorder Agent tool is a lightweight ECG sticker that can monitor dynamic electrocardiogram, heart rate, respiratory rate, etc., and has an API interface for the Agent large model to call.

[0215] S7.3: After multiple rounds of intelligent follow-up visits based on the big model, medical staff are reminded to conduct manual follow-up visits, including:

[0216] Determine the patient's risk hazards through S1-S7.2, calculate the initial determination follow-up score according to the constructed follow-up formula, if the set follow-up threshold is reached, the agent will follow up the patient for the first round, communicate with the patient through a virtual phone, obtain the patient's daily habits and intervention level based on the communication content with the patient, and generate a risk report summary. If the patient does not connect to communicate with the agent, the corresponding virtual phone follow-up weight is accumulated, and as the number of follow-up visits increases, the patient's repeated risk hazards will accumulate the corresponding follow-up weight, and the corresponding intervention level weight is adjusted according to the degree of intervention for the patient. During the follow-up process, if S7.22 detects that the number of differential risk segments or missing key risk segments reaches the set differential missing threshold, S4 and S7.21 will remind the patient to re-wear and improve the high-precision patient physiological data and the low-precision The accuracy of the patient's physiological data is improved, thereby improving the accuracy of risk detection for the patient. At the same time, in order to obtain the patient's current anxiety level during the follow-up period and fill it into the risk report during the interaction with the patient, the follow-up score reaches the preset manual follow-up threshold and enters the manual stage. By collating historical risk reports, a long-term risk report summary is formed and sent to the doctor to intervene in the patient. In the process of historical multiple rounds of follow-up, the agent large model intercepts the risk data prompt template according to the preset template to obtain the life risk data fragments with fewer risk hazards and the corresponding time of the life risk data fragments, so as to find the least life risk data fragments and the least life risk data fragments corresponding time for the risk hazards, and also find a suitable time interval for confirming the risk hazards in S7.22, so as to improve the calculation efficiency, thereby improving the accuracy of determining risk hazards for the next round of follow-up;

[0217] Determining the risk hazard of the patient through S1-S7.2, and calculating the risk hazard score value determined for the first time according to the constructed follow-up formula, includes:

[0218] In the first round, the risk hazards of the patient are determined through S1-S7.2. According to the constructed follow-up formula, the follow-up formula is S=a1*b1+a2*b2+a3*b3+a4*b4, S is the follow-up score, a1 is the patient's age value, a2 is the risk hazard type value, a3 is the intervention degree value, a4 is the number of unconnected times value, b1, b2, b3 and b4 are weight coefficients, and the follow-up score is calculated for the first time. The b1 is initialized to 0.1. For example, a1 is initially set to 30 for 20-30 years old, 50 for 70-90 years old, etc., a2 is 30 for "lack of sleep",

[0219] "Not taking medication regularly" is 50 points, a3 and a3 are initially set to 0, and the initial weight coefficients corresponding to each parameter are set to 0.1, 0.5, 0.2 and 0.3. Therefore, through the return visit formula, the risk score of "not taking medication regularly" in the age group of 70-90 years old is determined to be S=35 points. If the risk is a serious problem such as "forgetting to charge the left ventricular assist device or abnormal ventricular assist pump circulation", the return visit score S is set to 100 points;

[0220] If the set return visit threshold is reached, the agent will return the patient for the first round, communicate with the patient through a virtual phone, obtain the patient's daily habits and intervention level based on the communication content with the patient, and generate a risk report summary, including: if the set return visit threshold is reached, the risk hazards determined for the patient are transmitted to a preset prompt template for intervening in the patient's daily behavior, the agent model executes corresponding tools according to the instructions of the prompt template for intervening in the patient's daily behavior, interacts with the patient through a virtual phone, obtains the patient's erroneous behaviors under the risk hazards, generates a risk report for the patient's erroneous behaviors under the risk hazards and the corresponding correct behaviors, and prompts the patient with the correct steps based on the risk report, and asks what the last prompt behavior was. Whether the patient's risk is corrected, and the degree of improvement of the physical condition, so as to achieve the effect of intervening in the patient's daily behavior, and generate a risk report summary and calculate the intervention degree value, for example: the risk hidden dangers of the patient determined are transmitted to the preset prompt template for intervening in the patient's daily behavior, and the agent model executes the corresponding tools according to the instructions of the prompt template for intervening in the patient's daily behavior, interacts with the patient, obtains the patient's wrong behavior under the risk hidden danger, generates a risk report for the patient's wrong behavior under the risk hidden danger and the corresponding correct behavior, and prompts the patient with the steps of correct behavior according to the risk report, and asks whether the last prompt behavior has been corrected, and the degree of improvement of the physical condition after the interference, so as to achieve the effect of intervening in the patient's daily behavior, including:

[0221] The intervention prompt template for the patient's daily behavior can be "

[0222] Correct Behavior: Get a Rehabilitation Plan via S2

[0223] Risk: "xxx"

[0224] The various tools preset by the agent include: the patient's wearable device agent tool, mobile device agent tool, etc.

[0225] You are an intelligent interactive robot. You need to rely on the above information, obtain the correct behavior corresponding to "xxx" in the rehabilitation plan according to the risk "xxx", determine whether to execute the tool preset by the Agent, ask the patient which guidance rule is not achieved under the correct behavior, and generate a risk report for the risk "xxx" and the guidance rule is not achieved under the correct behavior. "The risk can be "not taking medicine regularly" and the correct behavior is "medication management";

[0226] For example: the risk hazard "not taking medicine regularly" of the patient is transmitted to the preset prompt template for intervening the patient's daily behavior. The agent model calls the patient's wearable device or mobile device according to the instructions of the prompt template for intervening the patient's daily behavior, and interacts with the patient through voice or text messages, etc. According to the risk hazard of "not taking medicine regularly", it corresponds to the correct "medication management" behavior of the rehabilitation plan in S2, obtains the correct "medication management" behavior, and asks the specific details of "medication management" in turn. For example: through the wearable device or mobile device, ask the patient whether the medication has been stopped by voice or text message. If the patient answers no, then according to the setting of S252 "taking time management" in the "medication management" behavior of the rehabilitation plan, ask the patient about the time period for taking medication, and the setting of S251 "drug type and dosage" asks the patient about the dosage of medication, so as to collect the patient's answer content and fill it into the risk report, and also mark the correct medication time and dosage in the rehabilitation plan in the risk report, corresponding to the lower part of the actual medication time and dosage of the patient. After the patient's medication information is collected, it is stored in the risk report; secondly, the patient is prompted by voice according to the correct medication time and dosage in the rehabilitation plan through the patient's wearable device or mobile device. Finally, the patient is asked whether the last prompt behavior effectively corrected the patient's wrong behavior. For example, the last prompt behavior may be to stipulate that the patient takes medicine at 8-9 am and 6 pm, and the patient's last wrong behavior is to take medicine only at 6 pm. The patient is asked by voice or text message whether he takes medicine at 8-9 am. If the patient answers "yes", and the patient is asked about the degree of physical improvement of the patient after the interference, such as "not good", "good" or "very good", so as to achieve the effect of intervening in the patient's daily behavior. In the first round of follow-up, the question of whether the patient's last prompt behavior effectively corrected the patient's wrong behavior is omitted, and the second round begins to ask the patient whether the last prompt behavior effectively corrected the patient's wrong behavior;

[0227] If the patient fails to communicate with the agent, the corresponding virtual telephone return visit weight is accumulated, and as the number of return visits increases, the patient repeatedly has the same risk hazards, and the corresponding return visit weight is accumulated. According to the degree of intervention for the patient, the corresponding intervention degree weight is adjusted, including:

[0228] If the patient fails to communicate with the agent, the corresponding virtual phone call return weight is accumulated. For example, if the agent communicates with the patient through the virtual phone and the patient does not accept the agent's inquiry, the corresponding virtual phone call return weight is accumulated. If the initial weight coefficient is 0.2, it is increased to 0.3, and each time it is increased by 0.1 until the weight coefficient is 1;

[0229] As the number of return visits increases, the patient's risk hazards recur, and the corresponding return visit weights are also accumulated. For example, as the number of return visits increases, the patient's risk hazards are determined through S1-S7.2, and the same risk hazards recur. The risk hazard determined in the first return visit is "lack of sleep" with a weight coefficient of 0.5. The risk hazard determined in the second return visit is also "lack of sleep", then the weight coefficient is increased by 0.1 on the original basis, and the third return visit is also increased by 0.1 on the second basis, and so on for the fourth and fifth times, until the risk hazard weight coefficient is 1;

[0230] If the intervention degree is good, the corresponding weight is reduced. For example, by communicating with the patient, information about the patient after the last intervention is obtained from the communication, such as "bad", "good", "very good", etc. The weight coefficient of "bad" is increased by 0.1 on the original basis, the weight coefficient of "good" is reduced by 0.05 on the original basis, and the weight coefficient of "very good" is reduced by 0.1 on the original basis. The range of the numerical weight coefficient of the intervention degree is 0-1;

[0231] If the number of differential risk segments or missing key risk segments detected by S7.22 reaches the set differential missing threshold, the patient is reminded to put on the device again and improve the accuracy of the high-precision patient physiological data and the low-precision patient physiological data through S4 and S7.21, thereby improving the accuracy of risk detection for the patient, including:

[0232] During the multiple rounds of follow-up visits to the patient by the agent, if S7.22 detects that the number of differential risk segments or missing key risk segments reaches the set differential missing threshold, for example, if the differential risk segments reach more than 4 and the set differential missing threshold is 3, then S4 and S7.21 are used to remind the patient to wear the device again and improve the accuracy of the high-precision patient physiological data and the low-precision patient physiological data, thereby improving the accuracy of risk detection of the patient;

[0233] At the same time, in order to obtain the patient's current anxiety level during the interaction with the patient during the follow-up visit and fill it into the risk report, it includes:

[0234] At the same time, in order to obtain the patient's current anxiety level during the interaction with the patient during the follow-up time, a prompt template for obtaining the patient's current psychological anxiety level is preset. The agent model executes corresponding tools according to the instructions of the prompt template for obtaining the patient's current psychological anxiety level, inquires about the patient's anxiety level, thereby obtaining the patient's anxiety level data and filling it into the risk report, including: At the same time, in order to obtain the patient's current anxiety level during the interaction with the patient during the follow-up time, a prompt template for obtaining the patient's current psychological anxiety level is preset. The agent model executes corresponding tools according to the instructions of the prompt template for obtaining the patient's current psychological anxiety level, inquires about the patient's anxiety level, thereby obtaining the patient's anxiety level data and filling it into the risk report, including: The prompt template for the patient's current psychological anxiety level can be "a way to construct a question: please construct a question to ask the patient whether they can score their current anxiety level according to a scoring standard of 1 to 10, where 1 represents no anxiety and 10 represents extreme anxiety.

[0235] The various tools preset by Agent include: the patient-worn device agent tool,

[0236] The agent model calls the preset tool according to the instructions of the prompt word template, for example: the agent model constructs a question according to the instructions of the prompt word template: "Please rate your current anxiety level based on your feelings, using a rating scale of 1 to 10. 1 means you feel completely free of anxiety, and 10 means you feel extremely anxious. Please tell me your rating, for example, 'I rate 3'." The agent model calls the agent tool of the device worn by the patient and asks the patient, so as to collect the patient's answers and fill them into the risk report for reference by medical staff;

[0237] The revisit score calculated by the revisit formula reaches the preset manual revisit threshold, enters the manual stage, and forms a long-term risk report summary by collating historical risk reports, and sends it to the doctor to intervene in the patient, including: the revisit score calculated by the revisit formula reaches the preset manual revisit threshold, and the preset manual revisit threshold can be set to 80 points. When entering the manual stage, the historical risk reports are sorted, for example: the historical risk reports are input into the preset risk report summary prompt template, so that the agent large model generates a risk report summary, and sends the risk report summary to medical staff by email or SMS transmission, and reminds the medical staff to check, including:

[0238] Enter the risk report into the preset risk report summary prompt template,

[0239] The preset risk report summary prompt template can be "

[0240] Risk Report: Get the risk report generated above

[0241] Risk Report Summary Format:

[0242] “Dear medical staff,

[0243] I am sending you this risk report summary on behalf of the intelligent agent system for patient [name of medical institution]. The following is a summary of the risk assessment for patient [name of patient]:

[0244] Patient basic information: patient name, age and gender, etc.

[0245] Health status: Over the past [time frame], the patient [description of health status].

[0246] Anxiety level: Based on the patient's self-assessment, the anxiety level is [patient anxiety score].

[0247] Failure to perform incorrect behaviors as described in the rehabilitation plan instructions: The rehabilitation instructions that the patient failed to follow include [list of specific incorrect behaviors].

[0248] The effect of the correct behavioral intervention: Through [the correct behavioral intervention measures], the patient's daily behavior has shown [effect description], and the physical condition [has improved].

[0249] Change in Mental State: The patient's mental state changes from [previous mental state] to [current mental state].

[0250] Please refer to the full text of the risk report in the attachment for more detailed information. We recommend that you review the report as soon as possible after receiving this notification and take appropriate medical actions based on the content of the report.

[0251] Thank you for your concern and commitment to your patients' health.

[0252] Sincerely,

[0253] [Medical institution name] Intelligent agent system

[0254] [Date sent]"

[0255] ", so that the agent big model generates a long-term risk report summary, and the long-term risk report summary is sent to the doctor through email or text message to intervene in the patient;

[0256] In the process of historical multiple rounds of return visits, the agent model obtains the life risk data segment with fewer risk hazards and the corresponding time of the life risk data segment according to the preset interception risk data prompt template, so as to find the minimum life risk data segment and the minimum life risk data segment corresponding time for the risk hazard, and also finds a suitable time interval for confirming the risk hazard in S7.22, improves the calculation efficiency, and thus improves the accuracy of determining the risk hazard for the next round of return visits, including: in the process of historical multiple rounds of return visits, the agent model intercepts the risk data prompt template according to the preset interception risk data prompt template, which can be:

[0257] The current time interval for large models to call risk hazards is x hours;

[0258] The big model has currently reached the following round of risk identification: Round x;

[0259] Currently suspected risks: xxx risks;

[0260] Whether the patient has identified the current risk hazards through telephone follow-up: "Clear xxx risk hazards" or "No risk hazards exist";

[0261] Risk hazards correspond to key risk segments, and key risk segments correspond to time: Table x (for example, Table 1 in the embodiment)

[0262] You are an intelligent analysis robot. You need to judge whether the frequency of risk hazards confirmed by the comprehensive x-round big model is reliable based on whether the patient has identified the current risk hazards through telephone follow-up, and output the shortest time interval for calling the big model that can identify the risk hazards.

[0263] The method for judging whether the frequency of risk hazards confirmed by the comprehensive x-round big model is reliable is: according to whether the patient has clarified the current risk hazards through telephone follow-up, if it is clear, the life risk data fragments corresponding to the risk hazards determined in the previous round and the time-intercepted parts of the life risk data fragments are sent to the risk hazard similarity model to calculate whether the intercepted parts can achieve the same risk similarity. If the similarity reaches the same threshold, the time interval for the big model to call the risk hazards is shortened according to the time corresponding to the intercepted life risk data fragments. You can make multiple attempts to obtain the best time interval that meets the same risk similarity; if the patient cannot clarify the current risk hazards through telephone follow-up, the time interval for the big model to call the risk hazards is not adjusted. Then the time interval of the risk hazard is saved in the hospital database. "

[0264] The patient's wearable device agent tool has the function of interacting with the patient, such as voice interaction and SMS interaction, and has an API interface for the Agent large model to call

[0265] The mobile device agent tool is used to interact with the patient through voice, and there is an API interface for the Agent model to call

[0266] And its effect: The method proposed by the present invention is based on the agent scheduling system and smart wearable, which can obtain the vital signs data of patients with left ventricular assist device in real time, and remind medical staff to make regular follow-up visits through artificial intelligence algorithms such as deep learning, so that patients can receive rehabilitation treatment in time, reduce the occurrence of abnormal ventricular assist pump circulation, and reduce the postoperative risks of patients.

[0267] In this embodiment, a computer device includes a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the at least one program to execute the intelligent rehabilitation method for receiving a left ventricular assist device in the embodiment to achieve the same technical effect as described in the embodiment.

[0268] In this embodiment, a storage medium stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the intelligent rehabilitation method for receiving a left ventricular assist device in the embodiment to achieve the same technical effect as described in the embodiment.

[0269] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it may be directly fixed or connected to another feature, or it may be indirectly fixed or connected to another feature. In addition, the descriptions of up, down, left, right, etc. used in this disclosure are only relative to the relative positional relationship of the components of the present disclosure in the accompanying drawings. The singular forms of "a", "said" and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates other meanings. In addition, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as those generally understood by technicians in this technical field. The terms used in the specification of this embodiment are only for describing specific embodiments, not for limiting the present invention. The term "and / or" used in this embodiment includes any combination of one or more related listed items.

[0270] It should be understood that, although the term first, second, third etc. may be adopted to describe various elements in the present disclosure, these elements should not be limited to these terms. These terms are only used to distinguish the same type of elements from each other. For example, without departing from the scope of the present disclosure, the first element may also be referred to as the second element, and similarly, the second element may also be referred to as the first element. The use of any and all examples or exemplary language ("for example", "such as" etc.) provided by the present embodiment is only intended to better illustrate embodiments of the present invention, and unless otherwise required, the scope of the present invention will not be limited.

[0271] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in an assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed ASIC for this purpose.

[0272] In addition, the operations of the process described in this embodiment may be performed in any suitable order, unless otherwise indicated in this embodiment or otherwise clearly contradicted by the context. The process described in this embodiment (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as a code (e.g., executable instructions, one or more computer programs, or one or more applications) executed jointly on one or more processors, by hardware or a combination thereof. The computer program includes a plurality of instructions that may be executed by one or more processors.

[0273] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or a portion thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0274] The computer program can be applied to input data to perform the functions described in the present embodiment, thereby converting the input data to generate output data stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on the display.

[0275] The above is only a preferred embodiment of the present invention. The present invention is not limited to the above implementation. As long as the technical effect of the present invention is achieved by the same means, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the present invention. Within the scope of protection of the present invention, its technical scheme and / or implementation method may have various modifications and changes.

Claims

1. An intelligent rehabilitation method for patients receiving a left ventricular assist device, characterized in that: include: Obtain patient data through the hospital server and create a patient information table; Establishing a rehabilitation plan based on the patient information form, wherein the rehabilitation plan includes exercise rehabilitation, sleep time, dietary guidance, anxiety level assessment and classification, and medication management; Based on wearable devices, the patient's life physiological data is obtained, and according to the event marking method and in combination with the rehabilitation plan, life risk fragments and risk hazards are marked; Obtain a preset number of patients, obtain the motion detection start cycle, patient motion detection frequency, patient motion type, patient motion time, and life risk segments of each patient, and perform real-time detection on the patients in combination with high-precision professional detection equipment in the hospital to obtain high-precision patient physiological data. At the same time, perform real-time detection on the patients in combination with the wearable device to obtain low-precision patient physiological data, supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement, compare and count the differences between the high-precision patient physiological data and the low-precision patient physiological data, and obtain a list of differences in the patient physiological data life segments; According to the labeled life risk fragment data, combined with the LSTM deep learning algorithm, a patient life risk prediction model is trained; Obtain patient information tables of a preset number of patients, obtain life risk fragments and risk hazards of the preset patients, and construct a patient risk similarity information table; According to the patient risk similarity information table, a patient hemodynamic risk similarity model is constructed in combination with a similarity algorithm; According to the patient's life risk prediction model, predict the various life risk segments in the patient's life, combine the patient's hemodynamic risk similarity model to predict risk hazards, and obtain the possible risk hazards of the patient. According to the patient's physiological data life segment difference list, determine whether the patient's life risk segment is a differential risk segment, and convert the differential risk segment into prompt text. According to the preset key risk segments corresponding to the possible risk hazards of the patient, determine whether the patient's life risk segment lacks a certain key risk segment, and convert it into prompt text. Based on the large model prompt word template, combine the Agent to complete the patient's life situation, determine the patient's actual risk hazards, and remind medical staff to conduct manual return visits after multiple rounds of intelligent return visits based on the large model; Among them, the supplementation of high-precision patient physiological data and low-precision patient physiological data by data enhancement includes: determining the patient's electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, activity amount and the time period of the life risk segment by the wearable device in the life risk segment according to the life risk segment marked by the patient, and judging the activity amount detected by the wearable device to which the activity amount belongs, the movement detection start cycle, patient movement detection frequency, patient movement type and patient movement time, so as to determine the time period, movement detection start cycle, patient movement detection frequency, patient movement type and patient movement time of the life risk segment that the patient needs to simulate in the data enhancement; The patient maintains a life segment simulation of the movement detection start cycle, patient movement detection frequency, patient movement type, and patient movement time in the time period of the life risk segment within the preset days of admission. During the life segment simulation process, the high-precision professional detection equipment and the wearable device are combined to collect and detect high-precision patient physiological data and low-precision patient physiological data, so as to supplement the high-precision patient physiological data and the low-precision patient physiological data by data enhancement; Conduct similarity judgment on the captured risk segments to determine whether the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level are consistent with the previously marked segments; Determine the similarity between the low-precision patient physiological data obtained by data enhancement and the low-precision patient physiological data detected by the patient's annotated life risk fragments. If the similarity is lower than a threshold, repeat the method of supplementing the high-precision patient physiological data and the low-precision patient physiological data by data enhancement.

2. The intelligent rehabilitation method for patients receiving a left ventricular assist device according to claim 1, characterized in that: The wearable device is used to obtain the patient's life physiological data, and the life risk segments and risk hazards are marked according to the event marking method and the rehabilitation plan, including: The wearable device includes monitoring technology for blood oxygen saturation, blood pressure, respiration, and body position and movement. The parameters that can be detected include electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level; The wearable devices include wrist-worn, smart vest, and head-worn; The patient's life physiological data includes: based on the detection parameters obtained by the wearable device, recording the corresponding relationship between the detection parameters and time, that is, the correspondence between the detection parameters every minute and every second of every day, so as to obtain the patient's life physiological data; The event marking method is combined with the rehabilitation plan to mark life risk segments and potential risks, including: requiring the patient, family members or doctors to view the patient's life physiological data in the app, record the life physiological data by themselves, and combine the rehabilitation plan content to determine whether it is a life risk segment, and mark the time period when the life risk segment occurs. The life risk segment includes sports type risk segments and physiological value type risk segments; The risks include: failure to take medication regularly, dietary problems, drinking, smoking, swimming, postoperative complications, wound infection, fear of exercise, breathing difficulties, lack of sleep, fatigue, other discomfort, forgetting to charge the left ventricular assist device, abnormal ventricular assist pump circulation; postoperative complications include one or more of pneumonia, thrombosis and joint contracture; The sports type risk segment includes: the family members or doctors detect the running type data of the patient in the app based on the patient's life physiological data, including the number of steps, respiratory rate, and activity, to determine whether the patient's running status in this time period is normal, thereby marking the life risk segment; the physiological value type risk segment includes: the family members or doctors detect the physiological value type data of the patient in the app based on the patient's life physiological data, including electrocardiogram, electromyogram, heart rate, heart rate variability, blood pressure, blood oxygen saturation, and body surface temperature, to determine whether the patient's physiological value status in this time period is normal, thereby marking the life risk segment; The risk hazards also include: presetting key risk segments corresponding to each risk hazard, and the key risk segments include: the time point of occurrence of the key risk segments, the key movement type risk segments, the key physiological value type risk segments, the frequency of occurrence of related key risk segments, and key patient information.

3. The intelligent rehabilitation method for left ventricular assist device recipients according to claim 1, characterized in that: The method of obtaining a preset number of patients, obtaining the motion detection start cycle, patient motion detection frequency, patient motion type, patient motion time, and life risk segments of each patient, and performing real-time detection on the patient in combination with high-precision professional detection equipment in the hospital to obtain high-precision patient physiological data, and at the same time performing real-time detection on the patient in combination with the wearable device to obtain low-precision patient physiological data, supplementing the high-precision patient physiological data and the low-precision patient physiological data by data enhancement, and comparing and counting the differences between the high-precision patient physiological data and the low-precision patient physiological data to obtain a list of differences in patient physiological data life segments, specifically includes: The high-precision professional testing equipment in the hospital includes: monitoring blood pressure by listening to Korotkoff sounds based on manual blood pressure cuff, or using Doppler ultrasound to monitor blood pressure; The high-precision professional testing equipment in the hospital also includes the use of high-precision equipment to detect heart rate, electrocardiogram, electromyogram, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level; The wearable device is used to detect the patient in real time to obtain low-precision physiological data of the patient, including using the wearable device to detect the patient's electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity level; The comparing and counting the differences between the high-precision patient physiological data and the low-precision patient physiological data to obtain a list of differences in the patient's physiological data life segments includes: Determine whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold value. If so, the life risk segments and risk hazards corresponding to the difference are included in the patient physiological data life segment difference list; The determining whether the data difference of the electrocardiogram, electromyogram, heart rate, heart rate variability, respiratory rate, blood pressure, blood oxygen saturation, body surface temperature, number of steps, and activity amount detected by the high-precision patient physiological data and the low-precision patient physiological data exceeds a threshold also includes: determining whether the abnormalities of the high-precision patient physiological data and the low-precision patient physiological data are consistent, and including the life risk segments and potential risks corresponding to the differences into the patient physiological data life segment difference list.

4. The intelligent rehabilitation method for patients receiving a left ventricular assist device according to claim 1, characterized in that: The life risk existence prediction model of the patient is trained according to the labeled life risk segment data in combination with the LSTM deep learning algorithm, specifically including: obtaining the life physiological data of the patient with a preset number of life risk segment data and the life physiological data of the patient with ordinary life risk-free segment data, the patient information table data corresponding to the patient is used as the LSTM algorithm input parameter, and whether the life segment corresponds to the life risk segment is used as the output parameter to train the patient life risk existence prediction model; The ordinary life risk-free segment data corresponds to the preset number of life risk segment data, and the ordinary life risk-free segment data is the ordinary patient life physiological data that is not marked as risky.

5. The intelligent rehabilitation method for patients receiving a left ventricular assist device according to claim 1, characterized in that: The step of obtaining the patient information table of a preset number of patients, obtaining the life risk segments and risk hazards of the preset patients, and constructing the patient risk similarity information table specifically includes: Obtain patient information tables, corresponding life risk segments, and risk hazards of a preset number of patients, obtain life risk segments and corresponding risk hazards that occur in the life of each patient, and fill each field in the patient information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, the physiological value type or exercise type corresponding to the life risk segment, and the risk hazard into the patient risk similarity information table; According to the patient risk similarity information table, a patient hemodynamic risk hazard similarity model is constructed in combination with a similarity algorithm, specifically including: according to each field in the patient information table in the patient risk similarity information table, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or movement type corresponding to the life risk segment, a patient hemodynamic risk hazard similarity model is constructed using the bm25 similarity algorithm, the above-mentioned characteristic values ​​of the patient who needs to predict the risk hazard are input to perform similarity calculation with other users, the user with the highest similarity is obtained, the risk hazard of the user is judged, and the possible risk hazard of the patient who needs to predict the risk hazard is obtained.

6. The intelligent rehabilitation method for patients receiving a left ventricular assist device according to claim 1, characterized in that: The method further comprises: obtaining the patient's life data in combination with the wearable device, predicting the various life risk segments in the patient's life through the patient's life risk prediction model, obtaining the patient information table of the patient, predicting the various life risk segments in the patient's life according to each field in the patient information table of the patient, the time point corresponding to the life risk segment, each field of the patient's life physiological data corresponding to the life risk segment, and the physiological value type or exercise type corresponding to the life risk segment in combination with the patient's hemodynamic risk hazard similarity model, obtaining the possible risk hazards of the patient, and generating risk hazards and key risk segments corresponding to the risk hazards, as well as a corresponding time table; The method of judging whether the patient's life risk segment is a differential risk segment based on the patient's physiological data life segment difference list and texting the differential risk segment prompt specifically includes: obtaining the life risk segment within a recent preset time, and calling the corresponding tool through the preset similarity number prompt template, and calculating the number of similarities with the patient's physiological data life segment difference list. When the number of occurrences exceeds a threshold, the life risk segment and the difference segment list are transmitted to the preset re-wearing prompt template, and a request for the patient to re-wear the wearable device is initiated; the life risk segment of the patient is calculated to have a similarity with each life risk segment in the patient's physiological data life segment difference list. If the life risk segment has a similarity value greater than the threshold, the life risk segment of the patient is judged to be a differential risk segment, and the situation where the differential risk segment has an abnormal value is obtained, and the abnormal value situation is inserted into the differential risk segment prompt template; The method of determining whether the patient's life risk fragments lack a certain key risk fragment according to the preset key risk fragments corresponding to the potential risk hazards of the patient, and texting the prompt words, and completing the patient's life situation based on the large model prompt word template in combination with the Agent to determine the actual risk hazards of the patient, specifically includes: according to the table data where the patient's risk hazards are stored, the table data stores the preset key risk fragments corresponding to the patient's risk hazards, and based on the preset missing data prompt template, calling the agent large model to find and analyze whether the key risk fragments of the patient's risk hazards are missing; Calculate the similarity between the patient's life risk segment and each life risk segment in the patient's physiological data life segment difference list; if the similarity value is greater than the threshold life risk segment, determine that the patient's life risk segment is a differential risk segment; obtain the abnormal value of the differential risk segment, and put the abnormal value into the differential risk segment prompt template; Based on the agent tool and the preset completion data prompt template, the agent big model is called to complete the missing life risk fragments of the patient, and the risk hidden danger similarity value is recalculated based on the completed life risk fragment to obtain the risk hidden danger similarity value, and according to the preset risk hidden danger accuracy threshold, the risk hidden danger similarity value is less than the preset risk hidden danger accuracy threshold, and the agent big model finds suitable repetition time intervals for different risk hidden dangers according to the preset time interval prompt template, until the risk hidden danger similarity value is greater than the preset risk hidden danger accuracy threshold, thereby outputting the actual risk hidden danger of the patient.

7. The intelligent rehabilitation method for patients receiving a left ventricular assist device according to claim 1, characterized in that: After multiple rounds of intelligent follow-up visits based on the large model, the medical staff is reminded to conduct manual follow-up visits, which specifically include: determining the risk hazards of the patient, calculating the initial follow-up score according to the constructed follow-up formula, and if the set follow-up threshold is reached, the agent will follow up the patient in the first round, communicate with the patient through a virtual phone, obtain the patient's daily habits and intervention level based on the communication content with the patient, and generate a risk report summary. If the patient does not connect to communicate with the agent, the corresponding virtual phone follow-up weight is accumulated, and as the number of follow-up visits increases, the patient's repeated risk hazards will accumulate the corresponding follow-up weight, and the corresponding intervention level weight will be adjusted according to the degree of intervention for the patient. During the follow-up process, When the number of differential risk segments or missing key risk segments reaches the set differential missing threshold, the patient is reminded to put on the device again and improve the accuracy of the high-precision patient physiological data and the low-precision patient physiological data, the patient's current anxiety level is obtained and filled in the risk report, and the follow-up score reaches the preset manual follow-up threshold, entering the manual stage, and by collating historical risk reports, a long-term risk report summary is formed and sent to the doctor to intervene in the patient. During the historical multiple rounds of follow-up, the agent large model intercepts the risk data prompt template according to the preset template, obtains the corresponding time of the life risk data segment with fewer risk hazards and the life risk data segment with more risk hazards, and confirms the time interval for confirming the risk hazards.

8. A computer device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the at least one program to execute the method according to any one of claims 1 to 7.

9. A storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 7 when executed by the processor.

Images