Artificial intelligence-based nephropathy peritoneal dialysis full life cycle management device, system and method

Abstract

The invention relates to the technical field of medical instruments and peritoneal dialysis, in particular to a nephropathy peritoneal dialysis full life cycle management device, system and method based on artificial intelligence, and the method comprises the following steps: 1, obtaining life record information, medical record information and household medical equipment terminal data of a patient, wherein the household medical equipment terminal data comprises data from an automatic peritoneal dialysis machine; step 2, performing data verification, data checking and data spot check on the data; and step 3, generating suggestion information of daily guidance, diagnosis and treatment evaluation and special medical food by the artificial intelligence model according to the normal data provided in the step 2. According to the invention, the problems of incomplete nephropathy management chain, low data quality and lack of personalized management in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices and peritoneal dialysis, and in particular to an artificial intelligence-based full life cycle management device, system and method for peritoneal dialysis for kidney disease. Background Art

[0002] In the terminal stage of chronic kidney disease (CKD), i.e. the uremia stage, renal replacement therapy (including hemodialysis, peritoneal dialysis and kidney transplantation) becomes an important means to maintain the patient's life. Among them, peritoneal dialysis, especially automated peritoneal dialysis, is easier to operate than hemodialysis. Its advantage is that it can provide a dialysis effect that is closer to the physiological state. It is especially suitable for home environment, which significantly improves the patient's quality of life and freedom. It is not difficult to buy a home peritoneal dialysis machine, but the current application of home peritoneal dialysis in China is less than that in developed countries. The main reasons are: first, there is a lack of supporting home peritoneal treatment technical support and a scientific and systematic follow-up management system, which affects patients' timely access to professional medical support, thereby affecting the dialysis effect and health status; second, the patient's education and self-management ability is insufficient, resulting in a lack of understanding of peritoneal dialysis operation and management, affecting the dialysis effect and quality of life; third, the uneven distribution of medical resources and the lack of technical support restrict patients in remote and economically underdeveloped areas from carrying out home peritoneal dialysis treatment.

[0003] At present, some peritoneal dialysis remote monitoring systems (such as patent 202311316020.5) have been developed on the market and can be connected to peritoneal dialysis machines. They mainly involve remote monitoring of peritoneal dialysis, dialysis early warning, and disease classification diagnosis and evaluation. However, in actual use, such remote monitoring systems cannot provide full-chain medical technology support to form complete, scientific, and effective management. The problem is:

[0004] 1. Low data quality: The current peritoneal dialysis system mainly relies on manual operations of patients in data collection and processing, such as recording dialysis diaries and monitoring vital signs. This process is not only cumbersome and error-prone, which increases the burden on patients, but also makes it difficult to ensure the integrity, accuracy and timeliness of the data.

[0005] 2. Lack of personalized management: Existing peritoneal dialysis treatment programs often adopt standardized clinical pathways and lack personalized management. However, the conditions of kidney patients are often more complex and individual, and the effect of standardized treatment is not good;

[0006] 3. Imperfect management chain: The management of kidney disease involves treatment, nursing, diet, exercise and other aspects, and also covers hospitals, homes, outdoors and other scenarios. The current peritoneal dialysis system only involves home and daily care, and has not yet penetrated into the fields of diet, treatment, exercise and so on.

[0007] 4. Kidney disease detection and early warning are not timely and accurate enough: The existing system relies too much on patients' regular hospital visits for treatment effects and complication monitoring, and is unable to achieve early warning and real-time intervention, which increases the risk of worsening complications and increases patients' suffering and treatment costs. Summary of the invention

[0008] The present invention aims to provide an artificial intelligence-based method for the full life cycle management of peritoneal dialysis for renal disease, so as to solve the problems of imperfect renal disease management chain, low data quality and lack of personalized management in the prior art.

[0009] The present invention provides a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence, comprising the following steps:

[0010] Step 1: Obtain the patient's life record information, medical record information and home medical device terminal data, wherein the home medical device terminal data includes data from a peritoneal dialysis machine;

[0011] Step 2: Verify, check and spot check the data;

[0012] Among them, data verification includes presetting the threshold range of each data indicator. If the corresponding data does not meet the threshold range, it is regarded as abnormal data and transferred to data verification. Otherwise, it is regarded as normal data;

[0013] Data verification includes sending the current abnormal data to the service provider and receiving the verification results. If the verification passes, it will be converted to normal data. Otherwise, it will be abnormal data and return to data verification;

[0014] Data spot checks include periodically sampling the collected data and comparing it with the original data to check its accuracy;

[0015] Step 3: The artificial intelligence model generates daily guidance, diagnosis and treatment evaluation, and special medical food recommendation information based on the normal data provided in step 2;

[0016] Among them, the recommended information on daily guidance and special medical food is sent directly to the patient for examination; the recommended information on diagnosis and treatment evaluation is first sent to the service provider, and the diagnosis and treatment evaluation information is then sent to the patient after manual correction, supplementation and / or confirmation.

[0017] The beneficial effects of the present invention are: (1) A closed-loop system for full life cycle management: Peritoneal dialysis management lacks closed-loop management of the entire process from data collection to analysis, guidance, and dynamic adjustment, which is prone to problems such as data fragmentation and delayed adjustment of treatment plans. The present invention proposes a full life cycle management system, which collects dialysate parameters and patient vital signs in real time through a device, and combines dynamic analysis and personalized feedback in the cloud to form a closed-loop management of patient treatment. The closed-loop system significantly improves the consistency of data and the real-time nature of guidance, avoids the phenomenon of treatment information islands, and enables treatment plans to quickly adapt to changes in patients' conditions. (2) Complications such as infection and peritonitis during peritoneal dialysis are difficult to monitor in a timely manner, and traditional means can only perform reactive processing after the condition worsens. The system of the present invention collects key parameters such as the patient's life records and medical records of the dialysate, combines vital signs data, and uses an algorithm model to assess infection risks in real time and issue early warnings. And real-time analysis of patient dialysate data, vital signs and life records, generate personalized treatment, diet and exercise recommendations, and dynamically adjust according to treatment data, improve the individual adaptability of the treatment plan, optimize patient compliance and treatment effect, and avoid the risk of complications caused by mismatched treatment plans; (3) Multidisciplinary health coach collaboration: The system supports a multidisciplinary health coach collaboration model that combines online and offline, covering multiple roles such as specialists, nurses, nutritionists, and psychological counselors. With the help of intelligent decision-making assistance modules, it improves team collaboration efficiency, realizes hierarchical diagnosis and treatment, and ensures that patients can obtain comprehensive and integrated medical services while keeping costs under control. (4) The present invention adds a kidney disease insurance and special food module, with a built-in kidney disease risk actuarial model, which can provide patients with personalized insurance plans, combine disease management throughout the life cycle, reduce the incidence of endpoint events, reduce insurance compensation risks, and achieve a win-win situation for patients, insurance companies, and medical service providers; at the same time, according to the special nutritional needs of peritoneal dialysis patients, it provides personalized special medical food recommendations and management to help patients better control their diet, improve their condition, and improve their quality of life. (5) Patient intelligent interactive terminal: The system provides a convenient patient intelligent interactive terminal that supports voice interaction, knowledge Q&A, remote dialogue, prescription reception, and data collection, etc., to enhance patient participation and compliance, and promote effective communication between patients and the medical team. The system uses large language model (LLM) and retrieval enhancement (RAG) technology to provide knowledge Q&A services and patient education courses dedicated to kidney disease, helping patients better understand and manage their condition and improve their self-management capabilities.

[0018] Furthermore, in step 3, the diagnosis and treatment evaluation process includes two stages: disease warning and matching diagnosis and treatment recommendation: (1) Disease warning: using an algorithm model based on indicator thresholds, a risk-based warning model, or an algorithm model based on multimodal data to obtain a risk warning level; (2) Matching diagnosis and treatment recommendation: when the disease warning finds a risk, it relies on the diagnosis and treatment recommendation database for similarity matching, uses the matching diagnosis and treatment plan as a recommendation, and finally pushes the risk warning level and the diagnosis and treatment recommendation plan; otherwise, no diagnosis and treatment recommendation plan will be generated, only the disease warning result will be pushed. These models not only consider a single data item, but also combine multiple data items and comprehensive data, thereby improving the accuracy and reliability of disease warning.

[0019] Furthermore, when the algorithm model based on the indicator threshold is used in step 3, the model divides the risk warning level into three levels: low, medium, and high. When multiple warning thresholds are triggered at the same time, the highest level of risk shall prevail; the low-risk warning is pushed through the method of "patient-side system message + server-side system message", the medium-risk warning is pushed through the method of "patient-side system message + server-side system message + patient mobile phone text message", and the high-risk warning is pushed through the method of "patient-side system message + server-side system message + patient mobile phone text message + patient phone reminder" and dials 120 emergency alarm at the same time. This solution subdivides the risk warning level into three levels: low, medium, and high, and takes the highest level of risk as the basis according to the triggered warning threshold, ensuring the timeliness and effectiveness of the warning information. Through different push methods (such as patient-side system message, mobile phone text message, phone reminder, etc.), the system can take corresponding warning measures for different levels of risks to ensure that patients can receive necessary medical attention and intervention in a timely manner.

[0020] Furthermore, when the risk-based early warning model is used in step 3, the model adopts a time series convolution-recurrent neural network model. The input of the model is the patient's basic medical data, behavioral data, and time series data. The output of the model is the probability of adverse events occurring in the future. When the predicted risk probability of adverse events exceeds the set threshold, it is a medium risk. This prediction not only helps to discover potential medical risks in advance, but also provides doctors with a valuable time window to formulate preventive measures.

[0021] Furthermore, when the algorithm model based on multimodal data is used in step 3, the model input is the basic medical data, behavioral data, and image data of the patient during treatment, and the output of the model is the probability of adverse events in the future. When the predicted risk probability of adverse events exceeds the set threshold, it is a medium risk. The fusion analysis of multimodal data not only improves the accuracy of the prediction, but also provides doctors with more comprehensive patient health information.

[0022] Furthermore, the medium-risk warning is pushed via the method of "patient-side system message + server-side system message + patient mobile phone text message".

[0023] Furthermore, in step 3, the process of matching the recommended treatment plans includes comparing the patient's condition data with the cases in the recommended treatment plan database for similarity, selecting the plan with a similarity ≥ 80% as the recommended treatment plan and sending it to the service provider, and then pushing it to the patient after receiving the confirmation information from the service provider; if the similarity is <80%, one or more plans with the highest similarity will be sent to the service provider, and the plan selection results and supplementary and confirmation information on the result plan sent back by the service provider will be accepted, and then pushed to the patient together.

[0024] Furthermore, the special medical food module calculates special foods that meet the needs based on patient data through a special food matching model and pushes the results to the patient. If the patient decides to purchase, the purchase information will be pushed to the special food merchant.

[0025] The present invention also discloses a full life cycle management device for peritoneal dialysis of kidney disease, which is characterized by: comprising a processor and a memory; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the device executes a full life cycle management method for peritoneal dialysis of kidney disease based on artificial intelligence.

[0026] The present invention also discloses a full life cycle management system for peritoneal dialysis for renal disease, including a cloud intelligent platform, a patient end and a server end; bidirectional data transmission can be performed between the cloud intelligent platform and the patient end, between the cloud intelligent platform and the server end, and between the server end and the patient end, wherein:

[0027] (1) The patient side includes a patient intelligent interaction module and a home medical device module; the patient intelligent interaction module is used to collect the patient's life records, medical records and consultation questions in a way that the patient actively interacts with the patient, and the home medical device module is used to automatically collect the device data and usage data of home medical equipment;

[0028] (2) The cloud-based intelligent platform includes a data collection and storage module, a data quality control module, and a decision analysis module. The data collection and storage module performs structured processing and classification on the data transmitted from the patient end; the data quality control module performs data verification, data checking, and data spot checks on the processed data; the decision analysis module generates and pushes daily guidance, diagnosis and treatment evaluation, and special medical food recommendation information based on the data transmitted by the data quality control module through an artificial intelligence model;

[0029] (3) The service end includes an outpatient nursing module and an inpatient medical module, both of which can obtain the final normal data on the cloud intelligent platform. The outpatient nursing module is used for the service team to provide online medical consulting services to the patient side, which includes an online customer service sub-module and an online multidisciplinary health coach module; the online customer service sub-module is used for the customer service team to provide basic consulting services to the patient side through the cloud intelligent platform; the online multidisciplinary health coach module is used for the health coach team to provide manual services to the patient side, and to confirm, correct and / or supplement the information pushed by the cloud intelligent platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a schematic functional block diagram of a full life cycle management system for peritoneal dialysis for renal disease in an embodiment of the present invention;

[0031] Figure 2 This is a flow chart of the data quality control module acquiring and processing data in an embodiment of the present invention;

[0032] Figure 3 It is a schematic diagram of the structure of the time series convolution-recurrent neural network model (TCN-RNN);

[0033] Figure 4 Schematic diagram of the structure of a multimodal convolutional neural network (CNN). DETAILED DESCRIPTION

[0034] This example is a full life cycle management system for peritoneal dialysis based on artificial intelligence. Figure 1 As shown, the system includes a cloud-based intelligent platform, a patient terminal, and a server terminal, and the patient terminal and the server terminal are both connected to the cloud-based intelligent platform. Bidirectional data transmission is possible between the cloud-based intelligent platform and the patient terminal, between the cloud-based intelligent platform and the server terminal, and between the server terminal and the patient terminal.

[0035] 1. Patient side

[0036] Including patient intelligent interaction apps and home medical equipment terminals.

[0037] Home medical equipment terminals may include automatic peritoneal dialysis machines and other home medical equipment (blood pressure monitors, thermometers, pulse oximeters, blood glucose meters, weight scales, body composition analyzers, etc.), all of which are connected to the Internet through 4G, 5G, WiFi, Bluetooth or NB-IoT and other communication protocols, and upload data to the smart interactive terminal APP, and then to the cloud smart platform. Among them, the automatic peritoneal dialysis machine (APD) automatically uploads data such as dialysis treatment plans, ultrafiltration volume, equipment parameters, operating status, fault codes, etc., and other home medical devices automatically upload data such as heart rate, blood pressure, blood oxygen, body temperature, blood sugar, weight, and body composition.

[0038] The smart interactive terminal APP also has dialysis diaries, life records and medical records, which are filled out by the patients themselves (fill in text content) or uploaded (upload methods include taking photos, screenshots, etc.), and then the smart interactive terminal connects to the Internet through communication protocols such as 4G, 5G or WiFi, and uploads the data to the cloud intelligent platform. Among them, the dialysis diary collects symptoms and signs, such as various types of pain, dizziness, palpitations, dyspnea, fever, fatigue, loss of appetite, nausea and vomiting, constipation, skin itching, edema, urine volume and other information. Life records collect information on diet, exercise, sleep, etc. Medical records collect laboratory data such as dialysis adequacy assessment (Kt / V), residual renal function, peritoneal balance test, blood biochemistry, electrolytes, nutritional indicators, complete blood cell analysis, and records of in-hospital diagnosis and treatment, which are uploaded by patients after being tested in medical institutions.

[0039] The above data is divided according to the collection source. When the patient data is aggregated to the patient end, it must be marked to provide a data marking basis for subsequent analysis. The marking standards are as follows:

[0040]

[0041] In addition to data collection, the smart interactive terminal APP also has a communication section connected to the server. Patients can contact online customer service or online health coaches through the smart interactive terminal APP to consult on issues.

[0042] 2. Cloud-based intelligent platform

[0043] It includes data collection and storage, data quality control, and decision analysis modules. The logic of data collection and storage, and data quality control is as follows Figure 2 As shown:

[0044] (1) The data collection and storage module collects data from the patient (including data automatically uploaded by the device and data uploaded by the patient). The data automatically transmitted by the device is transferred to automatic data verification; and the data uploaded by the patient (such as pictures uploaded by taking photos, screenshots, etc.) When the collected data format is jpg, pdf and other graphics, structured extraction is performed using OCR recognition technology. To ensure the OCR extraction effect, manual verification is added when necessary.

[0045] (2) The data quality control module includes three functional modules: data verification, data checking, and data spot check. Its working process is as follows: Figure 1 As shown, the steps are as follows:

[0046] 1) Data Verification

[0047] The data verification module automatically verifies the data. The threshold range of each data indicator is preset in the data verification module, and the following two results appear:

[0048] a. Data that does not meet the threshold range is deemed to have failed verification, and abnormal data is transferred to the data verification module;

[0049] b. If the data meets the threshold range, it is deemed to have passed the verification and the data is stored in the database (i.e. Figure 2 Enter the data into the decision analysis module;

[0050] 2) Data verification

[0051] The data verification module is connected to the HC module on the server side, and the HC module manually verifies the abnormal data (i.e. Figure 2 If the verification result is normal, the data is converted to normal data and stored in the database. Otherwise, the data item with abnormal verification result is retrieved again and returned to the data verification module;

[0052] 3) Data spot check

[0053] The data spot check module regularly samples the incoming data and manually verifies the accuracy of the incoming data based on the original records. If any anomalies are found, they will be corrected, the causes will be found, and the system will be improved.

[0054] (3) Decision Analysis Module

[0055] The decision analysis module performs artificial intelligence decision analysis on the data output by the data quality control module. The specific method is as follows: the artificial intelligence model analyzes and evaluates the patient's condition and predicts the development of the disease, and generates daily guidance, diagnosis and treatment evaluation, and recommended information for special medical foods. Among them, the recommended information on daily guidance, kidney disease insurance and special medical foods with higher fault tolerance is sent directly to the patient side for the patient to view. The diagnosis and treatment evaluation information with lower fault tolerance is pushed to the service side, and the medical service team of the HC module corrects, supplements and confirms the information generated by artificial intelligence before sending it to the patient side. The higher fault tolerance in this embodiment means that the patient can have a certain degree of flexibility and freedom in the process of receiving guidance, and even if they occasionally deviate from the guidance, it will not have a serious impact on the overall health; on the contrary, the lower fault tolerance means that the deviation of the results from the correct situation will have a serious impact on the overall health.

[0056] The decision analysis module includes daily guidance module, diagnosis and treatment evaluation module, special medical food module and kidney disease insurance module. The daily guidance module has functional modules such as instrument operation, diet guidance, exercise guidance, sleep guidance, etc. The system presets relevant graphics, texts, videos and other multimedia courses for patients to view and learn. The diagnosis and treatment evaluation module has a disease warning module, a diagnosis and treatment plan module (including a diagnosis and treatment recommendation plan database) and a message push function module. The details are as follows:

[0057] a. Disease warning module

[0058] The disease warning module includes three algorithm models: an algorithm model based on indicator thresholds, a risk-based warning model, and an algorithm model based on multimodal data to obtain the disease warning risk level.

[0059] The disease warning module of this embodiment adopts an algorithm model based on indicator thresholds. The warning module is preset with warning threshold data, which is generally a specific warning threshold range and a corresponding warning risk level. Some indicators are specific risk characteristics. The specific warning trigger mode is:

[0060] When the value of a specific indicator falls into a specific warning threshold range or corresponds to a specific risk feature, the warning module will evaluate the indicator as the corresponding warning risk level, and will automatically match the diagnosis and treatment recommendation plan later. In this embodiment, the risk warning levels are divided into three levels: low, medium, and high. When multiple warning thresholds are triggered at the same time, the highest level of risk shall prevail; the low-risk warning is pushed through the "patient-side system message + server-side system message", the medium-risk warning is pushed through the "patient-side system message + server-side system message + patient mobile phone text message", and the high-risk warning is pushed through the "patient-side system message + server-side system message + patient mobile phone text message + patient phone reminder + 120 emergency phone alarm".

[0061] The corresponding risk levels of the core indicators of the algorithm model based on indicator thresholds are as follows:

[0062]

[0063]

[0064]

[0065] b. Diagnosis and treatment plan module

[0066] The diagnosis and treatment plan module includes a built-in diagnosis and treatment recommendation database. When the disease warning module assesses that there is a risk (regardless of the risk level), the diagnosis and treatment plan module starts matching the diagnosis and treatment recommendation. The process is as follows:

[0067] 1) Compare the patient's condition data with the cases in the diagnosis and treatment recommendation database for similarity, select the scheme with similarity ≥ 80% as the diagnosis and treatment recommendation scheme and send it to the HC module (the HC module is the service provider in this process). Since the HC module performs manual confirmation, the diagnosis and treatment scheme confirmed by the HC module is reasonable;

[0068] 2) If the similarity is less than 80%, one or more solutions with the highest similarity will be sent to the HC module, which will manually select, supplement or even replace the recommended solutions.

[0069] The above similarity comparison method can use cosine similarity:

[0070] First, the patient's basic medical data and warning information are converted into a numerical vector, represented by A; vector A and vector B pre-stored in the diagnosis and treatment recommendation database are subjected to cosine similarity calculations one by one:

[0071]

[0072] Among them, ‖A‖ represents the modulus of vector A, and A·B represents the inner product of vectors A and B.

[0073] The cosine similarity is between -1 and 1. The larger the value, the more similar the two are.

[0074] C. Message push function module

[0075] When risk warnings and treatment recommendations are generated, the cloud-based intelligent platform selects different push methods based on the warning risk level:

[0076] Low risk level: Low risk warning risks and corresponding diagnosis and treatment recommendations are pushed to patients through the "patient-side system message + server-side system message" method;

[0077] Low risk level: Medium risk warning and corresponding diagnosis and treatment recommendation plan are pushed through the method of "patient-side system message + server-side system message + patient mobile phone text message";

[0078] High-risk level: High-risk warnings and corresponding recommended treatment plans are pushed through the "patient-side system message + server-side system message + patient mobile phone text message + patient phone reminder" method, and the server-side customer service calls 120 for emergency assistance.

[0079] In addition, the special medical food module in the decision analysis module calculates special foods that meet the needs based on patient data through the special food matching model, and pushes the results to the patient. If the patient decides to purchase, the purchase information will be pushed to the special food merchant. The kidney disease insurance module calculates the insurance payment risk through the actuarial model based on the calculation results of the disease assessment function module, and pushes the risk control prediction results to the insurance company, which then determines whether it meets the insurance requirements. If it does, the corresponding insurance product will be recommended to the patient.

[0080] 3. Server

[0081] The service end includes in-hospital medical module and out-of-hospital nursing module.

[0082] The in-hospital medical module and the patient's hospital information system, after obtaining the relevant data from the diagnosis and treatment evaluation module, perform the following processing: send the patient's condition information to the attending physician, who can make judgments based on experience and inform the patient of precautions or whether he needs to go to the hospital for treatment through the patient end.

[0083] The outpatient care module includes an online customer service module and an online multidisciplinary health coach module. The online customer service module can answer patients' questions online or by phone based on the information transmitted by the cloud-based intelligent platform, and call 120 when necessary. The HC module can provide medical auxiliary consulting services to patients online or by phone based on the information transmitted by the cloud-based intelligent platform or patient feedback receiving module, such as dialysis guidance, dietary guidance, and treatment plan recommendations.

[0084] The definition of the online multidisciplinary health coach module (HC module for short) is: a medical service team composed of health professionals from different fields, including specialists, nurses, clinical pharmacists, nutritionists, sports rehabilitation therapists, psychological counselors, etc. Through remote video, telephone and mobile APP, etc., we communicate with patients, formulate personalized health plans according to the health status, goals and needs of patients, covering diet, exercise, mental health and chronic disease management, and provide comprehensive health management services.

[0085] Example 2

[0086] like Figure 3 As shown, the difference between this embodiment 2 and embodiment 1 is that the disease warning module adopts a risk-based warning model, specifically including a time series convolution-recurrent neural network model (TCN-RNN). The input of the model is the patient's basic medical data, behavioral data and time series data, which have been marked in the aforementioned data collection stage. The output of the model is the probability of a specific adverse event occurring in a future period of time, such as hypoglycemia risk, hyperkalemia risk, etc.

[0087] When the predicted risk probability of an adverse event exceeds the set threshold (for example, 0.5), the system will trigger an early warning. This type of early warning is fixedly configured as a medium-risk early warning. The next step is for the diagnosis and treatment plan module to generate a medical recommendation plan. Otherwise, if it does not exceed the set threshold, it is configured as risk-free.

[0088] The structure of the time series convolution-recurrent neural network model (TCN-RNN) is as follows Figure 3 As shown, the details are as follows:

[0089] (1) The input of the model includes: basic medical data, behavioral data and time series data: Time series data is a sequence of indicators to be predicted sorted by time. Each item in the sequence is a tuple (x, t), where x is the value of the indicator and t is the inspection time of the indicator. Basic medical data and behavioral data can be used as input of the model directly or after encoding conversion.

[0090] (2) Time series data is input to the time series convolution (TCN) layer: Time series convolution ensures that each time step of the output sequence depends only on the current and previous time steps, and not on future time steps. This is achieved by properly padding the input with zero values ​​in the convolution operation. The formula is expressed as:

[0091]

[0092] Among them, y(t) is the output sequence, w(k) is the convolution kernel, K is the size of the convolution kernel, It is a sequence consisting of K-1 zeros padded in front of the input sequence x(t). The time series convolution layer can be 1 or more layers.

[0093] (3) The activation layer generally uses activation functions commonly used in neural networks, such as:

[0094] The ReLU activation function is expressed as:

[0095]

[0096] Hyperbolic tangent (Tanh) activation function, the formula is expressed as:

[0097]

[0098] The Sigmoid activation function is expressed as:

[0099]

[0100] (4) The output of the time series convolution layer is transformed by the activation function and then input into the recurrent neural network (RNN) layer. The recurrent neural network has an internal loop structure that can retain information at each time step of the sequence, thereby capturing the temporal dependency and sequential relationship in the data.

[0101] The recurrent neural network layer can use the gated recurrent unit (GRU), which is expressed as:

[0102] Update Gate: Z t =Sigmoid(W Z ·[h t-1 ,x t ]+b z )

[0103] Reset gate: r t =Sigmoid(W r ·[h t-1 ,x t ]+b r )

[0104] Candidate hidden states:

[0105] Hide status updates:

[0106] Where * represents element-wise multiplication, [] represents the concatenation of vectors, and W and b are weights and biases.

[0107] The number of recurrent network layers can be one or more.

[0108] (5) The output of the recurrent neural network layer is further input into a fully connected layer, and the formula is expressed as:

[0109] y=W·x+b

[0110] Among them, W and b are weights and biases, and the output of the fully connected layer is obtained through the activation function.

[0111] (6) The patient's basic medical data and behavioral data are sequentially passed through the fully connected layer and the activation layer to obtain an output, and the output is concatenated with the output obtained from the time series. The concatenated vector is then sequentially passed through the fully connected layer and the activation layer, and then transformed into a probability value between (0, 1) through the Sigmoid layer to obtain the final probability output.

[0112] (7) During the training phase, the model is trained using stochastic gradient descent (SGD) and back propagation. Dropout layers and regularization can be added to the model to prevent overfitting.

[0113] Example 3

[0114] like Figure 4 As shown, the difference between this embodiment 3 and embodiment 1 is that the disease warning module adopts an algorithm model based on multimodal data, specifically a multimodal convolutional neural network model (CNN), which collects basic medical data, behavioral data and image data during the patient's treatment. The probability of occurrence of adverse events, such as peritonitis, peritoneal dialysis catheter outlet infection, etc., is calculated through the algorithm model. When the predicted risk probability of adverse events exceeds the set threshold (such as 0.5), the system will trigger an early warning. This type of early warning is fixedly configured as a medium-risk early warning. The next step is for the diagnosis and treatment plan module to generate a medical recommendation plan. Otherwise, if it does not exceed the set threshold, it is configured as risk-free.

[0115] Convolutional neural networks can be connected using residual networks (ResNet), and the formula is:

[0116] H(x)=F(x)+x

[0117] The network structure is as follows Figure 4 As shown, the details are as follows:

[0118] (1) The input of the model is: basic medical data, behavioral data and image data.

[0119] The input image is a picture taken of a patient (such as a peritoneal dialysis catheter outlet picture). First, preprocessing is performed to adjust the resolution to a fixed size and standardize and normalize the pixel values.

[0120] Basic medical data and behavioral data can be used as input to the model directly or after encoding conversion.

[0121] (2) The preprocessed image is processed by a 7×7 convolutional layer, a batch normalization layer (BN), and a 3×3 max pooling layer.

[0122] (3) The output of the 3×3 max pooling layer enters 8 residual connection blocks in series. According to the data flow, each residual connection block consists of a 3×3 convolutional layer, a batch normalization layer, a ReLU activation layer, a 3×3 convolutional layer, a batch normalization layer, and a shortcut connection with a 1×1 convolutional layer. The shortcut connection and the trunk output are added and then passed through the ReLU activation layer;

[0123] (4) The output of the last residual block enters the global average pooling layer, and then enters the fully connected layer and the ReLU activation layer in sequence to obtain the feature output of the image.

[0124] (5) The patient’s basic medical data and behavioral data are first input into the fully connected layer, and then concatenated with the feature output of the image after passing through the ReLU activation layer to obtain the fused feature output.

[0125] (6) The fused feature output is input into the fully connected layer again, and finally the probability value of the event is output through the Softmax layer.

[0126] The Softmax formula is expressed as follows:

[0127]

[0128] Where e is the base of natural logarithm, x i is the i-th element in the input vector, and K is the dimension of the input vector, i.e., the number of categories. i is the i-th element in the output vector, indicating the probability that the input belongs to the i-th class.

[0129] (5) During the training phase, the model is trained using stochastic gradient descent (SGD) and back propagation.

[0130] Recommended diagnosis and treatment plan generation function module

[0131] When the patient's indicators trigger an early warning, the recommended treatment plan generation function module compares the patient's basic medical data and early warning information with the cases in the treatment recommendation plan database for similarity, and selects plans with a similarity ≥ 80% as treatment recommendations for confirmation by the manual team on the server; if the similarity is <80%, that is, the similarity with all cases is <80%, the health coach on the server is required to select, supplement and confirm.

[0132] The similarity comparison method can use cosine similarity:

[0133] First, the patient's basic medical data and warning information are converted into a numerical vector, represented by A; vector A and vector B pre-stored in the diagnosis and treatment recommendation database are subjected to cosine similarity calculations one by one:

[0134]

[0135] Among them, ‖A‖ represents the modulus of vector A, and A·B represents the inner product of vectors A and B.

[0136] The cosine similarity is between -1 and 1. The larger the value, the more similar the two are.

[0137] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by the present invention shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence, characterized in that The steps include: Step 1: Obtain the patient's life record information, medical record information and home medical device terminal data, wherein the home medical device terminal data includes data from a peritoneal dialysis machine; Step 2: Verify, check and spot check the data; Among them, data verification includes presetting the threshold range of each data indicator. If the corresponding data does not meet the threshold range, it is regarded as abnormal data and transferred to data verification. Otherwise, it is regarded as normal data; Data verification includes sending the current abnormal data to the service provider and receiving the verification results. If the verification passes, it will be converted to normal data. Otherwise, it will be abnormal data and return to data verification; Data spot checks include periodically sampling the collected data and comparing it with the original data to check its accuracy; Step 3: The artificial intelligence model generates daily guidance, diagnosis and treatment evaluation, and special medical food recommendation information based on the normal data provided in step 2; Among them, the recommended information on daily guidance and special medical food is sent directly to the patient for examination; the recommended information on diagnosis and treatment evaluation is first sent to the service provider, and the diagnosis and treatment evaluation information is then sent to the patient after manual correction, supplementation and / or confirmation.

2. According to claim 1, a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence, characterized in that: In step 3, the diagnosis and treatment evaluation process includes two stages: disease warning and matching diagnosis and treatment recommendation plan; (1) Disease warning: using an algorithm model based on indicator thresholds, a risk-based warning model, or an algorithm model based on multimodal data to obtain the risk warning level; (2) Matching treatment recommendations: When a risk is detected in the disease warning, similarity matching is performed based on the treatment recommendation database, and the matching treatment plan is used as the recommended plan. Finally, the risk warning level and treatment recommendation plan are pushed; otherwise, no treatment recommendation plan will be generated, and only the disease warning result will be pushed.

3. According to claim 2, a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence, characterized in that: When the algorithm model based on indicator threshold is used in step 3, the model divides the risk warning level into three levels: low, medium and high. When multiple warning thresholds are triggered at the same time, the highest level of risk shall prevail; The low-risk warning is pushed via the "patient-side system message + server-side system message" method, the medium-risk warning is pushed via the "patient-side system message + server-side system message + patient mobile phone text message" method, and the high-risk warning is pushed via the "patient-side system message + server-side system message + patient mobile phone text message + patient phone reminder" and simultaneously dials 120 emergency alarm.

4. According to claim 2, a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence is characterized by: When the risk-based early warning model is used in step 3, the model adopts a time series convolution-recurrent neural network model. The input of the model is the patient's basic medical data, behavioral data, and time series data. The output of the model is the probability of adverse events occurring in the future. When the predicted adverse event risk probability exceeds the set threshold, it is a medium risk.

5. According to claim 3, a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence is characterized by: When the algorithm model based on multimodal data is used in step 3, the model input is the basic medical data, behavioral data, and image data of the patient during treatment. The output of the model is the probability of adverse events occurring in the future. When the predicted adverse event risk probability exceeds the set threshold, it is medium risk.

6. A method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence according to claim 4 or 5, characterized in that: The medium-risk warning is pushed via the method of "patient-side system message + server-side system message + patient mobile phone text message".

7. A method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence according to claim 2, 3, 4 or 5, characterized in that: In step 3, the process of matching the diagnosis and treatment recommendation plan includes comparing the patient's condition data with the cases in the diagnosis and treatment recommendation plan database for similarity, selecting the plan with a similarity of ≥ 80% as the diagnosis and treatment recommendation plan and sending it to the delivery service provider, and then pushing the patient after receiving the confirmation information from the delivery service provider; If the similarity is less than 80%, one or more plans with the highest similarity will be sent to the service provider, and the plan selection result and supplementary and confirmation information on the result plan sent back by the service provider will be accepted and pushed to the patient together.

8. According to claim 7, a method for managing the whole life cycle of peritoneal dialysis for renal disease based on artificial intelligence is characterized by: The special medical food module calculates special foods that meet the needs based on patient data through a special food matching model and pushes the results to the patient. If the patient decides to purchase, the purchase information will be pushed to the special food merchant.

9. An artificial intelligence-based full life cycle management device for peritoneal dialysis for renal disease, characterized in that: It comprises a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the device executes the artificial intelligence-based full life cycle management method for renal peritoneal dialysis as described in any one of claims 1 to 8.

10. An artificial intelligence-based management system for the entire life cycle of peritoneal dialysis for renal disease, characterized in that: It includes the cloud intelligent platform, the patient end and the server end. Bidirectional data transmission is possible between the cloud intelligent platform and the patient end, between the cloud intelligent platform and the server end, and between the server end and the patient end. (1) The patient side includes a patient intelligent interaction module and a home medical device module; the patient intelligent interaction module is used to collect the patient's life records, medical records and consultation questions in a way that the patient actively interacts with the patient, and the home medical device module is used to automatically collect the device data and usage data of home medical equipment; (2) The cloud-based intelligent platform includes a data collection and storage module, a data quality control module, and a decision analysis module. The data collection and storage module performs structured processing and classification on the data transmitted from the patient end; the data quality control module performs data verification, data checking, and data spot checks on the processed data; the decision analysis module generates and pushes daily guidance, diagnosis and treatment evaluation, and special medical food recommendation information based on the data transmitted by the data quality control module through an artificial intelligence model; (3) The service end includes an outpatient nursing module and an inpatient medical module, both of which can obtain the final normal data on the cloud intelligent platform. The outpatient nursing module is used by the service team to provide online medical consultation services to the patient end, which includes an online customer service sub-module and an online multidisciplinary health coach module; The online customer service submodule is used for the customer service team to provide basic consulting services to patients through the cloud intelligent platform; The online multidisciplinary health coach module is used for the health coach team to provide manual services to the patient side, and to confirm, correct and / or supplement the information pushed by the cloud intelligent platform.

Images