Multi-task collaborative prediction medical follow-up visit method and system, terminal and medium

Abstract

The invention belongs to the technical field of medical follow-up visit, and particularly discloses a multi-task collaborative prediction medical follow-up visit method and system, a terminal and a medium, and the method comprises the steps: collecting and preprocessing multi-source medical data, and generating a target data set in a unified format; respectively inputting the target data set into a classification model, a regression model and a time sequence model to obtain a patient risk level, a key monitoring index short-term prediction result and an illness state long-term trend prediction result; according to the prediction result, a personalized follow-up visit plan is generated in combination with the disease feature library and the core monitoring index weight, and the follow-up visit interval and follow-up visit items are dynamically adjusted according to the change of the prediction result in the follow-up visit execution process; when the follow-up plan is executed, a follow-up task is pushed, feedback information is received, and the feedback information is analyzed and stored in the medical information system for subsequent dynamic adjustment and model updating. Multi-source data fusion of medical follow-up visit, multi-task collaborative prediction and dynamic optimization of follow-up visit plans are realized, and personalization and timeliness of follow-up visit are improved.

Description

Technical Field

[0001] This invention belongs to the field of medical follow-up technology, specifically relating to a multi-task collaborative prediction method, system, terminal, and medium for medical follow-up. Background Technology

[0002] With the development of medical informatization and intelligentization, patient follow-up work is gradually transforming from traditional manual methods to digital and networked methods. In the existing follow-up methods, medical institutions usually conduct health status follow-ups with patients through manual telephone, paper records, or single information systems. This method not only relies on the experience and memory of medical staff, consuming a lot of human resources, but is also prone to information omissions or incomplete records.

[0003] In addition, patients' medical data are usually scattered across multiple sources, such as hospital information systems (HIS), electronic medical record systems (EMR), laboratory imaging systems (LIS / PACS), and patient self-reported information. The formats, structures, and update frequencies of these different data sources vary considerably, and there is a lack of a unified integration and preprocessing mechanism, making it difficult to use this data directly for disease analysis and prediction.

[0004] In terms of disease assessment and follow-up plan development, existing technologies mostly rely on a single model or human experience for judgment, making it difficult to simultaneously take into account risk classification, indicator prediction, and disease trend analysis. The accuracy and stability of the prediction results are low. At the same time, there is a lack of ability to dynamically adjust the follow-up frequency and items based on the prediction results, which cannot respond to changes in the patient's condition in a timely manner, resulting in problems of delayed or excessive follow-up plans. Summary of the Invention

[0005] This invention addresses the problems in the prior art by providing a multi-task collaborative prediction method, system, terminal, and medium for medical follow-up. This solves the problem that most existing technologies rely on a single model or human experience for judgment, making it difficult to simultaneously consider risk classification, indicator prediction, and disease trend analysis, resulting in low accuracy and stability of prediction results. It also solves the problem that existing technologies cannot respond to changes in patients' conditions in a timely manner, leading to delayed or excessive follow-up plans.

[0006] The technical solution adopted in this invention is as follows: Firstly, this application provides a multi-task collaborative prediction method for medical follow-up, which includes the following steps: Step S1: Collect multi-source medical data and preprocess the multi-source medical data to generate a target dataset for algorithm analysis; Step S2: Input the target dataset into the classification model, regression model, and time series model respectively, and calculate the prediction results; the prediction results include risk level, short-term prediction results, and long-term trend prediction results of the disease. Step S3: Based on the analysis results, generate a personalized follow-up plan by combining the preset disease feature library and the weights of the monitoring indicators in the target data. Step S4: Based on the execution process of the personalized follow-up plan, dynamically adjust the follow-up interval and follow-up items.

[0007] Furthermore, in step S1, the patient's basic information, medical records, test results, and imaging reports are obtained by connecting to the hospital information system, electronic medical record system, and laboratory imaging system through interfaces. The system collects information from patients' self-reported symptoms, medication records, and lifestyle habits. The collected multi-source medical data is structured, standardized, and missing values ​​are filled to obtain a target dataset with a unified format.

[0008] Furthermore, in step S2, the target dataset is input into the classification model, which is a random forest classification model. During the construction of each decision tree, multiple features are randomly selected, and nodes are divided based on the Gini coefficient. Multiple decision trees are integrated, and the patient risk level classification result is output using a majority voting mechanism. The hyperparameters of the model are optimized using a 5-fold cross-validation method. The hyperparameters include tree depth and the number of leaf nodes. The target dataset is input into the regression model, which is a gradient boosting regression tree model. The base learner is trained based on features such as historical test indicators, medication dosage and symptom scores. The mean squared error is used as the loss function. The model parameters are updated by fitting negative gradients during the iteration process, and the short-term prediction results of key monitoring indicators are output after the iteration is completed. The target dataset is input into the time series model, which is a long short-term memory network model. The time series data of historical test indicators are processed in chronological order and passed through the input layer, long short-term memory network layer, dropout layer and fully connected layer in sequence to output the long-term trend prediction results of the disease. The Adam optimizer is used to train the model at the set learning rate.

[0009] Furthermore, new follow-up data is acquired within a preset time period, and the classification model, regression model, and time series model are updated and trained based on the new data.

[0010] Furthermore, in step S3, the disease feature database contains core monitoring indicators for multiple diseases, and presets the indicator weights of the core monitoring indicators based on time series. The initial follow-up interval is determined based on the risk level output by the random forest classification model; When the gradient boosting regression tree model predicts abnormal fluctuations in core monitoring indicators, the follow-up interval should be shortened. When the long short-term memory network model predicts a stable long-term trend, the follow-up interval should be extended.

[0011] Furthermore, step S4 includes the following steps: Step S4-1: Determine the basic follow-up interval based on the risk level; Step S4-2: Combine the core monitoring indicators and their weights for the corresponding diseases in the disease feature database to generate a personalized follow-up project combination; Step S4-3: Dynamically adjust the personalized follow-up plan based on changes in the prediction results.

[0012] Furthermore, when implementing a personalized follow-up plan, follow-up tasks are pushed to the patient and patient feedback information is received. The patient feedback information is parsed into structured data, stored in the medical information system, and used for subsequent dynamic adjustments and model updates.

[0013] Secondly, this application provides a multi-task collaborative prediction medical follow-up system for implementing the multi-task collaborative prediction medical follow-up method as described in the first aspect. The system includes: The multi-source data fusion module is used to connect to hospital information systems, electronic medical record systems, and laboratory imaging systems through interfaces to collect basic patient information, medical records, laboratory indicators, and imaging reports. It also receives symptom self-assessments, medication records, and lifestyle information proactively reported by patients. The module performs structured processing, standardization processing, and missing value imputation on the collected multi-source medical data to generate a target dataset in a unified format. The multi-task algorithm collaboration module is used to input the target dataset into the random forest classification model, the gradient boosting regression tree model, and the long short-term memory network model respectively, calculate the patient risk level, short-term prediction results of key monitoring indicators, and long-term trend prediction results of the disease, and dynamically update and train each model based on the newly added follow-up data within the preset time period. The follow-up plan generation module automatically generates and dynamically adjusts personalized follow-up plans, including follow-up intervals and follow-up item combinations, based on the risk level output by the classification model, the abnormal fluctuation judgment of the regression model, and the long-term trend analysis of the time series model, combined with the built-in disease feature library and the weights of core monitoring indicators. The follow-up execution and feedback module is used to push follow-up tasks to patients, receive and parse patient feedback information, store it in the medical information system, and use the feedback data for dynamic adjustment of the follow-up plan and continuous optimization of the model. The physician interaction module provides a visual interface for physicians to display algorithm prediction results and follow-up execution status. It supports physicians in manually adjusting and temporarily intervening in the follow-up plan, and the adjustment and intervention records are linked to the subsequent model optimization process.

[0014] Thirdly, this application provides a terminal, including: Memory, used to store medical follow-up simulation programs for multi-task collaborative prediction; A processor, used to implement the steps of the multi-task collaborative prediction medical follow-up method as described in the first aspect when executing the multi-task collaborative prediction medical follow-up device.

[0015] Fourthly, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the multi-task collaborative prediction medical follow-up method as described in the first aspect.

[0016] As can be seen from the above technical solutions, the advantages of the present invention are: By integrating multi-source data acquisition and preprocessing, multi-task collaborative prediction, personalized follow-up plan generation, and dynamic adjustment within a unified technical framework, this method effectively addresses issues in existing technologies such as fragmented follow-up information sources, inability to directly analyze data, single predictive models, and lack of dynamism in follow-up plans. This method can simultaneously collect and integrate multiple types of medical data from hospital information systems, electronic medical record systems, laboratory imaging systems, and patient data. Through structuring, standardization, and missing value processing, it generates a unified target dataset, providing high-quality input data for subsequent model analysis. By inputting the target dataset into classification, regression, and time series models respectively, short-term predictions of patient risk levels and key indicators, as well as long-term trend predictions of disease progression, can be obtained. This allows for a comprehensive consideration of the patient's immediate condition and development trend in follow-up plan development, improving the scientific rigor and relevance of the plan.

[0017] During the follow-up plan generation process, by combining the disease characteristic database and the weights of core monitoring indicators, the basic follow-up interval can be automatically determined based on the prediction results. The follow-up interval can be shortened when key indicators show abnormal fluctuations and extended when the disease trend is stable, thereby achieving real-time response to changes in the patient's condition and reducing delayed or excessive follow-up. By collecting patient feedback information during the follow-up execution phase and converting it into structured data, not only can the integrity of the follow-up records be ensured, but this data can also be used for subsequent dynamic adjustments and model updates, forming a data-driven closed-loop optimization mechanism.

[0018] This invention can significantly reduce the workload of manual follow-up, improve data utilization efficiency and prediction accuracy, enhance the matching degree between follow-up plans and patients' actual conditions, thereby improving the personalization, timeliness and scientific nature of medical follow-up, improving the medical service experience of patients in the recovery process, and providing medical institutions with more accurate follow-up decision support. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the steps of the multi-task collaborative prediction medical follow-up method in the embodiment. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, this invention provides a multi-task collaborative prediction method for medical follow-up, comprising the following steps: Step S1: Collect multi-source medical data and preprocess the multi-source medical data to generate a target dataset for algorithm analysis; By connecting to the hospital information system, electronic medical record system, and laboratory imaging system, we can obtain patients' basic information, medical records, laboratory indicators, and imaging reports. The system collects information from patients' self-reported symptoms, medication records, and lifestyle habits. The collected multi-source medical data is structured, standardized, and missing values ​​are filled to obtain a target dataset with a unified format.

[0023] Structured processing: BERT model is used to perform entity recognition on unstructured data (such as medical record text), extracting key information such as diagnostic terms, complications, and surgical procedures, and converting them into structured labels; Standardization process: Map test indicators (such as blood glucose and blood pressure) to clinical reference ranges to generate "normal / mildly abnormal / severely abnormal" level labels; Missing value handling: Missing data is filled using KNN interpolation. in, For missing values, For the k nearest neighbor sample values, The weight is inversely proportional to the distance.

[0024] Step S2: Input the target dataset into the classification model, regression model, and time series model respectively, and calculate the prediction results; the prediction results include risk level, short-term prediction results, and long-term trend prediction results of the disease. The target dataset is input into the classification model, which is a random forest classification model, used for classifying patient condition status (e.g., "high risk / medium risk / low risk"). Multiple features are randomly selected during the construction of each decision tree, and nodes are divided based on the Gini coefficient. Multiple decision trees are integrated and the patient risk level classification result is output using a majority voting mechanism. The hyperparameters of the model are optimized using a 5-fold cross-validation method, including tree depth and the number of leaf nodes. Input features: structured features such as patient age, disease type, laboratory test results, and surgical history; Constructing a decision tree: Randomly select m features ( ), Nodes are partitioned based on Gini coefficients, and the formula for Gini coefficients is:

[0025] in The proportion of samples in the k-th class; Integrate 100 decision trees and use a majority voting mechanism to output the classification results; Five-fold cross-validation is used to optimize hyperparameters (tree depth, number of leaf nodes).

[0026] The target dataset is input into the regression model, which is a gradient boosting regression tree model, used to predict the numerical changes of specific indicators (such as glycated hemoglobin values ​​within 3 months). The base learner is trained based on historical test indicators, medication dosage and symptom scores, etc. The mean squared error is used as the loss function. The model parameters are updated by fitting negative gradients during the iteration process, and the short-term prediction results of key monitoring indicators are output after the iteration ends. Input features: historical test indicators, medication dosage, symptom scores, etc. Initialize weak learner Where L is the loss function (using mean squared error) ); Iterative generation of base learners: Calculate the negative gradient ; Fitting basis learner to ; Update model = (η is the learning rate, set to 0.1); The regression model is output after 100 iterations.

[0027] The target dataset is input into the time series model, which is a long short-term memory network model. The time series data of historical test indicators are processed in chronological order and passed through the input layer, long short-term memory network layer, dropout layer and fully connected layer in sequence to output the long-term trend prediction results of the disease. The Adam optimizer is used to train the model at the set learning rate.

[0028] Input: Time series data of the patient's historical test results (sorted by timestamp); Network structure: Input layer (feature dimension d) → LSTM layer (64 neurons) → Dropout layer (0.2) → Fully connected layer (output dimension 1); Core formula (LSTM gating mechanism): Forgotten Gate:

[0029] Input Gate: ,

[0030] Cell state:

[0031] Output gate: , Where σ is the sigmoid function, W is the weight matrix, and b is the bias term; The optimizer used was Adam with a learning rate of 0.001 and 50 iterations. The network parameters were adjusted using the validation set loss.

[0032] Acquire new follow-up data within a preset time period, and update and train the classification model, regression model, and time series model based on the new data.

[0033] The model is automatically updated every 3 months. The newly collected follow-up data is divided into training and validation sets in a 7:3 ratio, and the old model is retrained and replaced to ensure predictive adaptability.

[0034] Step S3: Based on the analysis results, generate a personalized follow-up plan by combining the preset disease feature library and the weights of the monitoring indicators in the target data. The disease feature database contains core monitoring indicators for multiple diseases and presets the indicator weights of the core monitoring indicators based on time series. Disease feature database: It contains core monitoring indicators for 12 types of diseases, including tumors, chronic diseases, and postoperative diseases (such as tumor markers such as CEA and CA125 for tumor patients, and fasting blood glucose and glycated hemoglobin for diabetic patients), and preset indicator weights (such as the weight of surgical incision recovery within 3 months after surgery is 0.7).

[0035] The initial follow-up interval is determined based on the risk level output by the random forest classification model; When the gradient boosting regression tree model predicts abnormal fluctuations in core monitoring indicators, the follow-up interval should be shortened. When the long short-term memory network model predicts a stable long-term trend, the follow-up interval should be extended.

[0036] Basic frequency: The initial follow-up interval is determined based on the risk level output by the random forest (high risk: 3 days / time, medium risk: 1 week / time, low risk: 1 month / time). Dynamic adjustment: If GBRT predicts abnormal fluctuations in a certain indicator (such as blood pressure increase >20%), the interval will be automatically shortened by 50%; if LSTM predicts a stable long-term trend, the interval will be extended by 30%. Physician intervention mechanism: Physicians can view the algorithm output results through a visual interface, manually adjust follow-up items (such as adding "electrocardiogram examination") or trigger temporary follow-up, and record the reasons for the adjustment (the system automatically links to subsequent model optimization).

[0037] Individual analysis: Generates patient follow-up reports, including indicator change curves, risk warnings (such as "based on LSTM prediction, blood glucose may exceed 7.0 mmol / L within 6 months"), and intervention recommendations (such as "it is recommended to increase exercise frequency").

[0038] Group Analysis: Disease feature mining: The K-means clustering algorithm is used to group patients with the same disease and output the "high recurrence risk subgroup features" (such as age > 60 years old with hypertension). Comparison of treatment effects: The differences in follow-up indicators among different treatment regimens were analyzed using a t-test. Formula:

[0039] in The mean, Let be the variance and n be the sample size.

[0040] Scientific data output: The data is anonymized (identifying information such as names and ID numbers is removed) to generate a dataset that conforms to the CDISC standard and supports export in CSV / JSON format.

[0041] Step S4: Based on the execution process of the personalized follow-up plan, dynamically adjust the follow-up interval and follow-up items; Step S4-1: Determine the basic follow-up interval based on the risk level; Step S4-2: Combine the core monitoring indicators and their weights for the corresponding diseases in the disease feature database to generate a personalized follow-up project combination; Step S4-3: Dynamically adjust the personalized follow-up plan based on changes in the prediction results.

[0042] When implementing a personalized follow-up plan, follow-up tasks are pushed to the patient and patient feedback information is received. The patient feedback information is parsed into structured data, stored in the medical information system, and used for subsequent dynamic adjustments and model updates.

[0043] Patients can receive follow-up reminders via a mini-program (in both text and voice formats), upload photos of their symptoms (such as wound healing status) or voice descriptions (such as "I feel dizzy today"), and the system will automatically analyze and associate them with the corresponding indicators.

[0044] Healthcare professionals: The system displays a real-time "List of High-Risk Patients" (sorted by risk value), marks patients who have not completed follow-up visits, and provides functions such as "One-Click Follow-up Call" and "Automatic Generation of Follow-up Record Templates." Recorded content is automatically synchronized to the EMR system.

[0045] In some embodiments, this application provides a multi-task collaborative prediction medical follow-up system. The system includes a multi-source data fusion module, a multi-task algorithm collaboration module, a follow-up plan generation module, a follow-up execution and feedback module, and a physician interaction module. The functions and working processes of each module are as follows: The multi-source data fusion module connects to hospital information systems, electronic medical record systems, and laboratory imaging systems via standardized data interfaces to collect patients' basic information, medical records, laboratory indicators, and imaging reports. Simultaneously, this module also receives proactively reported data from patients, including symptom self-assessment information, medication records, and lifestyle information. After data collection, the module sequentially performs structuring, standardization, and missing value imputation operations on the multi-source medical data. It transforms unstructured text information into analyzable structured labels, standardizes and maps laboratory indicator values ​​to generate grade labels, and imputes missing values ​​using interpolation methods, thereby generating a unified format target dataset that provides high-quality input data for subsequent model analysis.

[0046] The multi-task algorithm collaboration module is used to input the target dataset into the random forest classification model, the gradient boosting regression tree model, and the long short-term memory network model, respectively. Specifically, the random forest classification model outputs the patient's risk level; the gradient boosting regression tree model generates short-term predictions of key monitoring indicators; and the long short-term memory network model outputs long-term trend predictions of the patient's condition. Furthermore, this module acquires new follow-up data within a preset time period and combines the new data with historical data to dynamically update and train the random forest classification model, the gradient boosting regression tree model, and the long short-term memory network model, maintaining the adaptability and accuracy of the prediction results.

[0047] The follow-up plan generation module automatically generates personalized follow-up plans based on the risk level output by the random forest classification model, the judgment results of abnormal fluctuations in key monitoring indicators by the gradient boosting regression tree model, and the long-term trend analysis results of the long short-term memory network model, combined with the built-in disease feature library and the weights of core monitoring indicators. The personalized follow-up plan includes the follow-up interval and the combination of follow-up items, and the follow-up interval and the combination of items are dynamically adjusted according to changes in the prediction results during subsequent operation.

[0048] The follow-up execution and feedback module is used to push follow-up tasks to patients according to the personalized follow-up plan. Push notifications can take the form of text, voice, or multimedia reminders. After receiving feedback information from the patient, the module parses the feedback information into structured data and stores it in the medical information system. Simultaneously, the structured feedback data is used for the dynamic adjustment of the personalized follow-up plan and the continuous optimization of the random forest classification model, gradient boosting regression tree model, and long short-term memory network model.

[0049] The physician interaction module provides a visual interface for physicians to view the prediction results of the multi-task algorithm collaboration module and the execution status of the follow-up execution and feedback module in real time. This module allows physicians to manually adjust the follow-up plan or add temporary interventions based on clinical judgment. Adjustments and intervention records are automatically linked to subsequent model optimization processes to continuously improve the accuracy of predictions and follow-ups.

[0050] Taking "postoperative follow-up of lung cancer patients" as an example, the system operation process is as follows: Data collection: Integrate patient HIS information (age 65 years, smoking history 30 years), EMR surgical record (lobectomy), LIS report (postoperative CEA value 5.2 ng / mL) and patient-reported data (daily cough count); Pretreatment: The tags "lobectomy" and "smoking history" were extracted, and the CEA value was mapped to "normal" (reference range 0-5 ng / mL, 5.2 is mildly abnormal); Algorithm operation: Random forest classifies it as "medium risk" (feature weights: smoking history 0.3, CEA value 0.25); GBRT predicts a CEA value of 5.8 ng / mL after one month; LSTM predicts that the CEA value will show a slow upward trend over the next 6 months; Plan generation: The initial follow-up interval is 1 week, which is adjusted to 5 days / time due to the predicted increase in CEA, including "CEA test + chest CT". Execution and Feedback: Patients upload voice recordings of their worsening cough via the mini-program. The system analyzes the recordings and pushes them to the doctor's end. The doctor then triggers a temporary follow-up and adjusts the next examination to include "inflammatory marker detection". Scientific analysis: Clustering data from a total of 100 postoperative patients showed that the recurrence risk of the subgroup with "smoking history > 20 years + CEA postoperative level > 5 ng / mL" was 2.3 times that of other groups, providing a basis for clinical research.

[0051] In some embodiments, this application provides a terminal, including: Memory for storing medical follow-up programs that enable multi-task collaborative prediction; A processor, used to implement the steps of the multi-task collaborative prediction medical follow-up method when executing the multi-task collaborative prediction medical follow-up system.

[0052] In some embodiments, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the multi-task collaborative prediction medical follow-up method.

[0053] It is understood that the systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can be a personal computer, a laptop computer, a personal digital assistant, a tablet computer, a wearable device, or any combination of these devices.

[0054] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0055] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0056] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0057] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0058] It should be understood that although the terms first, second, third, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of one or more embodiments of this specification, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "in response to a determination," or "when," or "in the event of a determination."

[0059] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of one or more embodiments of this specification.

Claims

1. A method of multi-task synergistic prediction of medical follow-up visits, characterized in that, The method comprises the following steps: Step S1, collecting multi-source medical data, and preprocessing the multi-source medical data to generate a target data set for algorithm analysis; Step S2, inputting the target data set into a classification model, a regression model and a time series model respectively, and calculating a prediction result; the prediction result includes a risk level, a short-term prediction result and a long-term trend prediction result of a disease condition; Step S3, generating a personalized follow-up plan according to the analysis result, combining a preset disease characteristic library and the weight of the monitoring indicators in the target data set; Step S4, dynamically adjusting the follow-up interval and follow-up project based on the execution process of the personalized follow-up plan.

2. The method of multi-task synergistic prediction of medical follow-up visits according to claim 1, characterized in that, In step S1, the interface is connected to the hospital information system, the electronic medical record system and the inspection image system to obtain the basic information, diagnosis and treatment records, inspection indicators and image reports of the patient; Through the patient end, the self-evaluation of symptoms, medication records and lifestyle information actively reported are collected; The collected multi-source medical data are subjected to structured processing, standardization processing and missing value filling to obtain a target data set in a unified format.

3. The method of claim 1, wherein, In step S2, the target data set is input into a classification model, which is a random forest classification model. In the construction process of each decision tree, multiple features are randomly selected, and the nodes are divided based on the Gini coefficient. The classification result of the risk level of the patient is output by integrating multiple decision trees and using the majority voting mechanism. The hyperparameters of the model are optimized by the 5-fold cross-validation method, including tree depth and leaf node number. The target data set is input into a regression model, which is a gradient boosting regression tree model. The base learner is trained based on historical inspection indicators, medication dosage and symptom scores. The mean square error is used as the loss function. The model parameters are updated by fitting the negative gradient in the iteration process, and the short-term prediction result of the key monitoring indicators is output after the iteration is completed. The target data set is input into a time series model, which is a long short-term memory network model. The time series data of the historical inspection indicators are processed in time sequence. The long-term trend prediction result of the disease condition is output by sequentially passing through the input layer, the long short-term memory network layer, the Dropout layer and the full connection layer. The model is trained by using the Adam optimizer at a set learning rate.

4. The method of multi-task synergistic prediction of medical follow-up visits according to claim 3, characterized in that, In a preset time period, new follow-up data are obtained, and the classification model, the regression model and the time series model are updated and trained based on the new data.

5. The method of multi-task synergistic prediction of medical follow-up visits according to claim 3, characterized in that, In step S3, the disease characteristic library contains the core monitoring indicators of multiple diseases, and the weight of the core monitoring indicators is preset based on the time series; The initial follow-up interval is determined according to the risk level output by the random forest classification model; When the gradient boosting regression tree model predicts that there is abnormal fluctuation of the core monitoring indicators, the follow-up interval is shortened; When the long short-term memory network model predicts that the long-term trend is stable, the follow-up interval is extended.

6. The method of multi-task synergistic predictive medical follow-up visit of any one of claims 1-5, characterized in that, Step S4 comprises the following steps: Step S4-1, determining the basic follow-up interval according to the risk level; Step S4-2, generating a personalized follow-up project combination by combining the core monitoring indicators and their weights of the corresponding disease in the disease characteristic library; Step S4-3, dynamically adjusting the personalized follow-up plan according to the changes in the prediction result.

7. The method of claim 1, wherein, When the personalized follow-up plan is executed, the follow-up task is pushed to the patient end and the patient feedback information is received, the patient feedback information is parsed into structured data, stored in the medical information system, and the structured data is used for subsequent dynamic adjustment and model updating.

8. A multi-task synergistic prediction medical follow-up system for implementing the multi-task synergistic prediction medical follow-up method according to claim 1, characterized in that, The system comprises: A multi-source data fusion module is configured to interface with a hospital information system, an electronic medical record system, and a test and image system through an interface, collect patient basic information, diagnosis and treatment records, test indexes, and image reports, receive actively reported symptom self-evaluation, medication records, and lifestyle information from the patient end, and perform structured processing, standardized processing, and missing value filling on the collected multi-source medical data to generate a target data set in a unified format; A multi-task algorithm coordination module is configured to input the target data set into a random forest classification model, a gradient boosting regression tree model, and a long short-term memory network model, calculate a patient risk level, a key monitoring index short-term prediction result, and a disease long-term trend prediction result, and dynamically update and train each model based on newly added follow-up data within a preset time period; A follow-up plan generation module is configured to automatically generate and dynamically adjust a personalized follow-up plan based on the risk level output by the classification model, the abnormal fluctuation judgment of the regression model, and the long-term trend analysis of the time series model, in combination with an embedded disease feature library and core monitoring index weights, including follow-up intervals and follow-up project combinations; A follow-up execution and feedback module is configured to push a follow-up task to the patient end, receive and parse patient feedback information, store the feedback data in the medical information system, and use the feedback data for dynamic adjustment of the follow-up plan and continuous optimization of the model; A physician interaction module is configured to provide a physician end visual interface for displaying algorithm prediction results and follow-up execution status, supporting manual adjustment and temporary intervention of the follow-up plan by the physician, and associating adjustment and intervention records to subsequent model optimization processes.

9. A terminal, characterized by comprising: The system comprises: A memory is configured to store a multi-task collaborative prediction medical follow-up simulation program; A processor is configured to execute the multi-task collaborative prediction medical follow-up device to implement the steps of the multi-task collaborative prediction medical follow-up method of claim 1.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions, and when a computer reads the computer instructions in the storage medium, the computer executes the multi-task collaborative prediction medical follow-up method of claim 1.

Images