A method, system, device, and medium for home continuation of care for a neurological condition

Abstract

The application relates to a home follow-up care method, system, device and medium for nervous system diseases. The method comprises the following steps: acquiring multi-source illness data of a patient, and constructing a patient illness state model based on the multi-source illness data; performing rehabilitation effect evaluation according to a preset rehabilitation expected baseline based on the patient illness state model, and obtaining rehabilitation effect data; performing risk prediction by using a machine learning model based on the patient illness state model and the rehabilitation effect data, and obtaining risk prediction data; and generating a comprehensive nursing scheme according to a preset comprehensive nursing scheme generation rule based on the patient illness state model, the rehabilitation effect data and the risk prediction data. The method can realize personalized home nursing for nervous system disease patients through intelligent data integration and analysis, and optimize nursing efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical care, and in particular relates to a method, system, device and medium for home-based continuous care of neurological diseases. Background Technology

[0002] With the development of technologies in the field of intelligent medical care, technologies such as remote monitoring and multi-source data acquisition are gradually being applied to the nursing scenarios of neurological diseases. These technologies enable real-time acquisition of patients' basic health data, promoting home-based continued care as an important method for the rehabilitation of discharged patients. Traditionally, home care for patients with neurological diseases relies heavily on a fixed nursing manual provided by medical staff upon discharge. Family members manually record the patient's vital signs, diet, and activity levels. Patients return to the hospital for regular follow-up visits, and medical staff adjust nursing recommendations based on the information from these visits. However, traditional methods have the following drawbacks: multi-source disease data is scattered and not effectively integrated and analyzed; rehabilitation effect assessment relies solely on experience and lacks personalized baseline support; risk prediction is lagging, making it difficult to identify potential risks such as disease deterioration in advance; nursing plans are highly generic and cannot adapt to the dynamic changes in individual patients' conditions, resulting in insufficient targeted and inefficient care, failing to meet the complex and continuously changing home rehabilitation nursing needs of patients with neurological diseases. Summary of the Invention

[0003] Therefore, it is necessary to provide a home-based continuing care method, system, device, and medium for a neurological disease that can solve the above problems.

[0004] Firstly, this application provides a method for home-based continuing care of neurological diseases, including:

[0005] Acquire multi-source disease data of patients and construct a patient disease status model based on the multi-source disease data;

[0006] Based on the patient's condition status model, the rehabilitation effect is evaluated according to the preset rehabilitation expectation baseline to obtain rehabilitation effect data;

[0007] Based on the patient's condition status model and rehabilitation effect data, a machine learning model is used to predict risks and obtain risk prediction data.

[0008] Based on the patient's condition model, rehabilitation effect data, and risk prediction data, a comprehensive nursing plan is generated according to the preset comprehensive nursing plan generation rules.

[0009] In one embodiment, the multi-source disease data includes basic vital signs data, cognitive and behavioral assessment data, and dynamic monitoring data;

[0010] Based on multi-source disease data, a patient disease status model is constructed, including:

[0011] Based on basic vital sign data, extract basic physiological parameters;

[0012] Based on cognitive behavioral assessment data, basic cognitive behavioral parameters are extracted;

[0013] Based on dynamic monitoring data, physiological parameter change features corresponding to basic physiological parameters and cognitive behavior change features corresponding to basic cognitive behavior parameters are extracted to obtain parameter change features;

[0014] Based on the characteristics of each parameter change, a correlation analysis is performed to obtain the correlation relationship of parameter changes;

[0015] Based on the correlation of parameter changes, topological structure modeling is performed to obtain the basic model framework;

[0016] By integrating basic physiological parameters, basic cognitive and behavioral parameters, physiological parameter change characteristics, and cognitive and behavioral change characteristics into a basic model framework, a patient condition status model is obtained.

[0017] In one embodiment, based on a patient's condition status model, rehabilitation effectiveness is assessed according to a preset rehabilitation expectation baseline to obtain rehabilitation effectiveness data, including:

[0018] Based on the expected baseline of rehabilitation, combined with basic physiological parameters and basic cognitive and behavioral parameters, rehabilitation baseline nodes are determined, and based on the rehabilitation baseline nodes and the expected baseline of rehabilitation, the set of phased expected baselines and the correlation between each baseline in the set of phased expected baselines are obtained.

[0019] Based on the characteristics of each parameter change, a set of parameter change curves is generated, and based on each parameter change curve, the deviation from the baseline corresponding to the phased expected baseline set is calculated to obtain baseline deviation data.

[0020] Based on the correlation of parameter changes, the similarity of the correlation with each baseline in the phased expected baseline set is calculated to obtain baseline similarity data;

[0021] A pre-defined weighted fusion evaluation function is used to determine the baseline deviation weight and baseline similarity weight based on the rehabilitation baseline nodes;

[0022] The rehabilitation effect data is obtained by weighting baseline deviation data, baseline similarity data, baseline deviation weight, and baseline similarity weight.

[0023] In one embodiment, the machine learning model includes a multimodal feature extraction layer, a dynamic feature enhancement layer, a spatiotemporal feature fusion layer, and a multi-task risk classifier;

[0024] Based on patient condition status models and rehabilitation outcome data, machine learning models are used for risk prediction, yielding risk prediction data, including:

[0025] The patient's condition status model and rehabilitation effect data are input into the multimodal feature extraction layer, and multidimensional temporal features are extracted using the sliding window method.

[0026] Multi-dimensional temporal features are input into a dynamic feature enhancement layer, and a multi-head attention mechanism is used to enhance the features, resulting in enhanced multi-dimensional temporal features.

[0027] The enhanced multi-dimensional temporal features are input into the spatiotemporal feature fusion layer for processing, generating a spatiotemporal correlated state representation vector.

[0028] The spatiotemporal feature fusion layer includes a dual-path LSTM network. One LSTM network is used to process the physiological dimension parameters in the multi-dimensional enhanced temporal features to obtain the physiological parameter time dependence. The other LSTM network is used to process the cognitive behavior dimension parameters in the multi-dimensional enhanced temporal features to obtain the cognitive behavior event sequence. The dual-path LSTM network is also used to fuse the physiological parameter time dependence and the cognitive behavior event sequence through an attention gating mechanism to obtain the spatiotemporal associated state representation vector.

[0029] The spatiotemporal correlation state representation vector is input into the multi-task risk classifier, and risk prediction data is generated according to the preset risk prediction index system.

[0030] The multi-task risk classifier comprises two parallel MLP classifiers. One MLP classifier uses the Softmax activation function to output a multi-class risk level probability distribution based on the spatiotemporal correlation state representation vector and the risk prediction index system. The other MLP classifier uses the Sigmoid activation function to output a probability calibration value corresponding to the multi-class risk level probability distribution based on the risk prediction index system. The two parallel MLP classifiers are also used to generate risk prediction data based on the multi-class risk level probability distribution and the probability calibration value using the weighted cross-entropy loss function.

[0031] In one embodiment, the formula for calculating the weighted rehabilitation effect data is as follows:

[0032]

[0033] in, The overall score for rehabilitation effectiveness. The fusion coefficient of physiological parameters Let be the baseline deviation of the i-th physiological parameter. The time decay coefficient, Let be the monitoring timestamp for the i-th physiological parameter. The baseline deviation weight for the i-th physiological parameter is... Let be the calibrated similarity of the j-th cognitive behavioral parameter. Let be the confidence coefficient of the j-th cognitive behavioral parameter. Let be the baseline similarity weight corresponding to the j-th cognitive behavioral parameter. The fusion coefficients are the cognitive-behavioral parameters.

[0034] In one embodiment, the comprehensive nursing plan generation rules include multi-terminal push rules, consultation-follow-up-review rules, and visual guidance rules;

[0035] Based on the patient's condition model, rehabilitation outcome data, and risk prediction data, a comprehensive nursing care plan is generated according to pre-defined comprehensive nursing care plan generation rules, including:

[0036] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, multi-terminal push rules are adopted to generate multi-terminal feedback data. These multi-terminal push rules include push rules for medical devices, community hospitals, patients, and patients' families.

[0037] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a consultation-follow-review rule is adopted to generate a consultation checklist, follow-up tasks, and review tasks.

[0038] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a visualization guidance rule is adopted to match the corresponding visualization guidance content from a pre-set visualization nursing guidance library.

[0039] By integrating feedback data from multiple sources, consultation lists, follow-up tasks, re-examination tasks, and visual guidance content, a comprehensive nursing plan is obtained.

[0040] In one embodiment, based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a visualization guidance rule is used to match corresponding visualization guidance content from a pre-set visualization nursing guidance library, including:

[0041] Based on patient condition models, rehabilitation outcome data, and risk prediction data, nursing needs characteristics are extracted.

[0042] Based on the preset nursing scenarios, the characteristics of nursing needs are classified to obtain scenario classification results;

[0043] Based on the scenario classification results, candidate guidance content for the corresponding scenario is associated with the visualized nursing guidance library;

[0044] Based on the risk levels in the risk prediction data, the candidate guidance content is prioritized and sorted to obtain the sorted candidate guidance content. Then, based on the sorted candidate guidance content, the top N candidate guidance content is selected to obtain the visualized guidance content.

[0045] Secondly, this application also provides a home-based continuing care system for neurological diseases, comprising:

[0046] The disease data modeling module is used to acquire multi-source disease data of patients and build a patient disease status model based on the multi-source disease data;

[0047] The rehabilitation effect assessment module is used to assess the rehabilitation effect based on the patient's condition status model and according to the preset rehabilitation expectation baseline, and obtain rehabilitation effect data.

[0048] The risk prediction and analysis module is used to perform risk prediction based on the patient's condition status model and rehabilitation effect data, and to obtain risk prediction data using a machine learning model.

[0049] The nursing plan generation module is used to generate comprehensive nursing plans based on the patient's condition status model, rehabilitation effect data, and risk prediction data, according to preset comprehensive nursing plan generation rules.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned method for home-based continuous care of a neurological disease.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for home-based continuous care of a neurological disease.

[0052] The aforementioned home-based continuing care method, system, equipment, and medium for a neurological disease acquires multi-source patient condition data and constructs a patient condition status model based on this data. This achieves multi-source integration and digital representation of the condition, laying a data foundation for personalized care. Based on this patient condition status model, rehabilitation effectiveness is evaluated according to a preset rehabilitation expectation baseline, yielding rehabilitation effectiveness data. This objectively quantifies rehabilitation progress and supports real-time monitoring and adjustment of nursing strategies. Combining the patient condition status model and rehabilitation effectiveness data, a machine learning model is used for risk prediction, obtaining risk prediction data and enhancing risk warning capabilities, thus enabling preventative care. Based on the patient condition status model, rehabilitation effectiveness data, and risk prediction data, a comprehensive nursing plan is generated according to preset comprehensive nursing plan generation rules, achieving automated and personalized customization of the nursing plan, improving nursing efficiency, and ensuring that nursing measures accurately match patient needs. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart of a home-based continuing care method for neurological diseases according to the present invention;

[0055] Figure 2 This is a structural diagram of a home-based continuing care system for neurological diseases according to the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, such as Figure 1 As shown, a home-based continuing care method for neurological diseases is provided. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and to an architecture including both a terminal and a server, and implemented through interaction between the terminal and the server. In the implementation environment of this application, the hardware architecture mainly includes the patient's home-based terminal device (such as a smart wearable device, mobile terminal, or dedicated medical monitoring instrument) and a server platform at a community hospital or medical institution. These components communicate via a network. The terminal device is responsible for collecting multi-source patient condition data in real time, including basic vital signs, cognitive behavioral assessment results, and dynamic monitoring information. The server is used to run algorithms for condition status modeling, rehabilitation effect assessment, risk prediction analysis, and nursing plan generation. The system achieves data transmission, centralized processing, and distributed feedback through the collaboration of the terminal and the server. Application scenarios include: when patients with neurological diseases require continued home care, terminal devices continuously monitor physiological parameters and behavioral data and upload them to the server; the server builds a patient condition status model based on the received data, conducts quantitative assessment of rehabilitation effects and machine learning risk prediction, generates a personalized comprehensive care plan, and then distributes the plan content to the medical device terminal, community hospital terminal, patient terminal and family terminal through multi-terminal push rules, realizing real-time interaction of consultation reminders, follow-up task allocation and visual guidance, thereby meeting the needs of personalized and efficient home care.

[0058] In this embodiment, the method includes the following steps:

[0059] S01. Obtain multi-source disease data of patients and construct a patient disease status model based on the multi-source disease data.

[0060] Multi-source disease data refers to various types of disease-related information collected from patients, including basic vital sign data (such as physiological parameters like heart rate and blood pressure), cognitive and behavioral assessment data (such as memory tests and behavioral observation results), and dynamic monitoring data (such as continuously collected physiological changes or activity records). This data can be obtained through medical devices, sensors, or manual assessment tools. In practice, multi-source data can be preprocessed to extract key parameters, such as basic physiological parameters from basic vital sign data and basic cognitive and behavioral parameters from cognitive and behavioral assessment data. Based on dynamic monitoring data, the characteristics of physiological parameter changes and cognitive and behavioral changes can be analyzed. Dynamic relationships between parameters can be revealed through correlation analysis, and a basic model framework can be generated using topological modeling. Various parameters and features can be integrated into this framework to form a comprehensive model that can digitally represent the patient's disease status. This process can also employ data integration or machine learning techniques, allowing the model to dynamically adapt to individual patient differences, providing a foundation for subsequent rehabilitation assessment and risk prediction.

[0061] S02, based on the patient's condition status model, evaluates the rehabilitation effect according to the preset rehabilitation expectation baseline, and obtains rehabilitation effect data.

[0062] The rehabilitation baseline refers to pre-set rehabilitation goals and standards, including expected value sequences of physiological parameters (such as heart rate and blood pressure) and cognitive behavioral parameters (such as memory test scores), which can be dynamically adjusted based on clinical guidelines or individualized historical data. Rehabilitation outcome data refers to comprehensive indicators reflecting patient rehabilitation progress generated through quantitative assessment, such as rehabilitation outcome scores or probability distributions. In practice, multi-source parameters (such as basic physiological parameters and basic cognitive behavioral parameters) in the patient's condition model can be compared with the rehabilitation baseline. A set of parameter change curves is generated, and the deviation from the interim expected baseline is calculated. Euclidean distance or dynamic time warping algorithms can be used to calculate baseline deviation data. The similarity between the parameter change correlation and the baseline correlation is calculated (cosine similarity or graph matching techniques can be used) to obtain baseline similarity data. A weighted fusion evaluation function (such as combining time decay coefficients and confidence weights) is used to synthesize these data to generate rehabilitation outcome data.

[0063] S03. Based on the patient's condition status model and rehabilitation effect data, a machine learning model is used to predict risks and obtain risk prediction data.

[0064] In this context, risk prediction data refers to comprehensive indicators reflecting a patient's potential risk, such as risk level probability distributions or calibration values, output by the machine learning model. The machine learning model in this application refers to a computational architecture capable of learning risk patterns from multi-source time-series data, including neural networks, ensemble learning, or deep learning models. In implementation, patient condition models and rehabilitation effect data can be used as inputs. Multi-dimensional time-series features are extracted through a feature extraction module (e.g., sliding window method, convolutional neural network, or autoencoder). Feature enhancement techniques (e.g., multi-head attention mechanism, self-attention, or gated recurrent unit) are used to optimize feature representations to capture dynamic changes and key patterns. A spatiotemporal fusion module (e.g., dual-path LSTM network, graph neural network, or Transformer structure) integrates time-series dependencies and parameter correlations to generate a unified spatiotemporal correlation state representation vector. A multi-task classifier (e.g., multilayer perceptron combined with Softmax and Sigmoid activation functions) can be used to output risk prediction data based on a preset risk indicator system (e.g., classification probability and calibration values).

[0065] S04. Based on the patient's condition model, rehabilitation effect data, and risk prediction data, a comprehensive nursing plan is generated according to the preset comprehensive nursing plan generation rules.

[0066] The comprehensive nursing plan generation rules are a pre-defined set of rules for customizing personalized nursing plans. These include multi-terminal push rules (such as push logic for medical devices, community hospitals, patients, and family members), consultation-follow-up-review rules (such as the process of generating consultation lists, follow-up tasks, and review tasks), and visual guidance rules (such as methods for matching guidance content from a pre-defined library). A comprehensive nursing plan refers to a personalized nursing plan formed by integrating multi-terminal feedback data, consultation lists, follow-up tasks, review tasks, and visual guidance content. In implementation, based on patient condition models, rehabilitation effect data, and risk prediction data, multi-terminal push rules are applied to generate feedback data for different terminals; consultation lists, follow-up tasks, and review tasks are automatically generated using consultation-follow-up-review rules; and corresponding visual guidance content is matched using visual guidance rules (such as extracting nursing need characteristics, classifying according to nursing scenarios, associating candidate guidance content, and sorting based on risk level); these outputs are integrated into a unified comprehensive nursing plan, realizing the entire process from data collection to plan generation.

[0067] In one embodiment, the multi-source disease data includes basic vital signs data, cognitive and behavioral assessment data, and dynamic monitoring data;

[0068] Based on multi-source disease data, a patient disease status model is constructed, including:

[0069] S11, based on basic vital sign data, extracts basic physiological parameters;

[0070] S12, Extract basic cognitive behavior parameters based on cognitive behavior assessment data;

[0071] S13, Based on dynamic monitoring data, extract the physiological parameter change features corresponding to the basic physiological parameters, and extract the cognitive behavior change features corresponding to the basic cognitive behavior parameters to obtain parameter change features;

[0072] S14. Based on the characteristics of each parameter change, perform a change correlation analysis to obtain the parameter change correlation relationship;

[0073] S15. Based on the correlation of parameter changes, topological structure modeling is performed to obtain the basic model framework;

[0074] S16 integrates basic physiological parameters, basic cognitive and behavioral parameters, physiological parameter change characteristics, and cognitive and behavioral change characteristics into a basic model framework to obtain a patient condition status model.

[0075] For example, from basic vital sign data, basic physiological parameters such as heart rate, blood pressure, and blood oxygen saturation can be extracted through data preprocessing, denoising, and normalization. From cognitive behavioral assessment data, basic cognitive behavioral parameters such as memory scores, executive function scores, and frequency of daily behaviors can be extracted using the quantitative algorithms corresponding to standardized assessment scales. Based on dynamic monitoring data, time series analysis methods can be used to extract physiological parameter change characteristics such as trend slope and fluctuation amplitude corresponding to basic physiological parameters, as well as cognitive behavioral change characteristics such as change frequency and improvement rate corresponding to basic cognitive behavioral parameters, and integrate them to form parameter change characteristics. Pearson correlation coefficient and mutual information algorithm can be used to perform change correlation analysis on each parameter change characteristic to determine the positive and negative correlations between different characteristics. Using parameter change characteristics as nodes and correlation relationships as edges, graph theory modeling tools can be used to perform topological structure modeling to obtain the basic model framework. According to the correlation logic between nodes and edges in the topological structure, basic physiological parameters, basic cognitive behavioral parameters, physiological parameter change characteristics, and cognitive behavioral change characteristics are filled into the basic model framework to complete the patient's condition status model.

[0076] In one embodiment, based on a patient's condition status model, rehabilitation effectiveness is assessed according to a preset rehabilitation expectation baseline to obtain rehabilitation effectiveness data, including:

[0077] S21. Based on the expected baseline of rehabilitation, combined with basic physiological parameters and basic cognitive and behavioral parameters, determine the rehabilitation baseline nodes, and obtain the set of phased expected baselines and the correlation between each baseline in the set of phased expected baselines according to the rehabilitation baseline nodes and the expected baseline of rehabilitation.

[0078] S22. Based on the characteristics of each parameter change, a set of parameter change curves is generated, and based on each parameter change curve, the deviation from the baseline corresponding to the phased expected baseline set is calculated to obtain baseline deviation data.

[0079] S23. Based on the correlation of parameter changes, calculate the similarity of the correlation with each baseline in the phased expected baseline set to obtain baseline similarity data.

[0080] S24. Using a preset weighted fusion evaluation function, the baseline deviation weight and baseline similarity weight are determined based on the rehabilitation baseline nodes.

[0081] S25, based on baseline deviation data, baseline similarity data, baseline deviation weight, and baseline similarity weight, weighted data are used to obtain rehabilitation effect data.

[0082] Specifically, when evaluating rehabilitation effectiveness, a pre-set rehabilitation expectation baseline is first used as a benchmark. This is combined with basic physiological and cognitive behavioral parameters from the patient's condition model. Clinical rehabilitation standards and the patient's initial condition data can be referenced to determine rehabilitation baseline nodes. The rehabilitation expectation baseline is then divided into sets of staged expectation baselines based on these nodes. Graph theory analysis can be used to clarify the dependencies or influence relationships between the staged baselines. Based on the characteristics of parameter changes, a time series fitting algorithm is used to generate a set of parameter change curves. Euclidean distance or dynamic time warping algorithms are used to calculate the deviation of each curve from the corresponding baseline in the staged expectation baseline set, obtaining baseline deviation data. Based on the correlation of parameter changes, a cosine similarity algorithm can be used to calculate the matching degree between the curve and the correlation between each baseline in the staged expectation baseline set, generating baseline similarity data. A pre-set weighted fusion evaluation function is called. Based on the rehabilitation criticality of the rehabilitation baseline nodes (e.g., the weight of acute phase nodes is higher than that of stable phase nodes), the analytic hierarchy process (AHP) can be used to determine the baseline deviation weight and baseline similarity weight. By weighting baseline deviation data and baseline similarity data with their respective weights, quantitative rehabilitation effect data is obtained, which objectively reflects the degree to which the patient's rehabilitation progress matches the expected goals.

[0083] In one embodiment, the machine learning model includes a multimodal feature extraction layer, a dynamic feature enhancement layer, a spatiotemporal feature fusion layer, and a multi-task risk classifier;

[0084] Based on patient condition status models and rehabilitation outcome data, machine learning models are used for risk prediction, yielding risk prediction data, including:

[0085] S31, the patient's condition status model and rehabilitation effect data are input into the multimodal feature extraction layer, and multi-dimensional temporal features are extracted using the sliding window method;

[0086] S32, input the multi-dimensional temporal features into the dynamic feature enhancement layer, and use a multi-head attention mechanism to enhance the features to obtain enhanced multi-dimensional temporal features;

[0087] S33, the enhanced multi-dimensional temporal features are input into the spatiotemporal feature fusion layer for processing, generating a spatiotemporal related state representation vector;

[0088] The spatiotemporal feature fusion layer includes a dual-path LSTM network. One LSTM network is used to process the physiological dimension parameters in the multi-dimensional enhanced temporal features to obtain the physiological parameter time dependence. The other LSTM network is used to process the cognitive behavior dimension parameters in the multi-dimensional enhanced temporal features to obtain the cognitive behavior event sequence. The dual-path LSTM network is also used to fuse the physiological parameter time dependence and the cognitive behavior event sequence through an attention gating mechanism to obtain the spatiotemporal associated state representation vector.

[0089] S34, input the spatiotemporal correlation state representation vector into the multi-task risk classifier, and generate risk prediction data according to the preset risk prediction index system;

[0090] The multi-task risk classifier comprises two parallel MLP classifiers. One MLP classifier uses the Softmax activation function to output a multi-class risk level probability distribution based on the spatiotemporal correlation state representation vector and the risk prediction index system. The other MLP classifier uses the Sigmoid activation function to output a probability calibration value corresponding to the multi-class risk level probability distribution based on the risk prediction index system. The two parallel MLP classifiers are also used to generate risk prediction data based on the multi-class risk level probability distribution and the probability calibration value using the weighted cross-entropy loss function.

[0091] For example, patient condition status models and rehabilitation effect data can be input into a machine learning model containing a multimodal feature extraction layer, a dynamic feature enhancement layer, a spatiotemporal feature fusion layer, and a multi-task risk classifier. The multimodal feature extraction layer can use a sliding window method with a window size of 30 minutes and a step size of 10 minutes to extract multi-dimensional temporal features. The dynamic feature enhancement layer can use an 8-head multi-head attention mechanism to assign high weights to key features, resulting in enhanced multi-dimensional temporal features. In the dual-path LSTM network (Long Short-Term Memory network, a type of recurrent neural network), one path processes physiological dimension parameters, outputting the time dependence of physiological parameters, while the other path processes cognitive behavior dimension parameters, generating a sequence of cognitive behavior events. The weights of the two outputs are adjusted and fused through an attention gating mechanism to obtain a spatiotemporally related state representation vector. The vector is input into a multi-task risk classifier. The MLP (Multilayer Perceptron, a supervised machine learning algorithm based on a feedforward artificial neural network) classifier, which uses the Softmax activation function (an activation function for multi-classification problems that converts any real-valued vector output by a neural network into a probability distribution), outputs a multi-class risk level probability distribution based on different risk scenarios and a preset prediction index system of low, medium, and high risk levels. The MLP classifier, which uses the Sigmoid activation function (an S-shaped activation function that maps the input to the (0,1) interval, used for binary classification problems), outputs the corresponding probability calibration value. The two classifiers can be fused using a weighted cross-entropy loss function with a weight ratio of 0.7:0.3 to generate risk prediction data.

[0092] In one embodiment, S41, the formula for calculating the weighted rehabilitation effect data is:

[0093]

[0094] Among them, The overall score for rehabilitation effectiveness. The fusion coefficient of physiological parameters Let be the baseline deviation of the i-th physiological parameter. The time decay coefficient, Let be the monitoring timestamp for the i-th physiological parameter. The baseline deviation weight for the i-th physiological parameter is... Let be the calibrated similarity of the j-th cognitive behavioral parameter. Let be the confidence coefficient of the j-th cognitive behavioral parameter. Let be the baseline similarity weight corresponding to the j-th cognitive behavioral parameter. The fusion coefficients are the cognitive-behavioral parameters.

[0095] Specifically, the core purpose of this formula is to weight and fuse relevant assessment data of physiological parameters and cognitive behavioral parameters to quantify the patient's rehabilitation effect and output a comprehensive rehabilitation effect score. This provides an objective quantitative basis for subsequent risk prediction and the generation of comprehensive care plans. The formula uses physiological parameter fusion coefficients. Fusion coefficient with cognitive behavioral parameters Balance the weights of the two dimensions of evaluation, among which and The values ​​are all in the range of 0-1 and satisfy the following conditions: The calculation of physiological parameters includes the baseline deviation of the i-th physiological parameter. With time decay term (λ is taken as 0.01-0.05,) Multiplying by the data (in days) reflects the impact of the timeliness of the monitoring data on the assessment, and is weighted by the corresponding baseline deviation. Weighted summation yields the physiological dimension assessment results; for the cognitive behavioral parameters, the calibrated similarity of the j-th cognitive behavioral parameter is calculated. With confidence coefficient (Values ​​range from 0.8 to 1.0, set according to the reliability of the evaluation data) multiplied, and then multiplied by the corresponding baseline similarity weight. The weighted summation yields the assessment results for the cognitive-behavioral dimension. The assessment results for the two dimensions are then fused using α and β weighting respectively, ultimately resulting in... This formula achieves systematic integration of multi-dimensional assessment data by associating baseline deviation data, baseline similarity data, and weights determined by rehabilitation baseline nodes, thereby ensuring the comprehensiveness and accuracy of rehabilitation effect assessment.

[0096] In one embodiment, the comprehensive nursing plan generation rules include multi-terminal push rules, consultation-follow-up-review rules, and visual guidance rules;

[0097] Based on the patient's condition model, rehabilitation outcome data, and risk prediction data, a comprehensive nursing care plan is generated according to pre-defined comprehensive nursing care plan generation rules, including:

[0098] S51, based on the patient's condition status model, rehabilitation effect data and risk prediction data, adopts multi-terminal push rules to generate multi-terminal feedback data. Among them, the multi-terminal push rules include push rules for medical devices, push rules for community hospitals, push rules for patients, and push rules for patients' family members.

[0099] S52, based on the patient's condition status model, rehabilitation effect data and risk prediction data, adopts the consultation-follow-review rule to generate consultation list, follow-up task and review task;

[0100] S53, based on the patient's condition status model, rehabilitation effect data and risk prediction data, adopts visualization guidance rules to match the corresponding visualization guidance content from the preset visualization nursing guidance library;

[0101] S54 integrates multi-terminal feedback data, consultation checklists, follow-up tasks, re-examination tasks, and visual guidance content to obtain a comprehensive nursing plan.

[0102] For example, when generating a comprehensive nursing care plan, it is based on preset multi-terminal push rules, consultation-follow-up-review rules, and visual guidance rules, combined with patient condition status models, rehabilitation effect data, and risk prediction data, and executed step by step. According to the multi-terminal push rules, suggestions for adjusting monitoring thresholds of push parameters to medical devices can be sent, patient rehabilitation progress and high-risk warning information can be pushed to community hospitals, daily nursing operation reminders can be sent to patients, and key nursing assistance matters can be pushed to family members, generating multi-dimensional, multi-terminal feedback data. According to the consultation-follow-up-review rules, the core issues of the consultation list are determined based on the lagging rehabilitation progress items in the rehabilitation effect data. The follow-up frequency is set based on the risk level of the risk prediction data (e.g., twice a week for high risk, once a week for medium risk, and once every two weeks for low risk). The re-examination items and time nodes are determined based on the abnormalities of key parameters in the patient condition status model, generating corresponding consultation lists, follow-up tasks, and re-examination tasks. Through visual guidance rules, text, image, and video guidance content adapted to the patient's nursing needs is matched from a preset visual nursing guidance library. By integrating multi-terminal feedback data, consultation lists, follow-up tasks, re-examination tasks, and visual guidance content according to the logic of "priority-execution sequence-responsible entity", a comprehensive nursing plan that can be directly implemented is formed, realizing the dynamic matching of nursing measures with the individual patient's condition.

[0103] In one embodiment, based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a visualization guidance rule is used to match corresponding visualization guidance content from a pre-set visualization nursing guidance library, including:

[0104] S61, based on the patient's condition status model, rehabilitation effect data and risk prediction data, extract nursing needs characteristics;

[0105] S62, Based on the preset nursing scenarios, classify the nursing needs characteristics to obtain scenario classification results;

[0106] S63, Based on the scenario classification results, associate the candidate guidance content of the corresponding scenario in the visualized nursing guidance library;

[0107] S64. Based on the risk level in the risk prediction data, prioritize the candidate guidance content to obtain the ranked candidate guidance content. Then, based on the ranked candidate guidance content, select the top N candidate guidance content to obtain the visualized guidance content.

[0108] Specifically, when matching visualized guidance content, core nursing needs features can be extracted using principal component analysis (PCA) algorithms in feature engineering. These features include key needs information across dimensions such as physiological care, rehabilitation training, cognitive intervention, and safety protection. Based on pre-defined nursing scenarios such as medication guidance, limb rehabilitation, cognitive training, and dietary management, the extracted nursing needs features can be categorized using a K-nearest neighbor classification algorithm to obtain clear scenario classification results. Based on these classification results, a scenario tag matching mechanism is used to retrieve all candidate guidance content for the corresponding scenario from a pre-defined visualized nursing guidance library (which stores resources such as illustrated tutorials, operation videos, and step-by-step diagrams categorized by scenario). Referring to the high, medium, and low risk levels in the risk prediction data, a weighted ranking method (0.6 for high risk, 0.3 for medium risk, and 0.1 for low risk) is used to prioritize the candidate guidance content. The preset N value is 3-5 (which can be adjusted according to the clinical nursing priority). The top N candidate guidance contents are selected to form a visual guidance content that is adapted to the individual needs of patients, so as to achieve targeted guidance of the plan.

[0109] The aforementioned home-based continuing care method for neurological diseases acquires multi-source disease data, including basic vital signs, cognitive and behavioral assessments, and dynamic monitoring. It extracts basic parameters and changing characteristics, and constructs a patient disease status model through correlation analysis and topological modeling. This addresses the problem of fragmented and ineffective integration of multi-source disease data in traditional methods. Based on this model and a pre-set rehabilitation expectation baseline, rehabilitation effect data is obtained by determining rehabilitation baseline nodes, calculating baseline deviation and similarity, and weighted fusion. This avoids reliance on experience-based judgment in rehabilitation effect assessment and addresses the lack of personalized baseline support. The method employs a multi-modal feature extraction layer and a dynamic feature enhancement layer. The machine learning model, consisting of a dual-channel LSTM spatiotemporal feature fusion layer and a multi-task risk classifier, generates risk prediction data by inputting disease status models and rehabilitation effect data, overcoming the shortcomings of traditional technologies in risk prediction lag. Based on preset multi-terminal push, consultation-follow-review, and visualization guidance rules, it integrates multi-terminal feedback data, consultation lists, follow-up tasks, review tasks, and visualization guidance content to generate a comprehensive nursing plan. This solves the problem that nursing plans cannot adapt to the dynamic changes in individual patients' conditions, enabling personalized home care for patients with neurological diseases, improving the targeting and efficiency of nursing care, and meeting the complex and continuously changing home rehabilitation nursing needs of patients.

[0110] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0111] Based on the same inventive concept, this application also provides a home-based continuing care system for neurological diseases to implement the aforementioned method for home-based continuing care of neurological diseases. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of a home-based continuing care system for neurological diseases provided below can be found in the limitations of the home-based continuing care method for neurological diseases described above, and will not be repeated here.

[0112] In one exemplary embodiment, such as Figure 2 As shown, a home-based continuing care system for neurological disorders is provided, comprising:

[0113] The disease data modeling module 101 is used to acquire multi-source disease data of patients and construct a patient disease status model based on the multi-source disease data;

[0114] The rehabilitation effect assessment module 102 is used to assess the rehabilitation effect based on the patient's condition status model and according to the preset rehabilitation expectation baseline, and obtain rehabilitation effect data.

[0115] The risk prediction and analysis module 103 is used to perform risk prediction based on the patient's condition status model and rehabilitation effect data, and to obtain risk prediction data by using a machine learning model.

[0116] The nursing plan generation module 104 is used to generate a comprehensive nursing plan based on the patient's condition status model, rehabilitation effect data, and risk prediction data, according to the preset comprehensive nursing plan generation rules.

[0117] In one embodiment, the multi-source disease data in the disease data modeling module 101 includes basic vital sign data, cognitive behavioral assessment data, and dynamic monitoring data;

[0118] The disease data modeling module 101 is also used for:

[0119] Based on basic vital sign data, extract basic physiological parameters;

[0120] Based on cognitive behavioral assessment data, basic cognitive behavioral parameters are extracted;

[0121] Based on dynamic monitoring data, physiological parameter change features corresponding to basic physiological parameters and cognitive behavior change features corresponding to basic cognitive behavior parameters are extracted to obtain parameter change features;

[0122] Based on the characteristics of each parameter change, a correlation analysis is performed to obtain the correlation relationship of parameter changes;

[0123] Based on the correlation of parameter changes, topological structure modeling is performed to obtain the basic model framework;

[0124] By integrating basic physiological parameters, basic cognitive and behavioral parameters, physiological parameter change characteristics, and cognitive and behavioral change characteristics into a basic model framework, a patient condition status model is obtained.

[0125] In one embodiment, the rehabilitation effect assessment module 102 is further configured to:

[0126] Based on the expected baseline of rehabilitation, combined with basic physiological parameters and basic cognitive and behavioral parameters, rehabilitation baseline nodes are determined, and based on the rehabilitation baseline nodes and the expected baseline of rehabilitation, the set of phased expected baselines and the correlation between each baseline in the set of phased expected baselines are obtained.

[0127] Based on the characteristics of each parameter change, a set of parameter change curves is generated, and based on each parameter change curve, the deviation from the baseline corresponding to the phased expected baseline set is calculated to obtain baseline deviation data.

[0128] Based on the correlation of parameter changes, the similarity of the correlation with each baseline in the phased expected baseline set is calculated to obtain baseline similarity data;

[0129] A pre-defined weighted fusion evaluation function is used to determine the baseline deviation weight and baseline similarity weight based on the rehabilitation baseline nodes;

[0130] The rehabilitation effect data is obtained by weighting baseline deviation data, baseline similarity data, baseline deviation weight, and baseline similarity weight.

[0131] In one embodiment, the machine learning model in the risk prediction and analysis module 103 includes a multimodal feature extraction layer, a dynamic feature enhancement layer, a spatiotemporal feature fusion layer, and a multi-task risk classifier;

[0132] Risk prediction and analysis module 103 is also used for:

[0133] The patient's condition status model and rehabilitation effect data are input into the multimodal feature extraction layer, and multidimensional temporal features are extracted using the sliding window method.

[0134] Multi-dimensional temporal features are input into a dynamic feature enhancement layer, and a multi-head attention mechanism is used to enhance the features, resulting in enhanced multi-dimensional temporal features.

[0135] The enhanced multi-dimensional temporal features are input into the spatiotemporal feature fusion layer for processing, generating a spatiotemporal correlated state representation vector.

[0136] The spatiotemporal feature fusion layer includes a dual-path LSTM network. One LSTM network is used to process the physiological dimension parameters in the multi-dimensional enhanced temporal features to obtain the physiological parameter time dependence. The other LSTM network is used to process the cognitive behavior dimension parameters in the multi-dimensional enhanced temporal features to obtain the cognitive behavior event sequence. The dual-path LSTM network is also used to fuse the physiological parameter time dependence and the cognitive behavior event sequence through an attention gating mechanism to obtain the spatiotemporal associated state representation vector.

[0137] The spatiotemporal correlation state representation vector is input into the multi-task risk classifier, and risk prediction data is generated according to the preset risk prediction index system.

[0138] The multi-task risk classifier comprises two parallel MLP classifiers. One MLP classifier uses the Softmax activation function to output a multi-class risk level probability distribution based on the spatiotemporal correlation state representation vector and the risk prediction index system. The other MLP classifier uses the Sigmoid activation function to output a probability calibration value corresponding to the multi-class risk level probability distribution based on the risk prediction index system. The two parallel MLP classifiers are also used to generate risk prediction data based on the multi-class risk level probability distribution and the probability calibration value using the weighted cross-entropy loss function.

[0139] In one embodiment, the weighted formula for calculating the rehabilitation effect data in the rehabilitation effect evaluation module 102 is as follows:

[0140]

[0141] in, The overall score for rehabilitation effectiveness. The fusion coefficient of physiological parameters Let be the baseline deviation of the i-th physiological parameter. The time decay coefficient, Let be the monitoring timestamp for the i-th physiological parameter. The baseline deviation weight for the i-th physiological parameter is... Let be the calibrated similarity of the j-th cognitive behavioral parameter. Let be the confidence coefficient of the j-th cognitive behavioral parameter. Let be the baseline similarity weight corresponding to the j-th cognitive behavioral parameter. The fusion coefficients are the cognitive-behavioral parameters.

[0142] In one embodiment, the comprehensive nursing plan generation rules in the nursing plan generation module 104 include multi-terminal push rules, consultation-follow-review rules, and visual guidance rules;

[0143] The nursing plan generation module 104 is also used for:

[0144] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, multi-terminal push rules are adopted to generate multi-terminal feedback data. These multi-terminal push rules include push rules for medical devices, community hospitals, patients, and patients' families.

[0145] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a consultation-follow-review rule is adopted to generate a consultation checklist, follow-up tasks, and review tasks.

[0146] Based on the patient's condition status model, rehabilitation effect data, and risk prediction data, a visualization guidance rule is adopted to match the corresponding visualization guidance content from a pre-set visualization nursing guidance library.

[0147] By integrating feedback data from multiple sources, consultation lists, follow-up tasks, re-examination tasks, and visual guidance content, a comprehensive nursing plan is obtained.

[0148] In one embodiment, the care plan generation module 104 is further configured to:

[0149] Based on patient condition models, rehabilitation outcome data, and risk prediction data, nursing needs characteristics are extracted.

[0150] Based on the preset nursing scenarios, the characteristics of nursing needs are classified to obtain scenario classification results;

[0151] Based on the scenario classification results, candidate guidance content for the corresponding scenario is associated with the visualized nursing guidance library;

[0152] Based on the risk levels in the risk prediction data, the candidate guidance content is prioritized and sorted to obtain the sorted candidate guidance content. Then, based on the sorted candidate guidance content, the top N candidate guidance content is selected to obtain the visualized guidance content.

[0153] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a home-based continuing care method for a neurological disease as described above.

[0154] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0155] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0156] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for continued home care of neurological diseases, characterized in that, The method includes: Acquire multi-source disease data of patients, and construct a patient disease status model based on the multi-source disease data; Based on the patient's condition model, the rehabilitation effect is evaluated according to the preset rehabilitation expectation baseline to obtain rehabilitation effect data; Based on the patient's condition status model and the rehabilitation effect data, a machine learning model is used to predict risks and obtain risk prediction data. Based on the patient's condition model, the rehabilitation effect data, and the risk prediction data, a comprehensive nursing plan is generated according to the preset comprehensive nursing plan generation rules.

2. The method according to claim 1, characterized in that, The multi-source disease data includes basic vital signs data, cognitive and behavioral assessment data, and dynamic monitoring data; The process of constructing a patient condition status model based on the multi-source disease data includes: Based on the aforementioned basic vital sign data, basic physiological parameters are extracted; Based on the cognitive behavior assessment data, basic cognitive behavior parameters are extracted; Based on the dynamic monitoring data, physiological parameter change features corresponding to the basic physiological parameters are extracted, and cognitive behavior change features corresponding to the basic cognitive behavior parameters are extracted to obtain parameter change features; Based on the variation characteristics of each parameter, a variation correlation analysis is performed to obtain the parameter variation correlation relationship; Based on the correlation of the parameter changes, topological structure modeling is performed to obtain the basic model framework; The basic physiological parameters, the basic cognitive-behavioral parameters, the physiological parameter change characteristics, and the cognitive-behavioral change characteristics are integrated into the basic model framework to obtain the patient's condition status model.

3. The method according to claim 2, characterized in that, The rehabilitation effect assessment, based on the patient's condition model and a preset rehabilitation expectation baseline, yields rehabilitation effect data, including: Based on the expected rehabilitation baseline, combined with the basic physiological parameters and the basic cognitive behavioral parameters, rehabilitation baseline nodes are determined, and according to the rehabilitation baseline nodes and the expected rehabilitation baseline, the set of phased expected baselines and the correlation between each baseline in the set of phased expected baselines are obtained. Based on the variation characteristics of each parameter, a set of parameter variation curves is generated, and based on each parameter variation curve, the deviation from the baseline corresponding to the set of expected baselines in the stage is calculated to obtain baseline deviation data. Based on the correlation of the parameter changes, the similarity of the correlation with each baseline in the phased expected baseline set is calculated to obtain baseline similarity data; Using a preset weighted fusion evaluation function, the baseline deviation weight and baseline similarity weight are determined based on the rehabilitation baseline nodes; The rehabilitation effect data is obtained by weighting the baseline deviation data, the baseline similarity data, the baseline deviation weight, and the baseline similarity weight.

4. The method according to claim 3, characterized in that, The machine learning model includes a multimodal feature extraction layer, a dynamic feature enhancement layer, a spatiotemporal feature fusion layer, and a multi-task risk classifier; The risk prediction data obtained by using a machine learning model based on the patient's condition status model and the rehabilitation effect data includes: The patient's condition model and the rehabilitation effect data are input into the multimodal feature extraction layer, and multidimensional temporal features are extracted using the sliding window method. The multi-dimensional temporal features are input into the dynamic feature enhancement layer, and a multi-head attention mechanism is used to enhance the features to obtain enhanced multi-dimensional temporal features. The enhanced multi-dimensional temporal features are input into the spatiotemporal feature fusion layer for processing to generate a spatiotemporal correlation state representation vector. The spatiotemporal feature fusion layer includes a dual-path LSTM network. One LSTM network processes the physiological dimension parameters in the multi-dimensional enhanced temporal features to obtain the physiological parameter time dependence. The other LSTM network processes the cognitive behavior dimension parameters in the multi-dimensional enhanced temporal features to obtain the cognitive behavior event sequence. The dual-path LSTM network is also used to fuse the physiological parameter time dependence and the cognitive behavior event sequence through an attention gating mechanism to obtain the spatiotemporal associated state representation vector. The spatiotemporal correlation state representation vector is input into the multi-task risk classifier, and risk prediction data is generated according to the preset risk prediction index system. The multi-task risk classifier comprises two parallel MLP classifiers. One MLP classifier uses the Softmax activation function to output a multi-class risk level probability distribution based on the spatiotemporal correlation state representation vector and the risk prediction index system. The other MLP classifier uses the Sigmoid activation function to output a probability calibration value corresponding to the multi-class risk level probability distribution based on the risk prediction index system. The two parallel MLP classifiers are also used to generate the risk prediction data using a weighted cross-entropy loss function, based on the multi-class risk level probability distribution and the probability calibration value.

5. The method according to claim 3, characterized in that, The formula for calculating the weighted rehabilitation effect data is as follows: in, The overall score for rehabilitation effectiveness. The fusion coefficient of physiological parameters Let be the baseline deviation of the i-th physiological parameter. The time decay coefficient, Let i be the monitoring timestamp of the i-th physiological parameter. The baseline deviation weight for the i-th physiological parameter is... Let be the calibrated similarity of the j-th cognitive behavioral parameter. Let be the confidence coefficient of the j-th cognitive behavioral parameter. Let be the baseline similarity weight corresponding to the j-th cognitive behavioral parameter. The fusion coefficients are the cognitive-behavioral parameters.

6. The method according to claim 1, characterized in that, The rules for generating the comprehensive nursing plan include multi-terminal push rules, consultation-follow-up-review rules, and visual guidance rules. The process involves generating a comprehensive nursing care plan based on the patient's condition model, the rehabilitation effect data, and the risk prediction data, according to preset comprehensive nursing care plan generation rules. This includes: Based on the patient's condition status model, the rehabilitation effect data, and the risk prediction data, multi-terminal push rules are used to generate multi-terminal feedback data, wherein the multi-terminal push rules include push rules for medical devices, push rules for community hospitals, push rules for patients, and push rules for patients' family members. Based on the patient's condition status model, the rehabilitation effect data, and the risk prediction data, the consultation-follow-review rule is used to generate a consultation list, follow-up task, and review task. Based on the patient's condition status model, the rehabilitation effect data, and the risk prediction data, the visualization guidance rules are used to match the corresponding visualization guidance content from the preset visualization nursing guidance library. The comprehensive nursing plan is obtained by integrating the multi-terminal feedback data, the consultation list, the follow-up tasks, the re-examination tasks, and the visual guidance content.

7. The method according to claim 6, characterized in that, Based on the patient's condition model, the rehabilitation effect data, and the risk prediction data, the visualization guidance rules are used to match corresponding visualization guidance content from a preset visualization nursing guidance library, including: Based on the patient's condition status model, the rehabilitation effect data, and the risk prediction data, nursing needs characteristics are extracted; Based on the preset nursing scenarios, the nursing needs characteristics are classified to obtain scenario classification results; Based on the scenario classification results, the candidate guidance content for the corresponding scenario in the visualized nursing guidance library is associated; Based on the risk levels in the risk prediction data, the candidate guidance content is prioritized to obtain the ranked candidate guidance content. Based on the ranked candidate guidance content, the top N candidate guidance content is selected to obtain the visualized guidance content.

8. A home-based continuing care system for neurological diseases, characterized in that, The system includes: The disease data modeling module is used to acquire multi-source disease data of patients and construct a patient disease status model based on the multi-source disease data; The rehabilitation effect assessment module is used to assess the rehabilitation effect based on the patient's condition status model and according to the preset rehabilitation expectation baseline, and obtain rehabilitation effect data. The risk prediction and analysis module is used to perform risk prediction based on the patient's condition status model and the rehabilitation effect data, using a machine learning model to obtain risk prediction data. The nursing plan generation module is used to generate a comprehensive nursing plan based on the patient's condition status model, the rehabilitation effect data, and the risk prediction data, according to preset comprehensive nursing plan generation rules.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images