System for assessing health status of elderly people based on multi-modal data

Abstract

The application relates to the technical field of data processing, and discloses an old person health state evaluation data processing system based on multi-modal data, which comprises a medical data integration module, an across-modal causal fusion processing module, a health state evolution modeling module, a backtracking evaluation report module and a decision output module. The medical data integration module acquires electronic medical record data and medical examination reports through a FHIR interface and extracts structured health indexes. The across-modal causal fusion processing module fuses monitoring data and medical texts through an image, a text and an image sandwich architecture. The health state evolution modeling module maps a health feature vector into three-dimensional state indexes of physiological functions, cognitive levels and motor abilities. The backtracking evaluation report module and the decision output module are used for backtracking evaluation and decision output. Through the image, the text and the image sandwich architecture, deep semantic fusion of multi-modal features is realized under the constraint of medical pathology rules, feature weight adaptive distribution is dynamically guided based on a medical causal graph, a high-dimensional fusion vector retaining key pathological information is generated, the semantic integration capability of health data is significantly improved, and the reliability of discrimination and decision is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a data processing system for assessing the health status of the elderly based on multimodal data. Background Technology

[0002] Faced with the accelerating trend of global population aging, the complex health problems of the elderly, such as multiple diseases and dynamic changes in health status, are becoming increasingly prominent. Real-time and accurate health status assessment has become a key link in achieving proactive intervention and delaying functional decline. In response to the health management needs of the elderly, it is necessary to build a scientific and quantitative assessment system to achieve early warning and intervention of health risks. In this context, by integrating multi-dimensional health information flow, the scientific nature, timeliness and effectiveness of elderly health management can be improved. Traditional multimodal health assessment solutions often use feature splicing or attention-weighted fusion to process image and text data.

[0003] However, in current technologies, the lack of pathological rules guiding the feature allocation of image features and text semantics reduces clinical interpretability, resulting in reduced accuracy and reliability of health status assessment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a data processing system for assessing the health status of the elderly based on multimodal data. This system solves the problem that the lack of pathological rule-guided feature allocation in the fusion of image features and text semantics reduces clinical interpretability, thereby decreasing the accuracy of health status assessment and the reliability of decision-making.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a data processing system for assessing the health status of the elderly based on multimodal data, comprising:

[0006] The medical data integration module acquires electronic medical record data and medical examination reports through the FHIR interface and extracts structured health indicators;

[0007] The cross-modal causal fusion processing module fuses monitoring data with medical text through image, text, and image mezzanine architecture, generating multi-dimensional health feature vectors under the constraints of medical causal graphs.

[0008] The health status evolution modeling module maps health feature vectors into three-dimensional state indicators of physiological function, cognitive level, and motor ability, and calculates risk time trajectories.

[0009] The retrospective assessment report module analyzes the causal chain of historical health events and generates dynamic assessment reports by combining risk time trajectories.

[0010] The decision output module is used to distribute the evaluation results to medical terminals and family devices.

[0011] Preferably, the cross-modal causal fusion processing module includes:

[0012] The image feature extraction unit is configured to process medical image data through a three-dimensional convolutional network and output an anatomical structure feature matrix.

[0013] The text encoding unit is configured to use the BioBERT model to encode medical record text into semantic vectors.

[0014] The causal constraint fusion unit segments image features into front and back sub-matrices and embeds them into text vectors to form a mezzanine structure. It applies the disease development path constraint feature weight allocation from the medical causal graph to output a multidimensional health feature vector.

[0015] Preferably, the disease development path constraints in the causal constraint fusion unit include:

[0016] Path feature mapping extracts the disease progression path corresponding to the current diagnosis from the medical causal graph. , path nodes Encoded as path feature vector ,in, It is a 300-dimensional embedding vector based on node types, including symptoms, diseases, and complications, and the data source is a medical knowledge graph mapped from the ICD-11 encoding library;

[0017] Attention constraint mechanism to calculate text feature vectors and Cosine similarity is used as a weighting constraint factor: ;

[0018] in, This is a 768-dimensional medical record text vector output by BioBERT. For dynamic weighting factors;

[0019] Weighted fusion and reorganization of the feature matrix:

[0020] in, Slice the first 1 / 3 of the image features. Slice the image into two-thirds slices to capture its features. For the Hadamah accumulation, mandatory constraints are imposed on the current For end-stage heart failure .

[0021] Preferably, the health state evolution modeling module includes:

[0022] The state mapping unit transforms multidimensional health feature vectors into physiological state indices, cognitive state indices, and motor ability indices through linear transformation.

[0023] The degradation analysis unit analyzes the changing trends of various state indices through an aging trajectory model, which includes degradation patterns in three dimensions: physiological, cognitive, and motor.

[0024] The risk calculation unit monitors the rate of decline of the athletic ability index. When the athletic ability index is continuously monitored to accelerate its decline beyond a preset threshold, it generates a fall risk time window warning and establishes a correlation with the extracted recent fluctuation data of body movement.

[0025] Preferably, the degradation analysis unit includes:

[0026] Trend extraction involves analyzing the temporal change patterns of various health indices using convolutional neural networks to extract degenerative feature fragments in physiological, cognitive, and motor dimensions.

[0027] Aging trajectory prediction, based on extracted degenerative features, uses a linear mixed-effects model to predict the evolution curve of health status over the next 90 days.

[0028] Preferably, the retrospective evaluation report module includes:

[0029] The timeline construction unit establishes a unified timeline for the evolution of health status by integrating risk time trajectories and health indicators.

[0030] The causal chain analysis unit, based on medical causal graphs, analyzes the causal relationships between health events and calculates the contribution of each event to the current health risk.

[0031] The visualization reporting engine integrates risk timelines with causal analysis results to generate dynamic assessment reports that include timeline heatmaps, event correlation network diagrams, and risk prediction curves.

[0032] Preferably, the causal chain analysis unit includes:

[0033] Causal graph construction involves generating a causal influence network graph based on medical causal graphs, with the current health event as the root node.

[0034] Contribution quantification is performed by calculating the contribution of each health event to the current risk using the following formula:

[0035] in, Preset weights for disease pathways in the causal graph. The risk change rate provided for the health status evolution modeling module The time interval from the time the event occurred to the present. This is a critical event indicator; it is 1 if it is a critical pathological stage, and 0 otherwise.

[0036] Causal chain generation involves constructing a multi-level causal chain with weighted identifiers to filter event nodes with a contribution greater than 0.15.

[0037] Preferably, the decision output module includes:

[0038] The medical report distribution unit sends the assessment report to the medical terminal. The report includes the causal chain of the health event and the risk time trajectory.

[0039] The family member early warning distribution unit extracts key risk information from the assessment report, generates simplified early warning notifications, and pushes them to family members' devices.

[0040] Preferably, the family member early warning distribution unit includes:

[0041] A voice announcement will be made when the risk level exceeds the threshold of the health indicator.

[0042] Adjust the voice broadcast volume according to the user's age;

[0043] The notification interface displays a risk timeline graph.

[0044] A data processing method for assessing the health status of the elderly based on multimodal data, the method comprising the following steps:

[0045] S1. Obtain electronic medical record data and medical examination reports through the FHIR interface, and extract structured health indicators;

[0046] S2. Employs an image, text, and image-mezzanine architecture to fuse monitoring data with medical text, generating multi-dimensional health feature vectors under the constraints of a medical causal graph.

[0047] S3. Map the health feature vectors to three-dimensional state indicators of physiological function, cognitive level and motor ability, and calculate the risk time trajectory;

[0048] S4. Analyze the causal chain of historical health events and generate a dynamic assessment report by combining the risk time trajectory;

[0049] S5. Distribute the assessment results to medical terminals and family devices.

[0050] This invention provides a data processing system for assessing the health status of the elderly based on multimodal data. It has the following beneficial effects:

[0051] 1. This invention achieves deep semantic fusion of multimodal features under the constraints of medical pathology rules through an image, text and image sandwich architecture. Based on the dynamic guidance of medical causal graph, feature weights are adaptively allocated to generate a high-dimensional fusion vector that retains key pathological information, which significantly improves the semantic integration capability of health data and enhances the reliability of discrimination and decision-making.

[0052] 2. This invention establishes a quantitative indicator system for physiological function, cognitive level and motor ability, and constructs a personalized aging trajectory prediction model by combining the time-series degeneration law. Through the linkage analysis of multi-dimensional state indicators, it accurately captures the critical point of health risk and realizes the paradigm shift from static assessment to dynamic prediction.

[0053] 3. This invention constructs a quantifiable contribution analysis model through medical causal mapping, integrates risk time trajectories and historical health events, and generates an assessment report with causal traceability through dynamic calculation of event weights, time decay factors and key pathological markers, thereby enhancing the clinical reliability of the assessment report. Attached Figure Description

[0054] Figure 1 This is an architecture diagram of the elderly health status assessment data processing system based on multimodal data according to the present invention;

[0055] Figure 2 This is a flowchart of the data processing method for assessing the health status of the elderly based on multimodal data according to the present invention. Detailed Implementation

[0056] The technical solution of the present invention will now be clearly and completely described 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.

[0057] Please see the appendix Figure 1 This invention provides a data processing system for assessing the health status of the elderly based on multimodal data, comprising:

[0058] The medical data integration module acquires electronic medical record data and medical examination reports through the FHIR interface and extracts structured health indicators;

[0059] The cross-modal causal fusion processing module fuses monitoring data with medical text through image, text, and image mezzanine architecture, generating multi-dimensional health feature vectors under the constraints of medical causal graphs.

[0060] The health status evolution modeling module maps health feature vectors into three-dimensional state indicators of physiological function, cognitive level, and motor ability, and calculates risk time trajectories.

[0061] The retrospective assessment report module analyzes the causal chain of historical health events and generates dynamic assessment reports by combining risk time trajectories.

[0062] The decision output module is used to distribute the evaluation results to medical terminals and family devices.

[0063] Specifically, the medical data integration module integrates multi-source heterogeneous medical information through a standardized FHIR interface, transforming electronic medical records and medical examination reports into structured health indicators, and providing unified format clinical data input support for subsequent processing modules;

[0064] The cross-modal causal fusion processing module achieves deep feature fusion of multimodal medical data through a sandwich architecture of images, text, and images, under the guidance of medical causal graphs, and generates multi-dimensional comprehensive feature vectors that represent health status, providing a unified high-quality feature expression for subsequent health status evolution modeling;

[0065] The health status evolution modeling module maps multi-dimensional health feature vectors to a three-dimensional indicator system of physiological function, cognitive level and motor ability through a dynamic quantitative model, constructs a health status trajectory based on time series evolution, predicts individualized risk development paths, generates visualized risk time evolution curves, and realizes continuous dynamic monitoring and risk warning of the health status of the elderly.

[0066] The retrospective assessment report module quantifies the causal relationship paths between historical health events, combines the risk time trajectory output by health status evolution modeling, integrates time series characteristics and causal inference results to generate an interactive assessment report, realizes the visual tracking of the health risk evolution process, and provides a comprehensive assessment output with temporal continuity and causal interpretability for medical decision-making.

[0067] The decision output module distributes dynamic evaluation reports to medical terminals and family devices, enabling efficient delivery of evaluation conclusions and providing information output channels that are tailored to the roles of different user groups.

[0068] The cross-modal causal fusion processing module includes:

[0069] The image feature extraction unit is configured to process medical image data through a three-dimensional convolutional network and output an anatomical structure feature matrix.

[0070] The text encoding unit is configured to use the BioBERT model to encode medical record text into semantic vectors.

[0071] The causal constraint fusion unit segments image features into front and back sub-matrices and embeds them into text vectors to form a mezzanine structure. It applies the disease development path constraint feature weight allocation from the medical causal graph to output a multidimensional health feature vector.

[0072] Specifically, the image feature extraction unit analyzes the spatial structure information of medical images through a three-dimensional convolutional neural network, extracts the morphological and texture features of anatomical tissues from multimodal medical images, and transforms the original image data into a high-dimensional anatomical structure representation matrix.

[0073] The text encoding unit uses the BioBERT pre-trained language model to parse the clinical semantics of medical record texts. Through medical entity recognition and relation extraction technology, unstructured text is encoded into a vector that preserves medical semantics.

[0074] The causal constraint fusion unit constructs a sandwich fusion structure of image features and text semantics based on the disease development path defined by the medical causal graph: it decouples image features into different subspaces, embeds pathologically related text semantic vectors, and outputs multidimensional health feature vectors through dynamic weight allocation.

[0075] The disease progression path constraints in the causal constraint fusion unit include:

[0076] Path feature mapping extracts the disease progression path corresponding to the current diagnosis from the medical causal graph. , path nodes Encoded as path feature vector ,in, It is a 300-dimensional embedding vector based on node types, including symptoms, diseases, and complications, and the data source is a medical knowledge graph mapped from the ICD-11 encoding library;

[0077] Attention constraint mechanism to calculate text feature vectors and Cosine similarity is used as a weighting constraint factor: ;

[0078] in, This is a 768-dimensional medical record text vector output by BioBERT. For dynamic weighting factors;

[0079] Weighted fusion and reorganization of the feature matrix:

[0080] in, Slice the first 1 / 3 of the image features. Slice the image into two-thirds slices to capture its features. For the Hadamah accumulation, mandatory constraints are imposed on the current For end-stage heart failure .

[0081] Specifically, path feature mapping extracts the disease development path sequence of the current diagnosis from the medical causal graph, maps the symptom, disease and complication nodes in the path into embedding vectors, establishes a continuous feature representation of the disease evolution process, and provides structured knowledge guidance for disease development for subsequent feature fusion.

[0082] The attention constraint mechanism calculates the semantic similarity between text features and path features, generates dynamic weight factors that reflect clinical semantic relevance, and achieves automatic alignment between medical record text description and disease development path, ensuring that key pathological features receive reasonable weight allocation during the fusion process.

[0083] The weighted fusion unit uses a mezzanine recombination strategy to decouple image features into front and back feature subspaces and embed constrained text semantic vectors between them. It achieves local weighting of the back features through Hadamard product and enforces a key feature retention mechanism when diagnosing specific pathological stages. Finally, it outputs multidimensional fusion features with clinical interpretability.

[0084] The health status evolution modeling module includes:

[0085] The state mapping unit transforms multidimensional health feature vectors into physiological state indices, cognitive state indices, and motor ability indices through linear transformation.

[0086] The degradation analysis unit analyzes the changing trends of various state indices through the aging trajectory model, which includes degradation patterns in three dimensions: physiological, cognitive, and motor.

[0087] The risk calculation unit monitors the rate of decline of the athletic ability index. When the athletic ability index is continuously monitored to accelerate its decline beyond a preset threshold, it generates a fall risk time window warning and establishes a correlation with the extracted recent fluctuation data of body movement.

[0088] Specifically, the state mapping unit projects multidimensional health feature vectors onto a three-dimensional indicator space of physiological function, cognitive function, and motor ability through linear transformation, establishing a standardized quantitative indicator system for health status and realizing interpretable mapping from comprehensive features to specific health dimensions.

[0089] The degradation analysis unit tracks the evolution trends of physiological, cognitive, and motor state indices based on a predefined aging trajectory model, captures the degradation patterns of each dimension through time series pattern recognition technology, and establishes a dynamic trajectory model of the evolution of health status over time.

[0090] The risk calculation unit continuously monitors the short-term rate of change of the athletic ability index. When it detects that the acceleration decline continuously exceeds the preset warning value, it automatically triggers the fall risk prediction mechanism, generates a risk window warning signal, and cross-validates it with the body movement fluctuation characteristics.

[0091] The degradation analysis unit includes:

[0092] Trend extraction involves analyzing the temporal change patterns of various health indices using convolutional neural networks to extract degenerative feature fragments in physiological, cognitive, and motor dimensions.

[0093] Aging trajectory prediction, based on extracted degenerative features, uses a linear mixed-effects model to predict the evolution curve of health status over the next 90 days.

[0094] Specifically, trend extraction uses convolutional neural networks to automatically identify key change patterns in health index time series data, extract feature segments that characterize physiological degeneration, cognitive decline, and weakened motor function, capture local mutations and trend inflection points in the evolution of health status, and provide refined input features for modeling aging trajectories.

[0095] Aging trajectory prediction is based on a linear mixed-effects model that integrates population degradation patterns with individual-specific parameters. It uses fixed effects to characterize age-related basic decline paths and random effects to calibrate individual health offsets, generating personalized evolution curves of future health status and achieving quantitative prediction of health risks within 90 days.

[0096] The retrospective assessment report module includes:

[0097] The timeline construction unit establishes a unified timeline for the evolution of health status by integrating risk time trajectories and health indicators.

[0098] The causal chain analysis unit, based on medical causal graphs, analyzes the causal relationships between health events and calculates the contribution of each event to the current health risk.

[0099] The visualization reporting engine integrates risk timelines with causal analysis results to generate dynamic assessment reports that include timeline heatmaps, event correlation network diagrams, and risk prediction curves.

[0100] Specifically, the time axis construction unit integrates discrete time point monitoring data into a continuous health status evolution sequence, associates risk trajectories with health indicators under a unified time benchmark, establishes a traceable historical status change map, and achieves precise positioning of key nodes and panoramic visualization of status evolution.

[0101] The causal chain analysis unit calculates the strength of causal effects between health events based on predefined medical pathology association rules, quantifies the contribution weight of each event to the current risk status, identifies the main pathogenic factors and their transmission pathways, and forms a verifiable causal logical chain.

[0102] The visualization report engine integrates the risk timeline and causal analysis results into a heatmap, reveals the interaction effects of multiple events by combining a correlation network diagram, uses prediction curves to depict future risk trends, and outputs an interactive dynamic assessment report with temporal correlation and causal traceability.

[0103] The causal chain analysis unit includes:

[0104] Causal graph construction involves generating a causal influence network graph based on medical causal graphs, with the current health event as the root node.

[0105] Contribution quantification is performed by calculating the contribution of each health event to the current risk using the following formula:

[0106] in, Preset weights for disease pathways in the causal graph. The risk change rate provided for the health status evolution modeling module The time interval from the time the event occurred to the present. This is a critical event indicator; it is 1 if it is a critical pathological stage, and 0 otherwise.

[0107] Causal chain generation involves constructing a multi-level causal chain with weighted identifiers to filter event nodes with a contribution greater than 0.15.

[0108] Specifically, the causal graph construction automatically generates a multi-level causal influence network based on the pathological association rules of medical causal graphs, taking the current health event as the starting point for tracing the source, and constructs an event-driven tree-like topology to realize the visual tracing of the development path of health problems.

[0109] Contribution quantification is achieved by integrating path preset weights, real-time risk change rates, and time decay factors, combined with key pathological stage identifiers, to calculate the quantitative contribution value of each event to the current health risk, and to establish an objective causative factor ranking mechanism.

[0110] Causal chain generation involves filtering core influencing factor nodes based on preset contribution thresholds, constructing multi-level causal chains with weighted labels, and automatically generating simplified causal networks with clear main paths and prominent key nodes.

[0111] The decision output module includes:

[0112] The medical report distribution unit sends the assessment report to the medical terminal. The report includes the causal chain of the health event and the risk timeline.

[0113] The family member early warning distribution unit extracts key risk information from the assessment report, generates simplified early warning notifications, and pushes them to family members' devices.

[0114] Specifically, the medical report distribution unit encapsulates the complete causal chain and risk time trajectory of health events in the dynamic assessment report into a standard medical data format, and transmits it to the medical terminal through a secure communication protocol, providing professional report support for clinical decision-making with multi-dimensional analysis;

[0115] The family member early warning distribution unit reconstructs key risk information requiring urgent intervention from the report into life-oriented early warning instructions, which are then pushed to family member users' devices via mobile communication interfaces to achieve immediate response to family care risks.

[0116] The family early warning distribution unit includes:

[0117] A voice announcement will be made when the risk level exceeds the threshold of the health indicator.

[0118] Adjust the voice broadcast volume according to the user's age;

[0119] The notification interface displays a risk timeline graph.

[0120] Specifically, when a health risk is detected to exceed a safety threshold, a voice alarm system is triggered. The volume is adjusted to ensure clarity by using the user's age parameter. At the same time, a risk trajectory change view is loaded on the notification interface. Multi-dimensional perception channels are established to collaboratively output early warning information, enabling three-dimensional communication and immediate response guidance of key risk situations.

[0121] Please see the appendix Figure 2 A data processing method for assessing the health status of the elderly based on multimodal data, the method comprising the following steps:

[0122] S1. Obtain electronic medical record data and medical examination reports through the FHIR interface, and extract structured health indicators;

[0123] S2. Employs an image, text, and image-mezzanine architecture to fuse monitoring data with medical text, generating multi-dimensional health feature vectors under the constraints of a medical causal graph.

[0124] S3. Map the health feature vectors to three-dimensional state indicators of physiological function, cognitive level and motor ability, and calculate the risk time trajectory;

[0125] S4. Analyze the causal chain of historical health events and generate a dynamic assessment report by combining the risk time trajectory;

[0126] S5. Distribute the assessment results to medical terminals and family devices.

[0127] Specifically, by establishing an analytical foundation through standardized extraction of medical data, and under the constraints of pathological rules, an innovative sandwich architecture is adopted to integrate multimodal features to generate high-dimensional vectors, which are then mapped into three-dimensional quantitative health indicators to predict risk trajectories. Combined with a time-series causal backtracking mechanism, a dynamic assessment model is constructed. Finally, a dual-track distribution strategy is used to achieve the delivery of professional reports and simplified warnings, forming an end-to-end technical closed loop of multi-source medical data collection, fusion, modeling, assessment, and decision-making. This provides dynamic assessment support for elderly health management that combines clinical interpretability and timeliness, significantly improving the accuracy of risk prediction and the effectiveness of intervention measures.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data processing system for assessing the health status of the elderly based on multimodal data, characterized in that, include: The medical data integration module acquires electronic medical record data and medical examination reports through the FHIR interface and extracts structured health indicators; The cross-modal causal fusion processing module fuses monitoring data with medical text through image, text, and image mezzanine architecture, generating multi-dimensional health feature vectors under the constraints of medical causal graphs. The health status evolution modeling module maps health feature vectors into three-dimensional state indicators of physiological function, cognitive level, and motor ability, and calculates risk time trajectories. The retrospective assessment report module analyzes the causal chain of historical health events and generates dynamic assessment reports by combining risk time trajectories. The decision output module is used to distribute the evaluation results to medical terminals and family devices.

2. The elderly health status assessment data processing system based on multimodal data according to claim 1, characterized in that, The cross-modal causal fusion processing module includes: The image feature extraction unit is configured to process medical image data through a three-dimensional convolutional network and output an anatomical structure feature matrix. The text encoding unit is configured to use the BioBERT model to encode medical record text into semantic vectors. The causal constraint fusion unit segments image features into front and back sub-matrices and embeds them into text vectors to form a mezzanine structure. It applies the disease development path constraint feature weight allocation from the medical causal graph to output a multidimensional health feature vector.

3. The elderly health status assessment data processing system based on multimodal data according to claim 2, characterized in that, The disease progression path constraints in the causal constraint fusion unit include: Path feature mapping extracts the disease progression path corresponding to the current diagnosis from the medical causal graph. , path nodes Encoded as path feature vector ,in, It is a 300-dimensional embedding vector based on node types, including symptoms, diseases, and complications, and the data source is a medical knowledge graph mapped from the ICD-11 encoding library; Attention constraint mechanism to calculate text feature vectors and Cosine similarity is used as a weighting constraint factor: ; in, This is a 768-dimensional medical record text vector output by BioBERT. For dynamic weighting factors; Weighted fusion and reorganization of the feature matrix: in, Slice the first 1 / 3 of the image features. Slice the image into two-thirds slices to capture its features. For the Hadamah accumulation, mandatory constraints are imposed on the current For end-stage heart failure .

4. The elderly health status assessment data processing system based on multimodal data according to claim 1, characterized in that, The health state evolution modeling module includes: The state mapping unit transforms multidimensional health feature vectors into physiological state indices, cognitive state indices, and motor ability indices through linear transformation. The degradation analysis unit analyzes the changing trends of various state indices through an aging trajectory model, which includes degradation patterns in three dimensions: physiological, cognitive, and motor. The risk calculation unit monitors the rate of decline of the athletic ability index. When the athletic ability index is continuously monitored to accelerate its decline beyond a preset threshold, it generates a fall risk time window warning and establishes a correlation with the extracted recent fluctuation data of body movement.

5. The elderly health status assessment data processing system based on multimodal data according to claim 4, characterized in that, The degradation analysis unit includes: Trend extraction involves analyzing the temporal change patterns of various health indices using convolutional neural networks to extract degenerative feature fragments in physiological, cognitive, and motor dimensions. Aging trajectory prediction, based on extracted degenerative features, uses a linear mixed-effects model to predict the evolution curve of health status over the next 90 days.

6. The elderly health status assessment data processing system based on multimodal data according to claim 1, characterized in that, The retrospective assessment report module includes: The timeline construction unit establishes a unified timeline for the evolution of health status by integrating risk time trajectories and health indicators. The causal chain analysis unit, based on medical causal graphs, analyzes the causal relationships between health events and calculates the contribution of each event to the current health risk. The visualization reporting engine integrates risk timelines with causal analysis results to generate dynamic assessment reports that include timeline heatmaps, event correlation network diagrams, and risk prediction curves.

7. The elderly health status assessment data processing system based on multimodal data according to claim 6, characterized in that, The causal chain analysis unit includes: Causal graph construction involves generating a causal influence network graph based on medical causal graphs, with the current health event as the root node. Contribution quantification is performed by calculating the contribution of each health event to the current risk using the following formula: in, Preset weights for disease pathways in the causal graph. The risk change rate provided for the health status evolution modeling module The time interval from the time the event occurred to the present. This is a critical event indicator; it is 1 if it is a critical pathological stage, and 0 otherwise. Causal chain generation involves constructing a multi-level causal chain with weighted identifiers to filter event nodes with a contribution greater than 0.

15.

8. The elderly health status assessment data processing system based on multimodal data according to claim 1, characterized in that, The decision output module includes: The medical report distribution unit sends the assessment report to the medical terminal. The report includes the causal chain of the health event and the risk time trajectory. The family member early warning distribution unit extracts key risk information from the assessment report, generates simplified early warning notifications, and pushes them to family members' devices.

9. The elderly health status assessment data processing system based on multimodal data according to claim 8, characterized in that, The family early warning distribution unit includes: A voice announcement will be made when the risk level exceeds the threshold of the health indicator. Adjust the voice broadcast volume according to the user's age; The notification interface displays a risk timeline graph.

10. A data processing method for assessing the health status of the elderly based on multimodal data, characterized in that, The method for the elderly health status assessment data processing system based on multimodal data as described in any one of claims 1-9 includes the following steps: S1. Obtain electronic medical record data and medical examination reports through the FHIR interface, and extract structured health indicators; S2. Employs an image, text, and image-mezzanine architecture to fuse monitoring data with medical text, generating multi-dimensional health feature vectors under the constraints of a medical causal graph. S3. Map the health feature vectors to three-dimensional state indicators of physiological function, cognitive level and motor ability, and calculate the risk time trajectory; S4. Analyze the causal chain of historical health events and generate a dynamic assessment report by combining the risk time trajectory; S5. Distribute the assessment results to medical terminals and family devices.

Images