Cognitive function and body function linkage evaluation and rehabilitation training system for old people

Abstract

The invention relates to the technical field of elderly health intelligence, and discloses an elderly cognitive function and body function linkage evaluation and rehabilitation training system. According to the system, user behavior and physiological signals are obtained through multi-modal perception, and a sequence is formed through behavior slicing and situation marking; mining an individualized linkage mode of cognition and body activities by utilizing situation correlation analysis, and establishing an exclusive function baseline; by dynamically comparing real-time data with a baseline, detecting linkage abnormity and tracing to a specific cognition and body function dimension intersection point; and generating a personalized multi-dimensional rehabilitation path according to the intersection point, and converting the personalized multi-dimensional rehabilitation path into an equipment control instruction to drive an interaction terminal to execute training. According to the scheme, personalized quantitative evaluation of the cognition-body function cooperation mode is realized, and accurate and directional rehabilitation induction and intensive training can be carried out aiming at the identified specific linkage imbalance node.

Description

Technical Field

[0001] This invention relates to the field of intelligent technology for elderly health, specifically to an assessment and rehabilitation training system that links cognitive and physical functions in the elderly. Background Technology

[0002] Traditional geriatric functional assessment and rehabilitation techniques typically treat cognitive and physical functions as separate domains. Current techniques generally employ standardized cognitive scales and separate physical ability tests for assessment, subsequently presenting the results side-by-side or through simple weighted integration. During the rehabilitation training phase, training tasks are primarily designed to address the individual functional deficits identified during screening. This approach, based on the assumption of functional separation, fails to effectively address the inherent synergy between cognition and physical activity in daily life.

[0003] Existing technical solutions have shortcomings. At the assessment level, separate assessments fail to depict the dynamic coordination and mutual influence between cognitive and physical activities in specific contexts. The user baseline established is actually a collection of isolated functional indicators, rather than a holistic outline reflecting their inherent collaborative working pattern. This static, fragmented baseline is ill-suited to sensitively capture early, subtle functional declines caused by imbalances in the linkage mechanism. At the intervention level, training targeting single functions or fixed combinations of training cannot accurately address the core issue leading to a decline in daily living abilities: the breakage or inefficiency of the linkage chain between specific cognitive and physical functional dimensions. Rehabilitation training lacks specificity and is therefore inefficient.

[0004] The purpose of this invention is to solve the problem of how to construct a personalized functional baseline that reflects an individual's cognitive-physical inherent linkage pattern, and on this basis, to achieve precise source tracing and rehabilitation induction for specific linkage dysfunction points. Summary of the Invention

[0005] The purpose of this invention is to provide an assessment and rehabilitation training system for the linkage between cognitive function and physical function in the elderly, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a system for the coordinated assessment and rehabilitation training of cognitive and physical functions in the elderly, the system comprising:

[0007] The multimodal data sensing module is used to acquire the user's behavioral and physiological synchronous signal stream;

[0008] The behavior slicing and context labeling module is used to slice and label the synchronization signal stream to form a continuous sequence of behavior segments with context labels;

[0009] The context association analysis module is used to analyze the continuous behavioral segment sequence through the context association engine, identify the inherent linkage pattern between cognitive activities and physical activities, and generate a user-specific functional baseline profile based on the inherent linkage pattern.

[0010] The dynamic comparison and anomaly tracing module is used to dynamically compare the real-time collected user signal stream with the user-specific functional baseline profile, detect abnormal linkage events that deviate from the functional baseline profile, and perform source tracing and analysis on the abnormal linkage events to locate the intersection of the corresponding cognitive function dimension and physical function dimension.

[0011] The rehabilitation pathway generation and execution module is used to construct a personalized multidimensional rehabilitation induction pathway based on the intersection points, convert the multidimensional rehabilitation induction pathway into a specific device control sequence, and execute the device control sequence through an environmental interaction terminal.

[0012] Preferably, the multimodal data sensing module acquires the user's behavioral and physiological synchronization signal stream in the following manner:

[0013] Non-contact sensing units and wearable sensing units are deployed within the user's activity space. The non-contact sensing units capture the user's overall motion trajectory and interactive object information to generate a spatial behavior data stream. The wearable sensing units collect the user's neurophysiological signals and joint kinematic signals to generate a physiological motion data stream. A hardware timestamp synchronization mechanism is used to align the time references of the spatial behavior data stream and the physiological motion data stream, and fuse them to generate the behavior and physiological synchronization signal stream.

[0014] The hardware timestamp synchronization mechanism is used to align the time bases of the spatial behavior data stream and the physiological movement data stream. Specifically, the method is as follows:

[0015] High-precision clock chips are integrated into the non-contact sensing unit and the wearable sensing unit respectively. At the start of data acquisition, the main controller broadcasts a unified time synchronization signal to all sensing units. Each sensing unit embeds the current timestamp of its local clock chip when generating data packets. At the data fusion node, all data packets are received and the timestamps are extracted. The time offset of each data packet is calculated, and a linear interpolation algorithm is applied to correct the time offset, so that the time reference of the spatial behavior data stream and the physiological motion data stream is aligned to the same time axis.

[0016] Preferably, the behavior slicing and context labeling module performs behavior slicing and context labeling on the synchronization signal stream to form a continuous sequence of behavior segments with context labels, specifically in the following manner:

[0017] Based on the motion energy mutation points in the spatial behavior data stream and the signal mode conversion points in the physiological motion data stream, time boundaries are automatically defined; according to the defined time boundaries, the behavior and physiological synchronization signal streams are cut into discrete behavior slices; a predefined context template is loaded for each behavior slice, the context template being jointly defined by activity type, environmental complexity, and task requirements; multiple behavior slices loaded with context templates are connected in chronological order to form the continuous behavior segment sequence with context labels.

[0018] Preferably, the context association analysis module analyzes the continuous behavioral segment sequence through a context association engine to identify the inherent linkage pattern between cognitive activities and bodily activities. Specifically, it extracts the context label and its contained signal features corresponding to each behavioral slice in the continuous behavioral segment sequence, constructs a multidimensional feature vector field, calculates the covariance relationship matrix between the cognitive feature vector cluster and the bodily feature vector cluster in the multidimensional feature vector field, and parses the stable feature co-occurrence rules and temporal dependence rules from the covariance relationship matrix. The feature co-occurrence rules and the temporal dependence rules together constitute the inherent linkage pattern.

[0019] The specific method for constructing the context association engine is as follows:

[0020] A multi-level feature extractor is configured to extract the semantic features and signal feature statistics of the context label layer by layer from the continuous behavioral segment sequence. A graph neural network model is built, and the semantic features and statistics are used as node attributes to construct a temporal relationship graph between behavioral segments. The graph neural network model is trained to learn the nonlinear mapping relationship between cognitive feature vector clusters and bodily feature vector clusters. A rule inference engine is deployed, a predefined cognitive-bodily association rule library is loaded, and the output of the graph neural network model is logically verified and supplemented to form the context association engine.

[0021] Preferably, the context association analysis module generates a user-specific functional baseline profile based on the inherent linkage pattern. Specifically, it maps the feature co-occurrence rules in the inherent linkage pattern to a standard cognitive-physical function coordinate system to form a functional linkage scatter plot; performs density clustering on the functional linkage scatter plot to determine the core functional linkage regions; and marks the transition probabilities between different core functional linkage regions on the time event axis according to the temporal dependency rules. The core functional linkage regions and their transition probabilities together define the user-specific functional baseline profile.

[0022] The method for performing density clustering on the functional linkage scatter plot to determine the core functional linkage region is as follows:

[0023] Density peak clustering algorithm is used to scan all data points in the functional linkage scatter plot, calculate the local density and relative distance of each data point, select the points with local density higher than the density threshold and relative distance greater than the distance threshold as cluster centers, and assign the remaining data points to the nearest cluster centers to form multiple clusters. The functional linkage range covered by each cluster is defined as the core functional linkage region.

[0024] Based on the aforementioned temporal dependency pattern, the transition probabilities between different core functional linkage areas are marked on the time event axis, specifically in the following manner:

[0025] Extract the core functional linkage region to which adjacent behavioral segments belong in the continuous behavioral segment sequence, count the transfer frequency between regions, calculate the ratio of each transfer frequency to the total number of transfers, obtain the transfer probability matrix, map the transfer probability matrix onto the time event axis, and mark the time and probability value of each transfer.

[0026] Preferably, the dynamic comparison and anomaly tracing module dynamically compares the real-time collected user signal stream with the user-specific functional baseline contour to detect abnormal linkage events that deviate from the functional baseline contour. Specifically, the real-time collected user signal stream is processed into a behavior segment to be compared in real time; the matching distance between the feature vector of the behavior segment to be compared and all core functional linkage areas in the functional baseline contour is calculated; if the minimum matching distance exceeds the adaptive distance threshold, a linkage deviation is determined to have occurred; when the number of consecutive linkage deviations reaches the event threshold, this consecutive linkage deviation is recorded as an abnormal linkage event.

[0027] Preferably, the dynamic comparison and anomaly tracing module performs source analysis on the abnormal linkage event to locate the intersection of its corresponding cognitive function dimension and bodily function dimension. Specifically, it decomposes all the behavioral segments to be compared contained in the abnormal linkage event, extracts the cognitive component feature decay rate and bodily component feature disorder degree in each behavioral segment to be compared; projects the cognitive component feature decay rate onto the cognitive function dimension model to identify the damaged cognitive sub-dimension; projects the bodily component feature disorder degree onto the bodily function dimension model to identify the damaged bodily sub-dimension; the intersection of the damaged cognitive sub-dimension and the damaged bodily sub-dimension is the intersection of the cognitive function dimension and the bodily function dimension.

[0028] The method of projecting the decay rate of the cognitive component features onto the cognitive function dimension model to identify the damaged cognitive sub-dimensions is as follows:

[0029] The cognitive function dimension model predefines multiple cognitive sub-dimensions and their normal feature ranges. The Euclidean distance between the feature decay rate of the cognitive component and the normal feature range of each cognitive sub-dimension is calculated. Cognitive sub-dimensions whose Euclidean distance exceeds the decay threshold are marked as damaged cognitive sub-dimensions.

[0030] The method of projecting the disorder of the body component features onto the body function dimension model to identify the damaged body sub-dimension is as follows:

[0031] The somatic function dimension model predefines multiple somatic sub-dimensions and their normal fluctuation ranges. It calculates the divergence value between the disorder degree of the somatic component features and the normal fluctuation range of each somatic sub-dimension, and marks the somatic sub-dimension with a divergence value exceeding the disorder threshold as the damaged somatic sub-dimension.

[0032] Preferably, the rehabilitation path generation and execution module constructs a personalized multidimensional rehabilitation induction path based on the intersection point. Specifically, it queries a pre-set rehabilitation strategy knowledge base to obtain a set of basic rehabilitation action units associated with the intersection point of the cognitive function dimension and the physical function dimension; assesses the user's current ability boundary conditions and adjusts the difficulty coefficient and load of the set of basic rehabilitation action units; and, based on the principle of neuroplasticity and the theory of motor learning stages, arranges the adjusted rehabilitation action units into a sequence with a specific time rhythm and intensity gradient, which is the personalized multidimensional rehabilitation induction path.

[0033] Preferably, the rehabilitation path generation and execution module transforms the multidimensional rehabilitation induction path into a specific device control sequence. Specifically, it parses each rehabilitation action unit in the multidimensional rehabilitation induction path, decomposing it into parameter requirements for the cognitive stimulation device and movement instructions for the physical training device; establishes a synchronous triggering timing relationship between the parameter requirements and the movement instructions; and encodes the parameter requirements and movement instructions with the synchronous triggering timing relationship into a standardized control protocol data packet to form the device control sequence.

[0034] The specific method for establishing the synchronous triggering timing relationship between the parameter requirements and the motion command is as follows:

[0035] The starting time and parameter adjustment sequence of the cognitive stimulation device are obtained by parsing the parameter requirements. The action sequence and duration of the physical training device are obtained by parsing the movement instructions. A timing coordinator is constructed to align the starting time with the start time of the action sequence. An event trigger is set so that when the cognitive stimulation device reaches a specific parameter value, a signal is immediately sent to the physical training device to trigger the corresponding movement instruction. A time window constraint is defined to ensure that the parameter requirements and the movement instructions are executed synchronously within the specified time window.

[0036] Preferably, the rehabilitation path generation and execution module executes the device control sequence through an environmental interaction terminal. Specifically, the environmental interaction terminal includes a visual projection device, an audio feedback device, and a mechanical guidance device. The visual projection device presents a dynamic visual task scene according to the device control sequence. The audio feedback device provides rhythmic guidance or semantic prompts according to the device control sequence. The mechanical guidance device applies auxiliary or resistance mechanical stimulation according to the device control sequence. The system monitors the user's reaction data under the visual task scene, rhythmic guidance or semantic prompts, and mechanical stimulation in real time, and fine-tunes the data packet content of the subsequent device control sequence accordingly.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] By mining context-labeled behavioral sequences through a context association engine, the system can identify the inherent dynamic temporal and efficacy correlation patterns between specific cognitive processes and concrete physical actions when an individual performs daily activities. The personal functional baseline constructed based on this is essentially a personalized model characterizing how cognition and physical function work together. This transforms the assessment of functional status from discrete, standardized individual scores to continuous monitoring of the dynamic stability of an individual's internal collaborative network. The resulting individualized reference benchmark can more fundamentally and sensitively reflect early signs of functional decline, especially those manifestations that are merely loosening of collaborative relationships rather than a decline in individual abilities.

[0039] By comparing real-time acquired signal streams with individual collaborative models, the system can identify abnormal deviations in the linkage pattern and pinpoint the anomalies to specific intersections of cognitive and physical dimensions through source analysis. The multidimensional rehabilitation induction path generated based on this precise intersection directly targets the specific functional linkage links that have failed. The rehabilitation tasks can thus be designed to simulate and emphasize the re-establishment and consolidation of this specific cognitive-physical synergy link, shifting the training focus from improving the absolute ability of a single function to restoring and optimizing efficient and safe coordination patterns between functions. This intervention method, which directly targets the repair of synergistic failures, is in principle more aligned with the essential need to restore complex daily activities. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating the working principle of the elderly cognitive function and physical function linkage assessment and rehabilitation training system described in this invention.

[0041] Figure 2 A flowchart for the behavior slicing and context labeling module;

[0042] Figure 3 Probability map of the transfer of core functional linkage areas;

[0043] Figure 4 A diagram showing the decay rate of cognitive component features and the identification of damaged sub-dimensions;

[0044] Figure 5 A flowchart for constructing a personalized, multidimensional rehabilitation induction pathway. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 This invention provides a system for assessing and rehabilitating the cognitive and physical functions of the elderly. The system includes: a multimodal data perception module responsible for capturing the user's behavioral and physiological synchronous signal streams during daily activities; a behavior slicing and context labeling module receiving the synchronous signal streams and segmenting them into a sequence of continuous behavioral segments with clear context labels; and a context association analysis module using a built-in context association engine to deeply mine historical behavioral segment sequences, identifying stable and inherent linkage patterns between the user's cognition and physical activities, and generating a functional baseline profile characterizing the user's typical functional state. During the real-time assessment phase, a dynamic comparison and anomaly tracing module continuously compares newly acquired real-time user signals with the aforementioned functional baseline profile, detecting abnormal linkage events deviating from the baseline, and performing source tracing analysis on these abnormal events to locate the intersection of the specific cognitive and physical function dimensions causing the abnormality. Based on this intersection, a rehabilitation path generation and execution module automatically constructs a personalized multidimensional rehabilitation induction path, transforming it into a device control sequence that can be understood and executed by an interactive environmental terminal, thereby achieving coordinated and targeted intervention of the user's cognition and physical functions.

[0047] Example 1: See Figure 2The multimodal data sensing module deploys non-contact and wearable sensing units within the physical space of the user's daily activities. The non-contact sensing units capture the user's spatial behavior data stream, while the wearable sensing units collect the user's physiological motion data stream. A hardware timestamp synchronization mechanism aligns the time base of the two data streams. At the start of data acquisition, the main controller broadcasts a unified time synchronization signal to all sensing units. Each sensing unit embeds the current timestamp of its local high-precision clock chip when generating data packets. At the data fusion node, all data packets are received, the timestamps are extracted, and the time offset is calculated. A linear interpolation algorithm is applied to correct this offset, merging the spatial behavior data stream and the physiological motion data stream into a behavior and physiological synchronization signal stream with a consistent time base.

[0048] The behavior slicing and context labeling module receives the aforementioned behavior and physiological synchronization signal streams and automatically delineates time boundaries based on the motion energy abrupt change points within the spatial behavior data stream substream and the signal mode transition points within the physiological motion data stream substream. According to these time boundaries, the behavior and physiological synchronization signal streams are segmented into discrete behavior slices. A predefined context template, jointly defined by activity type, environmental complexity, and task requirements, is loaded onto each behavior slice. Multiple behavior slices loaded with context templates are connected in their original chronological order to form a continuous sequence of behavior segments with context labels.

[0049] In practical implementation, the multimodal data perception module is deployed within the space where users perform daily activities such as walking indoors, picking up and placing items, or engaging in simple desktop activities. Non-contact sensing units, which can be depth camera arrays or millimeter-wave radars, are deployed within the user's activity space to capture the user's overall motion trajectory and information about objects the user interacts with, generating a spatial behavior data stream containing position coordinates, velocity vectors, and the categories of interacting objects. Wearable sensing units, including EEG signal acquisition devices and inertial measurement units, are worn on the user's body to collect the user's neurophysiological signals and joint kinematic signals, generating a physiological motion data stream containing specific EEG rhythm power, joint angles, and accelerations. A hardware timestamp synchronization mechanism is employed to align the time references of spatial behavior data streams and physiological motion data streams. Specifically, high-precision clock chips are integrated into both the non-contact sensing unit and the wearable sensing unit. At the start of data acquisition, the main controller broadcasts a unified time synchronization signal to all sensing units. Each sensing unit embeds the current timestamp of its local clock chip when generating a data packet containing sensor readings or image frames. At the data fusion node, all data packets from the non-contact sensing unit and the wearable sensing unit are received, and the timestamp within each data packet is extracted. Based on the reference of the high-precision clock chip, the time offset of each data packet relative to the unified starting point is calculated. A linear interpolation algorithm is applied to numerically correct data points with time offsets, aligning the spatial behavior data stream and the physiological motion data stream to the same time axis on a microscopic time scale, and fusing them to generate a strictly time-synchronized behavior and physiological synchronization signal stream.

[0050] In some embodiments, the behavior slicing and context labeling module slices and labels the fused behavior and physiological synchronous signal streams, forming a continuous sequence of behavior segments with context labels. In a specific implementation, time boundaries are automatically defined based on motion energy mutation points in the spatial behavior data stream substream and signal pattern transition points in the physiological motion data stream substream. The displacement change rate of the user's contour in the continuous frames of the spatial behavior data stream is calculated, and when the displacement change rate exceeds a preset threshold, it is marked as a motion energy mutation point. The energy spectrum or joint angle change rate of the EEG signal in the physiological motion data stream is analyzed, and when a significant switch in signal pattern occurs, it is marked as a signal pattern transition point. Based on the defined time boundaries, the continuous behavior and physiological synchronous signal stream is cut into a series of discrete, temporally continuous behavior slices. A predefined context template is loaded for each behavior slice. The predefined context template is defined by three dimensions: activity type, environmental complexity, and task requirements. Activity types include walking, sitting, grasping, and visual search. Environmental complexity is quantified by factors such as the number of obstacles identified in the spatial behavior data stream and the degree of light change. Task requirements are determined by whether the user is assigned to complete a specific instruction. Multiple behavior slices loaded with context templates defined by activity type, environmental complexity, and task requirements are connected in the original time sequence of collection to form a continuous sequence of behavior segments with context labels.

[0051] Optionally, when calculating the point of abrupt change in kinetic energy, the rate of change of displacement can be determined using the following formula:

[0052]

[0053] in: Indicates a point in time kinetic energy, This represents the number of keypoints used to calculate motion energy in the spatial behavior data stream. Indicates a point in time Time Spatial coordinate vectors of key points Indicates a point in time Time Spatial coordinate vectors of key points This represents the calculation of the Euclidean norm of a vector. When When the value exceeds the preset energy threshold, at time point It was identified as a point of sudden change in kinetic energy.

[0054] In some embodiments, when applying a linear interpolation algorithm, the data fusion node of the hardware timestamp synchronization mechanism, based on the timestamp from the wearable sensing unit, performs a data fusion operation. The collected physiological data points are then searched for in the spatial behavior data stream from the non-contact sensing unit to find the timestamps closest to those in the data. The two data points before and after, and a value calculated based on the time ratio. Aligned virtual space behavior data points. In practice, the time synchronization signal broadcast by the main controller achieves microsecond-level accuracy, and the drift error of the high-precision clock chip is controlled within the millisecond range per day to ensure the time alignment accuracy of multimodal data during long-term acquisition.

[0055] Optionally, when loading a context template, the behavior slice and context labeling module matches features extracted from the behavior slice with a predefined rule base. For example, if the spatial behavior data stream in the behavior slice shows that the user's center of mass moves periodically in the horizontal plane, and the physiological motion data stream shows that the knee joint angle changes periodically, then the activity type is matched as "walking". Environmental complexity is dynamically assigned based on the number of objects detected by the depth camera around the user within the same time period. Task requirements are determined based on whether clear action command keywords are identified from the audio stream.

[0056] Example 2: The context association analysis module constructs a context association engine. It is configured with a multi-level feature extractor to extract the semantic and signal feature statistics of context labels from a continuous sequence of behavioral segments layer by layer. A graph neural network model is built, using the aforementioned semantic features and statistics as node attributes to construct a temporal relationship graph between behavioral segments. This graph neural network model is trained to learn the nonlinear mapping relationship between cognitive feature vector clusters and bodily feature vector clusters. A rule-based inference engine is deployed, loading a predefined cognitive-bodily association rule library to logically verify and supplement the output of the graph neural network model, forming a complete context association engine. This context association engine analyzes the continuous sequence of behavioral segments, extracting the context label and its signal features for each behavioral slice to construct a multi-dimensional feature vector field. It calculates the covariance relationship matrix between the cognitive and bodily feature vector clusters in this field, and parses out stable feature co-occurrence rules and temporal dependency patterns from this covariance relationship matrix. These feature co-occurrence rules and temporal dependency patterns together constitute an inherent linkage pattern.

[0057] The contextual association analysis module generates a user-specific functional baseline profile based on inherent linkage patterns. It maps the feature co-occurrence rules of these patterns onto a standard cognitive-physical function coordinate system to form a functional linkage scatter plot. A density peak clustering algorithm is used to scan all data points in the functional linkage scatter plot, calculating the local density and relative distance of each data point. Points with local densities exceeding a preset density threshold and relative distances exceeding a preset distance threshold are selected as cluster centers. The remaining data points are assigned to the nearest cluster centers to form multiple clusters. The functional linkage range covered by each cluster is defined as the core functional linkage region. Based on temporal dependency patterns, the transition probabilities between different core functional linkage regions are marked on the time event axis. The transition frequencies of core functional linkage regions belonging to adjacent behavioral segments in a continuous behavioral segment sequence are statistically analyzed. The ratio of the transition frequency to the total number of transitions is calculated to obtain a transition probability matrix. This transition probability matrix is ​​mapped onto the time event axis, marking the time and probability value of each transition. The core functional linkage regions and their transition probabilities together define the user-specific functional baseline profile.

[0058] In its implementation, the context association analysis module analyzes continuous behavioral segment sequences with context labels by constructing a context association engine to identify the inherent linkage patterns between cognitive and physical activities. The context association engine is constructed by configuring a multi-level feature extractor. The first layer of the multi-level feature extractor extracts semantic features from the context labels of each behavioral slice in the continuous behavioral segment sequence, and these semantic features are converted into word vector representations. The second layer of the multi-level feature extractor extracts statistical measures of signal features from the behavioral and physiological synchronous signal streams contained in the behavioral slices. These statistical measures include mean, variance, and entropy. A graph neural network model is built, using the statistical measures of semantic and signal features as attributes of nodes in the graph neural network model. Temporal relationship edges between behavioral segments are constructed based on the temporal order of the continuous behavioral segment sequence, forming a temporal relationship graph. The graph neural network model is trained to learn the nonlinear mapping relationship between cognitive feature vector clusters and physical feature vector clusters. The cognitive feature vector clusters consist of statistical measures of signal features related to attention and memory, while the physical feature vector clusters consist of statistical measures of signal features related to balance and coordination.

[0059] A rule-based inference engine is deployed, loading a predefined cognitive-physical association rule base. This base contains domain knowledge rules such as "gait variability increases when performing complex decision-making tasks." The inference engine performs logical verification and supplementation on the output of the graph neural network model, forming a complete contextual association engine. The contextual association engine analyzes continuous behavioral segments, extracting the contextual label and its contained signal features corresponding to each behavioral slice. A multidimensional feature vector field with context and time dimensions is constructed. The covariance matrix between the cognitive and physical feature vector clusters in the multidimensional feature vector field is calculated. Stable feature co-occurrence rules and temporal dependencies are extracted from the covariance matrix. Examples of co-occurrence rules include "high-load memory tasks co-occur with an increased power pattern of theta waves in the prefrontal cortex of EEG when gait speed decreases." Examples of temporal dependencies include "within three seconds after completing a visual search task, the joint motion error rate of turning movements significantly decreases." These co-occurrence rules and temporal dependencies together constitute the user's stable, inherent interaction pattern in specific life situations.

[0060] In some embodiments, the contextual association analysis module generates a user-specific functional baseline profile based on the identified inherent linkage patterns. In a specific implementation, the feature co-occurrence rules in the inherent linkage patterns are mapped to a two-dimensional standard cognitive-physical function coordinate system with "executive function-balance ability" as the axis, forming a functional linkage scatter plot. Each point in the scatter plot represents the linkage state of cognitive and physical functions in a behavioral segment. Density clustering is performed on the functional linkage scatter plot to determine the core functional linkage region. A density peak clustering algorithm is used to scan all data points in the functional linkage scatter plot, calculating the local density and relative distance of each data point. Points with local density higher than a preset density threshold and relative distance greater than a preset distance threshold are selected as cluster centers. The remaining data points are assigned to the nearest cluster center, forming multiple clusters. The functional linkage range covered by each cluster in the standard cognitive-physical function coordinate system is defined as a core functional linkage region. For example, a core functional linkage region may cover a set of linkage state points of "moderate executive function demand and good balance control". Based on the temporal dependency patterns in the inherent linkage mode, the transition probabilities between different core functional linkage areas are marked on the time event axis. The core functional linkage areas to which adjacent behavioral segments belong in a continuous sequence of behavioral segments are extracted. The transition frequencies between these areas are statistically analyzed, and the ratio of each transition frequency to the total number of transitions is calculated to obtain a transition probability matrix describing the state transition patterns. This transition probability matrix is ​​then mapped onto the time event axis, marking the time and probability value of each transition. The core functional linkage areas and their transition probabilities together define the user-specific functional baseline profile. This functional baseline profile quantitatively describes the normal space and dynamic evolution patterns of the user's cognitive and bodily functional linkage under normal conditions without abnormal interference.

[0061] Optionally, when using the density peak clustering algorithm, data points Local density Calculated using the following formula:

[0062]

[0063] in: Representing data points Local density, Representing data points With data points The Euclidean distance between them It is a preset cutoff distance. It is an indicator function, when hour ,otherwise Data points relative distance Defined as To all local densities higher than The minimum distance between data points. Local density. and relative distance Points with higher average values ​​were selected as cluster centers.

[0064] See Figure 3 This is a time-series chart showing the probability of transfer in core functional linkage areas, used to illustrate the dynamic transfer patterns of different core functional areas in the cognitive-physical function linkage assessment of the elderly. This chart corresponds to the "functional baseline profile construction" stage in the cognitive-physical function linkage assessment of the elderly. The probability of transfer in core functional areas is a key component of the "user-specific functional baseline profile," reflecting the dynamic evolution of cognitive-physical function linkage; the peak transfer probability between 20 and 35 minutes may correspond to the switching period of the user's activity state. The functions of this type of chart include quantifying the normal transfer patterns of the user's cognitive-physical function linkage, providing a baseline reference for the detection of abnormal linkage events; assisting in the analysis of the temporal characteristics of the user's activity state, optimizing the rhythm and intensity of rehabilitation training; and verifying the rationality of the functional baseline profile, ensuring that the assessment results conform to the user's actual behavioral patterns.

[0065] Example 3: The dynamic comparison and anomaly tracing module processes the real-time collected user signal stream into a behavior segment to be compared, and calculates the matching distance between the feature vector of the behavior segment and all core functional linkage areas in the functional baseline contour. If the minimum matching distance exceeds the adaptive distance threshold dynamically calculated based on user historical data, a linkage deviation is determined to have occurred. When the number of consecutive linkage deviations reaches a preset event threshold, this consecutive linkage deviation is recorded as an abnormal linkage event.

[0066] The dynamic comparison and anomaly tracing module performs source analysis on abnormal linkage events, locating the intersection of the corresponding cognitive function dimension and physical function dimension. It decomposes all comparison behavior segments contained in the abnormal linkage event, extracting the cognitive component feature decay rate and physical component feature disorder degree from each comparison behavior segment. The cognitive component feature decay rate is projected onto the cognitive function dimension model, which predefines multiple cognitive sub-dimensions and their normal feature ranges. The Euclidean distance between the cognitive component feature decay rate and the normal feature range of each cognitive sub-dimension is calculated, and cognitive sub-dimensions with Euclidean distances exceeding a preset decay threshold are marked as damaged cognitive sub-dimensions. Similarly, the physical component feature disorder degree is projected onto the physical function dimension model, which predefines multiple physical sub-dimensions and their normal fluctuation ranges. The divergence value between the physical component feature disorder degree and the normal fluctuation range of each physical sub-dimension is calculated, and physical sub-dimensions with divergence values ​​exceeding a preset disorder threshold are marked as damaged physical sub-dimensions. The intersection of the damaged cognitive sub-dimensions and the damaged physical sub-dimensions is the intersection of the cognitive function dimension and the physical function dimension.

[0067] In practical implementation, the dynamic comparison and anomaly tracing module dynamically compares the real-time collected user signal stream with the user-specific functional baseline profile and detects abnormal linkage events that deviate from the functional baseline profile. Specifically, the module receives the real-time user signal stream from the multimodal data perception module and, following the same logic as the behavior slicing and context labeling module, processes the real-time user signal stream into behavior segments to be compared. Each behavior segment carries a feature vector extracted from the real-time signal. The matching distance between the feature vector of the behavior segment and all core functional linkage regions in the functional baseline profile is calculated. The matching distance is obtained by calculating the Mahalanobis distance between the feature vector of the behavior segment and the feature vector of the cluster center of each core functional linkage region. If the calculated minimum matching distance exceeds the adaptive distance threshold, it is determined that the behavior segment has experienced a linkage deviation. The adaptive distance threshold is dynamically calculated based on the distribution of matching distances between the user's historical behavior segments and the functional baseline profile, for example, set to twice the standard deviation of the historical average matching distance. When the number of consecutive deviations in the interaction reaches a preset event threshold, such as five consecutive behavioral segments, this consecutive deviation is recorded as an abnormal interaction event. An abnormal interaction event indicates that the user's cognitive and bodily function interaction pattern has significantly deviated from their normal state over a sustained period of time.

[0068] In some embodiments, the dynamic comparison and anomaly tracing module performs source analysis on abnormal linkage events, locating the intersection of the corresponding cognitive function dimension and physical function dimension. Specifically, it decomposes all comparison behavior segments contained in the abnormal linkage event, extracting the cognitive component feature decay rate and physical component feature disorder degree from each comparison behavior segment. The cognitive component feature decay rate is calculated by comparing the decrease in real-time cognitive feature values ​​with the corresponding cognitive feature baseline values ​​in the functional baseline profile. The physical component feature disorder degree is obtained by analyzing the complexity and randomness of the real-time physical feature sequence, such as calculating approximate entropy or detrended fluctuation analysis scaling index, and comparing it with the ordered pattern in the baseline state. The cognitive component feature decay rate is projected onto the cognitive function dimension model, which predefines multiple cognitive sub-dimensions and their normal feature ranges, including attention, working memory, executive function, and visuospatial ability. The Euclidean distance between the cognitive component feature decay rate and the normal feature range of each cognitive sub-dimension is calculated, and cognitive sub-dimensions with Euclidean distances exceeding a preset decay threshold are marked as damaged cognitive sub-dimensions. The disorder of bodily components is projected onto a bodily function dimension model, which predefines multiple bodily sub-dimensions, including balance, mobility, fine motor skills, and gross motor coordination, along with their normal fluctuation ranges. The divergence value between the disorder of bodily components and the normal fluctuation range of each bodily sub-dimension is calculated using Jensen-Shannon divergence. Bodily sub-dimensions with divergence values ​​exceeding a preset disorder threshold are marked as damaged bodily sub-dimensions. The intersection of damaged cognitive and bodily sub-dimensions represents the intersection point between cognitive and bodily function dimensions that leads to abnormal linkage events, such as the intersection of executive function and balance.

[0069] In calculating the decay rate of cognitive component features When this is the case, the following formula can be used:

[0070]

[0071] in: Indicates the decay rate of cognitive component features. This represents the cognitive feature baseline value extracted from the functional baseline profile that matches the current context of the behavioral segment to be compared. This represents the cognitive feature value extracted in real time from the behavioral segment to be compared. When The larger the value, the more significant the cognitive feature decay.

[0072] In some embodiments, a specific example scenario is a user getting up from the sofa indoors to walk to the kitchen to get a water cup. When the user gets up from the sofa, the real-time user signal stream is processed into a behavior segment to be compared. The matching distance between this behavior segment and the core functional linkage area of ​​"getting up-walking transition" in the functional baseline contour is calculated. When the user's functional state is normal, the matching distance is usually lower than the adaptive distance threshold. In an abnormal event, the user hesitated and exhibited slow gait initiation when getting up, and the calculated matching distance consistently exceeded the adaptive distance threshold. Five consecutive behavior segments triggered linkage deviation judgment, thus recording an abnormal linkage event related to "posture transition and initiation". Source analysis of this abnormal linkage event revealed that the decay rate of the extracted cognitive component features showed that the "executive function" sub-dimensional features were significantly lower than their normal feature range, and the disorder of the body component features showed that the fluctuations of the "balance function" and "mobility ability" sub-dimensional features exceeded their normal fluctuation range. After calculating the Euclidean distance and divergence value, "executive function" was determined to be a damaged cognitive sub-dimensional, and "balance function" was determined to be a damaged body sub-dimensional. The intersection of the damaged cognitive sub-dimension "executive function" and the damaged bodily sub-dimension "balance function" is located as the intersection of the cognitive function dimension and the bodily function dimension corresponding to this abnormal linkage event.

[0073] See Figure 4 This is a chart analyzing the feature decay rate and Euclidean distance of cognitive sub-dimensions, used to identify impaired cognitive sub-dimensions in tracing the causes of cognitive dysfunction in the elderly. The feature decay rate (approximately 32%) and Euclidean distance (approximately 6) of executive function are significantly higher than other dimensions; the Euclidean distance for working memory (approximately 3.5) slightly exceeds the threshold, while the remaining dimensions are all below the threshold. This chart corresponds to the tracing process of cognitive-physical dysfunction in the elderly: Euclidean distance is used to quantify the deviation of real-time features from the normal range; exceeding the threshold (3.0) indicates a "damaged cognitive sub-dimension." The functions of such charts include accurately locating impaired cognitive sub-dimensions; providing targeted evidence for the construction of subsequent rehabilitation pathways; quantifying the degree of cognitive dysfunction; and assisting in the objectivity and traceability of clinical assessments.

[0074] Example 4: See Figure 5 The rehabilitation pathway generation and execution module constructs a personalized, multidimensional rehabilitation induction pathway based on the identified intersection of cognitive and physical function dimensions. It queries a pre-built rehabilitation strategy knowledge base to retrieve a set of basic rehabilitation movement units associated with that intersection. The module assesses the user's current ability boundaries and adjusts the difficulty and workload of these basic rehabilitation movement units. Based on the principles of neuroplasticity and the theory of motor learning stages, the adjusted rehabilitation movement units are arranged into a sequence with a specific temporal rhythm and intensity gradient; this sequence constitutes the personalized, multidimensional rehabilitation induction pathway.

[0075] The rehabilitation pathway generation and execution module transforms the multidimensional rehabilitation induction pathway into a specific device control sequence. It analyzes each rehabilitation action unit within the pathway, decomposing it into parameter requirements for the cognitive stimulation device and movement commands for the physical training device. A synchronous triggering timing relationship is established between the parameter requirements and the movement commands. The parameter requirements are analyzed to obtain the start time point and parameter adjustment sequence of the cognitive stimulation device, while the movement commands are analyzed to obtain the movement sequence and duration of the physical training device. A timing coordinator is constructed to align the start time point with the start time of the movement sequence. Event triggers are set so that when the cognitive stimulation device reaches a specific parameter value, a signal is immediately sent to the physical training device to trigger the corresponding movement command. Time window constraints are defined to ensure that the parameter requirements and movement commands are executed synchronously within a specified time window. The parameter requirements and movement commands with synchronous triggering timing relationships are encoded into standardized control protocol data packets, forming the device control sequence.

[0076] In practice, the rehabilitation pathway generation and execution module constructs a personalized, multidimensional rehabilitation induction pathway based on the intersection of cognitive and physical function dimensions located by the dynamic comparison and anomaly tracing module. The module queries a pre-set rehabilitation strategy knowledge base to obtain a set of basic rehabilitation movement units associated with the intersection of cognitive and physical function dimensions. For example, for the intersection of "executive function and balance function," the rehabilitation strategy knowledge base might return basic rehabilitation movement units such as "dual-task walking," "weight transfer under visual tracking," and "variable rhythm gait training." The module assesses the user's current ability boundary conditions, including recent maximum heart rate, joint range of motion limits, and baseline accuracy of cognitive tasks, and adjusts the difficulty and workload of a set of basic rehabilitation movement units. For instance, the difficulty of "dual-task walking" is reduced from simultaneous mental arithmetic to simultaneous color recognition, and the support surface for "weight transfer under visual tracking" is adjusted from an unstable interface to a stable interface with added head movement. Based on the principles of neuroplasticity and the theory of motor learning stages, the adjusted rehabilitation movement units are arranged into sequences with specific time rhythms and intensity gradients. The sequences follow the principles of moving from single tasks to dual tasks, from stable environments to mild disturbances, and from low loads to progressive loads. This arranged sequence is the personalized multidimensional rehabilitation induction pathway.

[0077] In some embodiments, the rehabilitation path generation and execution module transforms a personalized multidimensional rehabilitation induction path into a specific device control sequence. In a specific implementation, each rehabilitation action unit in the multidimensional rehabilitation induction path is analyzed and decomposed into parameter requirements for the cognitive stimulation device and movement instructions for the physical training device. For a rehabilitation action unit named "gait crossing under auditory memory interference," the decomposed parameter requirements include "the cognitive stimulation device presents an audio sequence: length 5 digits, playback rate 1 digit / second," and the decomposed movement instructions include "the physical training device starts, speed set to 0.4 m / s, and a simulated obstacle with a height of 5 cm rises 3 seconds after the instruction is issued." A synchronous triggering timing relationship is established between the parameter requirements and the movement instructions. The parameter requirements are analyzed to obtain the start time point and parameter adjustment sequence of the cognitive stimulation device, and the movement instructions are analyzed to obtain the action sequence and duration of the physical training device. A timing coordinator is constructed to align the start time point of the cognitive stimulation device playing the audio sequence with the start time of the physical training device starting to run at a speed of 0.4 m / s. Configure an event trigger so that when the audio sequence from the cognitive stimulation device plays the third number, the trigger immediately sends a signal to the physical training device to trigger a motion command to raise a simulated obstacle. Define a time window constraint to ensure that the parameter requirements of the cognitive stimulation device and the motion commands of the physical training device are executed synchronously within a specified time window of approximately 10 seconds. Encode the parameter requirements and motion commands with synchronous triggering timing relationships into a standardized control protocol data packet, such as using JSON format, containing the device ID, command type, parameter value, and trigger timestamp, forming a device control sequence.

[0078] Optionally, when assessing the user's current ability boundary conditions and adjusting the load of the basic rehabilitation movement units, the adjusted training load can be calculated using the following formula. :

[0079]

[0080] in: This indicates the adjusted training load. This represents the nominal workload of basic rehabilitation movement units in the rehabilitation strategy knowledge base. This represents the decay rate of the relevant functional dimensions calculated from abnormal linkage events. It is the load attenuation coefficient. It is the basic load constant. and Based on individual user differences and preset recovery stages.

[0081] It is understandable that the process of constructing a rehabilitation induction path using the rehabilitation path generation and execution module is a personalized adaptation process based on rules and theories. Its core is to transform abstract, qualitative functional deficit intersections into specific, quantifiable, and executable action sequences that conform to rehabilitation principles. The adaptation process considers the user's real-time capability boundaries, avoiding overload or underload of training. It is also understandable that transforming the rehabilitation induction path into a device control sequence and establishing a precise synchronous triggering time relationship is key to achieving cognitive and physical linkage training. This synchronization ensures precise temporal coupling between cognitive tasks and physical movements, simulating real-world scenarios requiring coordinated cognitive and physical functions, thus potentially producing targeted intervention effects on identified abnormal linkage patterns. Refer to Table 1, the association table between basic rehabilitation action units and functional intersections in the rehabilitation strategy knowledge base.

[0082] Table 1: Partial Association Table of the Rehabilitation Strategy Knowledge Base

[0083] Intersection of cognitive function dimension and physical function dimension Related basic rehabilitation movement units Main training objectives Execution function and balancing function Dual-task walking (simple calculations are performed while walking) Improve the ability to maintain postural stability under cognitive load Working memory and mobility Delayed recall navigation (performs multi-step movement after remembering instructions) Improve path planning and execution capabilities under memory load Visual spatial ability and fine motor skills Visual pattern tracking and grasping (tracking and grasping moving targets) Enhance hand-eye coordination and visually guided fine motor skills

[0084] Example 5: The rehabilitation pathway generation and execution module executes the device control sequence through an environmental interaction terminal, which includes a visual projection device, an acoustic feedback device, and a mechanical guidance device. The visual projection device presents a dynamically changing visual task scene based on the parameters for cognitive stimulation in the device control sequence. The acoustic feedback device provides rhythmic guidance or semantic prompts synchronized with the task, based on the instructions in the device control sequence. The mechanical guidance device applies assistive or resistive mechanical stimulation based on the movement instructions for physical training in the device control sequence. The system monitors the user's response data in real time under the visual task scene, rhythmic guidance or semantic prompts, and mechanical stimulation, and fine-tunes the data packet content of subsequent device control sequences based on this real-time response data.

[0085] In practice, the rehabilitation pathway generation and execution module executes device control sequences through an environmental interaction terminal. This terminal includes a visual projection device, an acoustic feedback device, and a mechanical guidance device. The visual projection device presents a dynamic visual task scene based on the visual task parameters encoded in the device control sequence. The acoustic feedback device provides rhythmic guidance or semantic cues based on the audio instructions encoded in the device control sequence. The mechanical guidance device applies assistive or resistive mechanical stimulation based on the movement instructions encoded in the device control sequence. The rehabilitation pathway generation and execution module parses the device control sequence into control command streams for different devices and synchronously sends these streams to the visual projection device, acoustic feedback device, and mechanical guidance device. Upon receiving the command streams, the visual projection device generates a dynamically changing visual scene in the projection area in front of the user, such as a moving target, virtual obstacles to avoid, or a rapidly flashing pattern sequence to be recognized. The acoustic feedback device outputs rhythmic cues, voice commands, or disruptive auditory stimuli synchronized with the visual task through a speaker or bone conduction headphones. Mechanical guidance devices, such as exoskeletons worn on the lower limbs or damping adjustment systems integrated into smart treadmills, provide appropriate assistance at key phases of the user's limb movement according to the command flow to help complete the movement, or apply adjustable resistance to increase the training load.

[0086] In some embodiments, the visual projection device presents a dynamic visual task scene according to a device control sequence. The device control sequence contains precise parameters regarding the type of visual stimulus, its location, movement trajectory, duration, and rhythm. In a specific implementation, for a rehabilitation movement unit designed to train the intersection of working memory and mobility, the device control sequence might instruct the visual projection device to sequentially illuminate three number patterns at different locations on the floor, requiring the user to memorize and step on them in order. The visual projection device interprets the device control sequence, first projecting the number "5" at location A, which remains illuminated for 2 seconds before turning off; after a 1-second interval, it projects the number "2" at location B; and after another 1-second interval, it projects the number "8" at location C. The visual projection device uses a high refresh rate projection device to ensure clear, lag-free images, and the presentation of the dynamic visual task scene strictly follows the timeline and spatial coordinates defined in the device control sequence. An audio feedback device, based on the same device control sequence, might issue a "start memorizing" voice prompt when the first number is presented, and provide rhythmic "tick-tock" sounds during the user's movement to help control their step frequency. The mechanical guidance device, according to the equipment control sequence, applies a momentary, small-amplitude auxiliary torque at the hip joint via the exoskeleton to facilitate initiation if it detects slow initiation or unsteady gait during the user's movement from position A to position B.

[0087] Optionally, a performance-based dynamic adjustment strategy can be adopted when monitoring user responses in real time and fine-tuning subsequent device control sequences. The system captures new signal streams from the user during training in real time through a multimodal data sensing module and quickly calculates the completion quality index of the current rehabilitation action unit. Achieve quality targets The calculation formula is:

[0088]

[0089] in: This indicates the real-time completion quality index of the current rehabilitation movement unit. This represents the accuracy of the cognitive task decoded from real-time signals. This represents the error rate in task execution related to bodily movements. and These are the weighting coefficients for the cognitive and physical components, respectively. The system will calculate in real time. Compare with the preset expected performance threshold.

[0090] In some embodiments, the system monitors user response data in real time under visual task scenarios, rhythmic guidance or semantic cues, and mechanical stimulation, and fine-tunes the data packet content of subsequent device control sequences accordingly. In a specific implementation, when a user performs the aforementioned "memorizing pedaling numbers" task, the system analyzes the accuracy of the user's pedaling, reaction time, and gait stability in real time. If the system detects that the user completes the sequence accurately and quickly twice consecutively, the system achieves a satisfactory performance evaluation. If the threshold is exceeded, the system will increase the task difficulty in subsequent device control sequence data packets, such as increasing the number of digits to be memorized from 3 to 4, or shortening the duration of digit presentation. If a user's stamping error or significantly prolonged reaction time is detected, the quality indicators will be assessed. If the difficulty falls below the threshold, the system reduces the difficulty of subsequent data packets, for example, by reducing the number sequence to two or providing clearer voice prompts through the audio feedback device. The assistance level of the biomechanical guidance device is also dynamically fine-tuned based on the user's physical stability, reducing assistance when the user's posture is stable and instantly increasing assistance when a risk of imbalance is detected. This fine-tuning is continuous and automatic, allowing the rehabilitation induction path to adapt to the user's real-time performance.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0092] 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 system for the integrated assessment and rehabilitation training of cognitive and physical functions in the elderly, characterized in that, The system includes: The multimodal data sensing module is used to acquire the user's behavioral and physiological synchronous signal stream; The behavior slicing and context labeling module is used to slice and label the synchronization signal stream to form a continuous sequence of behavior segments with context labels; The context association analysis module is used to analyze the continuous behavioral segment sequence through the context association engine, identify the inherent linkage pattern between cognitive activities and physical activities, and generate a user-specific functional baseline profile based on the inherent linkage pattern. The dynamic comparison and anomaly tracing module is used to dynamically compare the real-time collected user signal stream with the user-specific functional baseline profile, detect abnormal linkage events that deviate from the functional baseline profile, and perform source tracing and analysis on the abnormal linkage events to locate the intersection of the corresponding cognitive function dimension and physical function dimension. The rehabilitation pathway generation and execution module is used to construct a personalized multidimensional rehabilitation induction pathway based on the intersection points, convert the multidimensional rehabilitation induction pathway into a specific device control sequence, and execute the device control sequence through an environmental interaction terminal.

2. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 1, characterized in that, The multimodal data sensing module acquires the user's behavioral and physiological synchronous signal stream in the following way: Non-contact sensing units and wearable sensing units are deployed within the user's activity space; the non-contact sensing units capture the user's overall movement trajectory and interaction object information, generating a spatial behavior data stream; The wearable sensing unit collects the user's neurophysiological signals and joint kinematic signals to generate a physiological motion data stream; A hardware timestamp synchronization mechanism is used to align the time references of the spatial behavior data stream and the physiological motion data stream, and then fuse them to generate the behavior and physiological synchronization signal stream. The hardware timestamp synchronization mechanism is used to align the time bases of the spatial behavior data stream and the physiological movement data stream. Specifically, the method is as follows: High-precision clock chips are integrated into the non-contact sensing unit and the wearable sensing unit respectively. At the start of data acquisition, the main controller broadcasts a unified time synchronization signal to all sensing units. Each sensing unit embeds the current timestamp of its local clock chip when generating data packets. At the data fusion node, all data packets are received and the timestamps are extracted. The time offset of each data packet is calculated, and a linear interpolation algorithm is applied to correct the time offset, so that the time reference of the spatial behavior data stream and the physiological motion data stream is aligned to the same time axis.

3. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 2, characterized in that, The behavior slicing and context labeling module performs behavior slicing and context labeling on the synchronization signal stream to form a continuous sequence of behavior segments with context labels, specifically as follows: Based on the motion energy mutation points in the spatial behavior data stream and the signal mode conversion points in the physiological motion data stream, time boundaries are automatically defined; according to the defined time boundaries, the behavior and physiological synchronization signal streams are cut into discrete behavior slices; a predefined context template is loaded for each behavior slice, the context template being jointly defined by activity type, environmental complexity, and task requirements; multiple behavior slices loaded with context templates are connected in chronological order to form the continuous behavior segment sequence with context labels.

4. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 1, characterized in that, The context association analysis module analyzes the continuous behavioral segment sequence through a context association engine to identify the inherent linkage pattern between cognitive activities and physical activities. Specifically, it extracts the context label and its signal features corresponding to each behavioral slice in the continuous behavioral segment sequence, constructs a multidimensional feature vector field, calculates the covariance relationship matrix between the cognitive feature vector cluster and the physical feature vector cluster in the multidimensional feature vector field, and parses the stable feature co-occurrence rules and temporal dependence rules from the covariance relationship matrix. The feature co-occurrence rules and the temporal dependence rules together constitute the inherent linkage pattern. The specific method for constructing the context association engine is as follows: A multi-level feature extractor is configured to extract the semantic features and signal feature statistics of the context label layer by layer from the continuous behavioral segment sequence. A graph neural network model is built, and the semantic features and statistics are used as node attributes to construct a temporal relationship graph between behavioral segments. The graph neural network model is trained to learn the nonlinear mapping relationship between cognitive feature vector clusters and bodily feature vector clusters. A rule inference engine is deployed, a predefined cognitive-bodily association rule library is loaded, and the output of the graph neural network model is logically verified and supplemented to form the context association engine.

5. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 4, characterized in that, The contextual association analysis module generates a user-specific functional baseline profile based on the inherent linkage pattern. Specifically, it maps the feature co-occurrence rules in the inherent linkage pattern to a standard cognitive-physical function coordinate system to form a functional linkage scatter plot; it performs density clustering on the functional linkage scatter plot to determine the core functional linkage regions; based on the temporal dependency rules, it marks the transition probabilities between different core functional linkage regions on the time event axis; the core functional linkage regions and their transition probabilities together define the user-specific functional baseline profile. The method for performing density clustering on the functional linkage scatter plot to determine the core functional linkage region is as follows: Density peak clustering algorithm is used to scan all data points in the functional linkage scatter plot, calculate the local density and relative distance of each data point, select the points with local density higher than the density threshold and relative distance greater than the distance threshold as cluster centers, and assign the remaining data points to the nearest cluster centers to form multiple clusters. The functional linkage range covered by each cluster is defined as the core functional linkage region. Based on the aforementioned temporal dependency pattern, the transition probabilities between different core functional linkage areas are marked on the time event axis, specifically in the following manner: Extract the core functional linkage region to which adjacent behavioral segments belong in the continuous behavioral segment sequence, count the transfer frequency between regions, calculate the ratio of each transfer frequency to the total number of transfers, obtain the transfer probability matrix, map the transfer probability matrix onto the time event axis, and mark the time and probability value of each transfer.

6. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 1, characterized in that, The dynamic comparison and anomaly tracing module dynamically compares the real-time collected user signal stream with the user-specific functional baseline contour to detect abnormal linkage events that deviate from the functional baseline contour. Specifically, the real-time collected user signal stream is processed into a behavior segment to be compared in real time; the matching distance between the feature vector of the behavior segment to be compared and all core functional linkage areas in the functional baseline contour is calculated; if the minimum matching distance exceeds the adaptive distance threshold, a linkage deviation is determined to have occurred; when the number of consecutive linkage deviations reaches the event threshold, this consecutive linkage deviation is recorded as an abnormal linkage event.

7. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 6, characterized in that, The dynamic comparison and anomaly tracing module performs source analysis on the abnormal linkage event to locate the intersection of its corresponding cognitive function dimension and physical function dimension. Specifically, it decomposes all the behavioral segments to be compared contained in the abnormal linkage event, extracts the cognitive component feature decay rate and physical component feature disorder degree in each behavioral segment to be compared; projects the cognitive component feature decay rate onto the cognitive function dimension model to identify the damaged cognitive sub-dimension; projects the physical component feature disorder degree onto the physical function dimension model to identify the damaged physical sub-dimension; the intersection of the damaged cognitive sub-dimension and the damaged physical sub-dimension is the intersection of the cognitive function dimension and the physical function dimension. The method of projecting the decay rate of the cognitive component features onto the cognitive function dimension model to identify the damaged cognitive sub-dimensions is as follows: The cognitive function dimension model predefines multiple cognitive sub-dimensions and their normal feature ranges. The Euclidean distance between the feature decay rate of the cognitive component and the normal feature range of each cognitive sub-dimension is calculated. Cognitive sub-dimensions whose Euclidean distance exceeds the decay threshold are marked as damaged cognitive sub-dimensions. The method of projecting the disorder of the body component features onto the body function dimension model to identify the damaged body sub-dimension is as follows: The somatic function dimension model predefines multiple somatic sub-dimensions and their normal fluctuation ranges. It calculates the divergence value between the disorder degree of the somatic component features and the normal fluctuation range of each somatic sub-dimension, and marks the somatic sub-dimension with a divergence value exceeding the disorder threshold as the damaged somatic sub-dimension.

8. The system for the integrated assessment and rehabilitation training of cognitive and physical functions in the elderly according to claim 1, characterized in that, The rehabilitation path generation and execution module constructs a personalized multidimensional rehabilitation induction path based on the intersection point. Specifically, it queries a pre-set rehabilitation strategy knowledge base to obtain a set of basic rehabilitation action units associated with the intersection point of the cognitive function dimension and the physical function dimension; it assesses the user's current ability boundary conditions and adjusts the difficulty coefficient and load of the set of basic rehabilitation action units; based on the principle of neuroplasticity and the theory of motor learning stages, it arranges the adjusted rehabilitation action units into a sequence with a specific time rhythm and intensity gradient, and the sequence is the personalized multidimensional rehabilitation induction path.

9. The elderly cognitive function and physical function linkage assessment and rehabilitation training system according to claim 8, characterized in that, The rehabilitation path generation and execution module transforms the multidimensional rehabilitation induction path into a specific device control sequence. Specifically, it parses each rehabilitation action unit in the multidimensional rehabilitation induction path, decomposing it into parameter requirements for the cognitive stimulation device and movement instructions for the physical training device; it establishes a synchronous triggering timing relationship between the parameter requirements and the movement instructions; and it encodes the parameter requirements and movement instructions with the synchronous triggering timing relationship into a standardized control protocol data packet to form the device control sequence. The specific method for establishing the synchronous triggering timing relationship between the parameter requirements and the motion command is as follows: The starting time and parameter adjustment sequence of the cognitive stimulation device are obtained by parsing the parameter requirements. The action sequence and duration of the physical training device are obtained by parsing the movement instructions. A timing coordinator is constructed to align the starting time with the start time of the action sequence. An event trigger is set so that when the cognitive stimulation device reaches a specific parameter value, a signal is immediately sent to the physical training device to trigger the corresponding movement instruction. A time window constraint is defined to ensure that the parameter requirements and the movement instructions are executed synchronously within the specified time window.

10. The system for the integrated assessment and rehabilitation training of cognitive and physical functions in the elderly according to claim 1, characterized in that, The rehabilitation pathway generation and execution module executes the device control sequence through an environmental interaction terminal. Specifically, the environmental interaction terminal includes a visual projection device, an audio feedback device, and a mechanical guidance device. The visual projection device presents a dynamic visual task scene according to the device control sequence. The sound feedback device provides rhythmic guidance or semantic cues according to the device control sequence; the mechanical guidance device applies auxiliary or resistive mechanical stimulation according to the device control sequence; and monitors the user's reaction data in real time under the visual task scenario, rhythmic guidance or semantic cues, and mechanical stimulation, and fine-tunes the data packet content of the subsequent device control sequence accordingly.

Images