Stroke patient double lower limb movement evaluation method and system

Abstract

The application discloses a kind of cerebral apoplexy patient double lower limbs movement evaluation method and system, it is related to motor function evaluation technical field.A kind of cerebral apoplexy patient double lower limbs movement evaluation system, including have: task guide module, motion acquisition module, motion scoring module, interference evaluation module, trend analysis module, combination analysis module and result output module.The application is matched by constructing the scoring model based on population category characteristics, and motion ability score is carried out using machine learning algorithm or fixed weighting combination, and personalized evaluation is realized according to different patient population characteristics, to improve the generalization ability and prediction accuracy of scoring model;By introducing patient age, gender, movement grade and other characteristics for grouping modeling, and automatically switching scoring strategy according to sample quantity, the scoring mechanism also has reliability and clinical consistency in the limited data amount of home scene.

Description

Technical Field

[0001] This invention relates to the field of motor function assessment technology, and in particular to a method and system for assessing the motor function of the lower limbs in stroke patients. Background Technology

[0002] Stroke is one of the leading causes of neurological dysfunction. Patients often experience lower limb motor function weakness, gait abnormalities, and cognitive decline during rehabilitation. Accurate assessment of a patient's lower limb motor ability is crucial for guiding rehabilitation programs and preventing complications such as falls.

[0003] Traditional rehabilitation assessment methods mainly rely on scale scoring, manual observation, or clinical gait analysis. These methods have significant limitations: they are highly subjective, infrequent, and fail to reflect the patient's true functional status in daily life. With the development of wearable sensing technology, some studies have attempted to use sensors such as plantar pressure and inertial measurement units to quantify the patient's movement status, improving the objectivity and continuity of the assessment.

[0004] However, current technologies still lack a systematic assessment method that can integrate multi-source motor and cognitive data, dynamically identify trends in ability fluctuations, and effectively assess rehabilitation risk status in unstructured environments such as the home. Specifically, current solutions are still imperfect in terms of data quality control, cognitive interference modeling, time series trend analysis, and risk warning mechanisms, and lack personalized scoring models and risk evolution path identification strategies that can adapt to different population types.

[0005] Therefore, there is an urgent need for a method for assessing lower limb movement in stroke patients that is suitable for home environments, enabling high-quality data collection, multimodal feature fusion, dynamic trend modeling, and personalized risk assessment, in order to support more scientific and precise rehabilitation interventions and remote medical monitoring. Summary of the Invention

[0006] This invention aims to propose a method and system for assessing the movement of both lower limbs in stroke patients. It has a highly robust data processing capability that adapts to the home environment, integrates cognitive task interference and motor performance characteristics, and combines ability fluctuation trends and risk evolution path modeling to achieve individualized, dynamic and predictable assessment of the patient's functional status.

[0007] A method for assessing lower limb motor function in stroke patients includes:

[0008] In a home setting, guide patients to perform lower limb assessment tasks, which include single-task and dual-task tasks; single-task tasks include only motor tasks, while dual-task tasks include simultaneous motor and cognitive tasks.

[0009] During the patient's exercise task, raw data including lower limb joint angles, gait cycle, walking acceleration, electromyographic activity, and plantar pressure were acquired, and the raw data were quality controlled.

[0010] Feature extraction is performed on the raw data to obtain motor function features including gait symmetry, duration of the stance phase, ankle range of motion, stride length, and walking speed. These features are then matched with a scoring model that corresponds to the basic features of the population category to calculate the motor ability score.

[0011] When the assessment task is a dual task, the patient's response time and accuracy in the cognitive task are collected simultaneously, the impact of cognitive task performance on motor function parameters is analyzed, and a cognitive interference index is generated.

[0012] The time-stamped athletic ability scores within a continuous assessment period are compiled into time series data, and the ability volatility is calculated using a non-uniform sampling adaptive analysis method.

[0013] Perform combined analysis of cognitive interference index and ability volatility, and output anomaly alerts;

[0014] The assessment results are output as motor ability score, cognitive interference index, ability volatility, and abnormal prompts.

[0015] As a preferred embodiment of the present invention, the quality control of the raw data includes:

[0016] Dynamic filtering based on environmental sensing parameters, including light intensity and ground flatness; signal drift correction based on historical data, including constructing a reference baseline based on historical stable sampling segments and performing data recalibration; and consistency comparison based on multi-source cross-validation processing, including real-time error comparison between acquisition channels to identify and remove abnormal data points.

[0017] As a preferred embodiment of the present invention, the dynamic filtering and multi-source cross-validation include:

[0018] Dynamic filtering: When the light intensity change exceeds the set threshold or the ground flatness fluctuation exceeds the set range, a high-pass filter or a weighted moving average filter is activated to filter the raw data in order to suppress noise interference caused by environmental changes.

[0019] Multi-source cross-validation processing performs real-time matching and comparison between acceleration data acquired from both sides of the lower limbs, and combines the synchronous signals from the inertial measurement unit and electromyography sensor for complementary verification. When the numerical deviation of a certain channel exceeds the set error tolerance, it is automatically identified as abnormal data and is removed or corrected.

[0020] As a preferred embodiment of the present invention, the scoring model is constructed based on machine learning algorithms or by a fixed weighted combination.

[0021] When the number of samples corresponding to a population category does not meet the preset threshold, a scoring model is constructed based on a fixed weighted combination associated with the basic features of the population category. When the number of samples meets the threshold condition, a machine learning algorithm is used to construct the scoring model. The machine learning algorithm is selected according to the sample features of the population category, including random forest, support vector machine, or logistic regression, and is used to train the scoring model based on motor function features and output the corresponding motor ability score. The basic features include the patient's motor ability level, age, gender, and disease stage.

[0022] As a preferred embodiment of the present invention, the fixed weighted combination scoring model includes:

[0023] For each population category, a corresponding fixed weighting combination is preset to replace the machine learning model for motor ability scoring when the number of samples does not meet the preset threshold. When the basic characteristics of the target patient match a certain population category, the fixed weighting combination associated with that population category is selected to perform weighted calculation on the patient's motor function characteristic parameters.

[0024] As a preferred embodiment of the present invention, the cognitive interference index includes:

[0025] During the patient's dual-task process, the motor function characteristics of the motor task and the response time and accuracy of the cognitive task are collected simultaneously. The perturbation of motor performance caused by the cognitive task is quantified, and the cognitive interference index is obtained, which includes a weighted combination of cognitive response delay rate and motor parameter deviation rate and is corrected by accuracy. This index is used to characterize the degree of influence of the cognitive task on lower limb motor ability.

[0026] The response delay rate is the percentage change in the dual-task response time relative to the baseline task response time; the motion parameter offset rate is the overall rate of change of motion function characteristics under dual-task conditions compared to motion function characteristics under baseline task conditions.

[0027] As a preferred embodiment of the present invention, the capability volatility includes:

[0028] The time series data is modeled using non-uniform sampling based on the timestamp information of athletic ability scores. A continuous trend curve is constructed using cubic spline interpolation, and the rate of change of the first derivative of the trend curve and the magnitude of local deviation are used as quantitative indicators of ability volatility.

[0029] As a preferred embodiment of the present invention, the combined analysis includes:

[0030] A two-dimensional risk state vector based on the cognitive interference index and ability volatility is constructed to represent the patient's current position in the cognitive-motor risk space. The trajectory trend of this vector is analyzed in continuous assessment cycles, and the trajectory direction change rate and risk area invasion angle are calculated. The trend synergy index of the cognitive interference index and ability volatility in multiple continuous cycles is calculated to identify risk synergy fluctuation patterns. The risk synergy fluctuation patterns, trajectory direction change rate, and risk area invasion angle constitute evolutionary features and are input into the risk evolution model to determine whether the patient conforms to the risk evolution path and output abnormal prompts.

[0031] The trend synergy indicators include the dynamic time-warped similarity of the rates of change of the two, the sliding window correlation coefficient, or the trend direction consistency score.

[0032] As a preferred embodiment of the present invention, the risk evolution model includes:

[0033] Cluster analysis was performed on longitudinal assessment data of large-scale stroke rehabilitation patients to construct multiple typical risk evolution paths, and corresponding abnormal prompts were set for each path to form a risk evolution model. In the risk evolution model, the initial evolution path was obtained by initial matching of risk synergistic fluctuation patterns, and the final risk evolution path was obtained by similarity matching of the initial evolution path by trajectory direction change rate and risk area invasion angle.

[0034] A system for assessing the movement of both lower limbs in stroke patients includes:

[0035] Task guidance module: Used to guide patients to perform lower limb assessment tasks in a home environment;

[0036] Motion acquisition module: Used to acquire raw data during the patient's performance of a motion task and to perform quality control;

[0037] The exercise scoring module is used to extract features from the raw data, obtain motor function features, and calculate the exercise ability score through the scoring model.

[0038] Interference assessment module: used to analyze the impact of cognitive task performance on motor function parameters and generate a cognitive interference index;

[0039] Trend Analysis Module: Used to assemble time-series data of time-stamped athletic ability scores within a continuous assessment period, and to calculate ability volatility using a non-uniform sampling adaptive analysis method;

[0040] Combined Analysis Module: Used to perform combined analysis of cognitive interference index and ability volatility, and output anomaly alerts;

[0041] Results output module: Used to output motor ability score, cognitive interference index, ability volatility and abnormal prompts as evaluation results.

[0042] The present invention has the following advantages:

[0043] This invention guides patients to perform lower limb assessment tasks, including single-task and dual-task assessments, in a home environment. This allows for high-frequency, low-interference assessments in patients' daily lives, effectively improving the ecological effectiveness and daily adaptability of rehabilitation assessments. Compared to the periodic assessments in traditional clinical settings that rely on professional equipment and personnel, this invention enables patients to conduct self-monitoring of functions in a natural environment, which helps improve rehabilitation compliance and long-term follow-up continuity.

[0044] This invention collects multi-source raw data on lower limb joint angles, gait cycles, walking acceleration, electromyographic activity, and plantar pressure, and performs quality control to ensure the accuracy and reliability of the assessment data in an unstructured environment. It reduces the impact of environmental interference on the results. Through dynamic filtering, signal drift correction, and multi-source cross-validation mechanisms, it effectively identifies and eliminates abnormal data caused by changes in lighting, ground material, or wearing errors, significantly improving the data stability and assessment usability in home scenarios.

[0045] This invention constructs a scoring model based on population category feature matching, uses machine learning algorithms or fixed weighted combinations to score motor ability, achieves personalized assessment based on the characteristics of different patient groups, and improves the generalization ability and prediction accuracy of the scoring model; by introducing features such as patient age, gender, and motor level for group modeling, and automatically switching the scoring strategy according to the number of samples, the scoring mechanism has reliability and clinical consistency in home scenarios with limited data.

[0046] This invention generates a cognitive interference index by simultaneously collecting cognitive task response time and accuracy under dual-task conditions and combining it with motor function parameters. This index quantifies the interference effect of cognitive tasks on motor ability and enhances the sensitivity of assessment to complications such as fall risk. This index not only reflects cognitive-motor coordination load capacity, but also reveals the functional vulnerabilities of patients when performing complex tasks, providing data support for the early identification of high-risk fall states and cognitive decline trends.

[0047] This invention constructs a time series of motor ability scores within a continuous assessment period, and uses a non-uniform sampling adaptation analysis method to identify the trend of ability change and ability fluctuation rate, dynamically capturing the fluctuation state of the patient's ability during the rehabilitation process, and realizing early warning of degeneration.

[0048] This invention constructs a two-dimensional risk state vector based on the cognitive interference index and the ability volatility, and analyzes features such as trajectory trends and trend synergy to identify synergistic fluctuation patterns. It accurately models the interactive risk mechanism between cognition and movement, enhancing the sensitivity and specificity of risk identification. By inputting synergistic fluctuation patterns, trajectory direction change rate, and risk area invasion angle into the risk evolution model, it matches the risk evolution path and outputs abnormal prompt information, realizing the state classification of rehabilitation paths and the early identification of high-risk development trends, supporting remote rehabilitation monitoring and individualized intervention. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the structure of a stroke patient's lower limb movement assessment system used in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0052] Example 1: A method for assessing the movement of both lower limbs in stroke patients, comprising the following steps:

[0053] Step S1: Guide the patient to perform a lower limb assessment task in a home environment. The assessment task includes single-task and dual-task; the single-task only includes a motor task, and the dual-task includes a simultaneous motor task and a cognitive task.

[0054] In this embodiment, the home environment refers to the scenario in which the patient is assessed in their living quarters outside of a medical institution, including typical areas such as bedrooms, living rooms, and balconies. This environment usually has unstructured characteristics such as irregular spatial layout, frequent changes in lighting, and diverse background noise. Therefore, the assessment task needs to have good environmental adaptability and remote guidance capability.

[0055] The guidance methods include text prompts, voice broadcasts, or visual demonstrations (such as app animations or video instructions) to help patients complete the sequence of actions independently without the presence of a professional. The guidance method is dynamically selected based on the patient's historical performance.

[0056] For example, when it is detected that the patient's accuracy in the previous assessment task was insufficient, the guided mode with video demonstration is preferred.

[0057] The evaluation tasks are divided into two categories according to their complexity: single-task and dual-task.

[0058] A single task refers to a sequence of lower limb movements that only involves motor tasks, including: high knees in place; slow straight walking; standing and turning; and standardized rehabilitation movements such as alternating forward and backward steps. This task is mainly used to assess the patient's basic lower limb motor abilities without cognitive interference.

[0059] Dual-task assessment involves performing the aforementioned motor tasks while simultaneously adding a mild cognitive task to simulate multi-task situations in daily life, such as "walking + thinking" or "walking + listening to speech." Cognitive tasks include, but are not limited to: continuous backward digit counting (e.g., counting down from 100); instruction recognition and response tasks (e.g., quickly identifying and naming a color or object after hearing a speech); and simple word memorization tasks (e.g., remembering three words while walking). Dual-task assessment significantly improves sensitivity to cognitive-motor coupling abilities and is a crucial step in assessing potential fall risk.

[0060] To ensure standardized movements and consistent results, assessment tasks should be performed uniformly after calibration. For example, in the high knee raise task, the maximum knee elevation angle should be collected for each attempt, and a message should be displayed indicating "movement completed" or "retry." The start and end times of the cognitive task should be triggered synchronously with the motor task to ensure that real-time interference from cognitive load on motor behavior can be accurately captured.

[0061] Step S2: During the patient's exercise task, acquire raw data including lower limb joint angles, gait cycle, walking acceleration, electromyographic activity, and plantar pressure, and perform quality control on the raw data;

[0062] In this embodiment, the raw data is the lower limb motor physiological and dynamic data synchronously collected by the patient through a multimodal sensor during the completion of the exercise task. This type of data forms the basis for subsequent feature extraction, ability scoring and risk modeling.

[0063] The methods and sources for obtaining raw data include:

[0064] Lower limb joint angles: acquired by inertial measurement units (IMUs) deployed at locations such as the knee and ankle joints, including dual-source fusion of gyroscopes and accelerometers. The output is the angle variation value in the sagittal and coronal planes, in degrees (°), used to analyze flexion-extension amplitude and gait periodicity. For example, the maximum right knee flexion angle is approximately 65°, which is the threshold for high knee raises.

[0065] Gait cycle: The start and end time of a complete gait, calculated by an array of plantar pressure sensors, including the duration of the stance and swing phases, measured in seconds or milliseconds. For example, the complete gait cycle of the left foot is 1.1 seconds, with the stance phase accounting for 64%.

[0066] Walking acceleration: Triaxial acceleration signals are acquired by a wearable IMU, covering sensor points on the legs, hips, and feet, with a data frequency of 50~200Hz and units of m / s². The trends in these signals are used to calculate cadence, symmetry, and identify abnormal gait rhythms. For example, a deviation of >20% in left and right x-axis acceleration indicates gait asymmetry.

[0067] Electromyography (EMG): Electromyographic signals of major lower limb muscle groups are acquired using surface electromyography electrodes. Commonly used muscles include the quadriceps femoris, hamstrings, and tibialis anterior. The sampling frequency is 500-1000 Hz, and the unit is mV. It reflects the intensity and coordination of muscle contraction. For example, a low EMG RMS value in the right quadriceps femoris indicates weakened knee extension strength.

[0068] Plantar pressure: The pressure values ​​per unit area of ​​the forefoot, heel, and midfoot are acquired by flexible pressure sensing pads (distributed array), measured in kPa, and used to detect support load transfer patterns and gait stability. For example, a difference of more than 20 kPa between the two feet indicates a risk of center of gravity shift.

[0069] To ensure the accuracy and stability of the collected data in a home environment, three quality control mechanisms are implemented:

[0070] Dynamic filtering (with environmental perception parameters involved): Configure an environmental perception unit to collect current light intensity (such as Lux value) and ground flatness (estimated by millimeter-wave sensor or visual depth camera); when the detected change in ambient light exceeds a set threshold (such as ΔLux>200), or the fluctuation of ground flatness exceeds a threshold (such as tilt angle>5°), a high-pass filter (to remove low-frequency drift) or a weighted moving average filter (to suppress high-frequency noise) is automatically activated to ensure data stability; for example, acceleration signal fluctuations are amplified by environmental changes, and the normal waveform is restored after filtering.

[0071] Signal drift correction (historical comparison modeling): Record stable movement segments of the user in the historical normal state and build a multi-dimensional reference baseline; if the current signal shows a continuous deviation trend (such as angle baseline drift >10°), a dynamic recalibration operation will be performed; for example, the ankle joint angle gradually shifts when walking on flat ground, and is reverted to the stable baseline after being detected.

[0072] Multi-source cross-validation (real-time consistency comparison): Real-time comparison of output data from multiple sensor channels (such as dual-leg IMU, EMG, pressure pad); if abnormal differences are detected between the data from one side channel and the opposite side and auxiliary channels (such as EMG) (such as deviation rate >25%), abnormal rejection or signal correction is performed; for example, the distortion of the right leg acceleration signal is identified by the electromyography-left leg acceleration cross-validation and automatically rejected.

[0073] Step S3: Extract features from the raw data to obtain motor function features including gait symmetry, duration of the stance phase, ankle range of motion, stride length, and walking speed. Match the scoring model with the basic features corresponding to the population category to calculate the motor ability score.

[0074] The scoring model is constructed based on machine learning algorithms or by a fixed weighted combination.

[0075] When the number of samples corresponding to a population category does not meet the preset threshold, a scoring model is constructed based on a fixed weighted combination associated with the basic features of the population category. When the number of samples meets the threshold condition, a machine learning algorithm is used to construct the scoring model. The machine learning algorithm is selected according to the sample features of the population category, including random forest, support vector machine, or logistic regression, and is used to train the scoring model based on motor function features and output the corresponding motor ability score. The basic features include the patient's motor ability level, age, gender, and disease stage.

[0076] The fixed-weighted combination scoring model includes:

[0077] For each population category, a corresponding fixed weighting combination is preset to replace the machine learning model for motor ability scoring when the number of samples does not meet the preset threshold. When the basic characteristics of the target patient match a certain population category, the fixed weighting combination associated with that population category is selected to perform weighted calculation on the patient's motor function characteristic parameters.

[0078] In this embodiment, motor function features are extracted from multimodal raw data that has undergone data quality control, aiming to quantify the core lower limb motor performance of stroke patients, and to map these performances into structured motor ability scores through a personalized scoring model.

[0079] The description and extraction methods of motor function characteristics include:

[0080] Gait symmetry: This indicates the consistency of the left and right lower limbs in terms of movement rhythm, amplitude, and frequency. It is calculated using the symmetry index (SI), with the formula: SI = |left step length − right step length| / (0.5 × (left step length + right step length)) × 100%, in percentages. High symmetry indicates good neural control, while low symmetry may suggest hemiplegia or motor incoordination.

[0081] Stance phase duration: This represents the proportion of the entire gait cycle that the foot bears weight on the ground, expressed as a percentage. It is determined by plantar pressure sensors detecting the start and end times of foot contact. For example, the left foot's stance phase duration is 0.72 seconds, accounting for 66% of the gait cycle.

[0082] Ankle range of motion: This represents the maximum difference between the flexion and extension angles of the ankle joint during gait, measured in degrees (°), and obtained by integrating the IMU angular velocity signal. The normal range is approximately 20–30°; limited range of motion indicates stiffness or movement disorders.

[0083] Stride length: The horizontal distance between two consecutive landings of the same foot, measured in cm, is estimated by IMU trajectory or calculated by plantar pressure space. Shortened stride length is associated with cognitive decline or fatigue.

[0084] Walking speed: Measured by dividing the total distance by the total time, in m / s, it is the most representative single indicator in rehabilitation assessment.

[0085] To improve scoring accuracy and individual fit, a dynamic matching mechanism for the scoring model is designed, which selects a scoring strategy based on the "patient's population category," including:

[0086] The basic characteristics of the population categories were defined by clinical experts and included: exercise capacity level (based on FMA scores, etc.); age range (e.g., young adults <45, middle-aged adults 45-65, elderly >65); gender (considering differences in muscle strength); and disease stage (acute phase, subacute phase, chronic phase, etc.). These basic characteristics were used to construct the matching logic between patients and historical samples.

[0087] If the number of historical training samples of the target patient’s population category is greater than or equal to a preset threshold (e.g., 50), a machine learning scoring model is used; if it is less than the threshold, a fixed weighted combination scoring model that matches the population category is selected.

[0088] The scoring model built by the machine learning algorithm has the following inputs: five categories of extracted motor function features; model type, which is automatically selected based on the features of the training samples; and output: a motor ability score (e.g., 0-100 points), used to track recovery trends or evaluate intervention effects.

[0089] Fixed weighted combination model: Each population group has a preset weight combination; the weights are derived from the historical characteristics sensitivity analysis of the corresponding population; input patient characteristics, perform weighted summation to output scores; automatically select the weight template with the highest matching degree with the basic characteristics of the target patient to perform scoring.

[0090] This dynamic switching mechanism ensures that scoring strategies are available in the early stages of sample accumulation, and that learning capabilities are available after sufficient samples have been collected, while also guaranteeing the interpretability of cross-population comparisons.

[0091] Step S4: When the assessment task is a dual task, the patient's response time and accuracy in the cognitive task are collected simultaneously, the impact of cognitive task performance on motor function parameters is analyzed, and a cognitive interference index is generated.

[0092] The cognitive interference index includes:

[0093] During the patient's dual-task process, the motor function characteristics of the motor task and the response time and accuracy of the cognitive task are collected simultaneously. The perturbation of motor performance caused by the cognitive task is quantified, and the cognitive interference index is obtained, which includes a weighted combination of cognitive response delay rate and motor parameter deviation rate and is corrected by accuracy. This index is used to characterize the degree of influence of the cognitive task on lower limb motor ability.

[0094] The response delay rate is the percentage change in the dual-task response time relative to the baseline task response time; the motion parameter offset rate is the overall rate of change of motion function characteristics under dual-task conditions compared to motion function characteristics under baseline task conditions.

[0095] To quantify the actual degree of interference of cognitive tasks on the lower limb motor ability of stroke patients when performing dual-task motor activities, this embodiment constructs a cognitive interference index, which reflects the motor-cognitive resource allocation ability and the stability of neural control through multimodal signal co-analysis.

[0096] Dual-task definition: In this embodiment, dual-task assessment requires patients to complete a cognitive task while performing standard lower limb motor tasks (such as walking, leg lifting, and turning).

[0097] During each task, the following data are collected synchronously: cognitive task data (response time, in milliseconds; accuracy rate, in %); motor function data (the same five types of parameters as in step S3); timestamps, execution stage markers (start / end / answer point), and other control data.

[0098] Baseline task definition: Patients are required to complete a purely motor task without cognitive tasks, thereby constructing an individualized baseline of motor and cognitive responses for comparative analysis.

[0099] The cognitive interference index is used to assess the combined effects of cognitive task loading on motor ability, and specifically includes the following components:

[0100] Cognitive response delay rate: ,in For the response time of dual tasks, The response time of the baseline task reflects the processing lag caused by increased cognitive load;

[0101] The motion parameter offset rate is calculated by comparing the rate of change of motion function characteristics under dual-task and single-task conditions: In this embodiment This represents five motor function characteristics. Subscripts representing motor function characteristics. and The first two tasks are respectively the dual-task and baseline task. Each motor function characteristic parameter.

[0102] Accuracy Correction Factor: If the accuracy of the cognitive task decreases significantly, it indicates that the task has not been performed effectively, and the interference index should be increased; otherwise, the interference evaluation should be weakened proportionally.

[0103] The final cognitive interference index is calculated as follows: CII = α × delay rate + β × offset rate (accuracy rate), where α and β are preset weighting coefficients (such as 0.5 / 0.5 or adjusted according to the characteristics of the population).

[0104] Based on historical data and expert settings, a personalized interference threshold is set for CII (e.g., >25% is considered significant interference).

[0105] The interference index is used to construct the subsequent "risk state vector"; to identify the "cognitive regression" rehabilitation pattern; to assist in assessing the patient's cognitive load capacity; and to guide the training rhythm and task arrangement.

[0106] Cognitive Interference Index (CII): refers to the average perturbation intensity of cognitive task loading on motor performance under dual-task conditions. It reflects the neural integration efficiency of stroke patients in resource conflict situations. CII is one of the important innovative indicators proposed in this invention, bridging the gap between traditional motor assessment and multi-task challenges in real life.

[0107] Step S5: Combine the time-stamped athletic ability scores within the continuous assessment period into time series data, and calculate the ability volatility using a non-uniform sampling adaptive analysis method;

[0108] The capacity volatility includes:

[0109] The time series data is modeled using non-uniform sampling based on the timestamp information of athletic ability scores. A continuous trend curve is constructed using cubic spline interpolation, and the rate of change of the first derivative of the trend curve and the magnitude of local deviation are used as quantitative indicators of ability volatility.

[0110] This step involves trend modeling and dynamic analysis of the patient's continuous motor ability scores over a period of time to identify the stability and magnitude of changes in their motor ability, thereby providing fluctuation characteristic input for subsequent risk assessment.

[0111] After each scoring task is executed in step S3, a sports ability score result with a timestamp will be output, and the record format is as follows: [score value, timestamp, task type (single task / double task), completion mark];

[0112] In home assessment scenarios, patients perform tasks at irregular frequencies (e.g., once a day or once every three days), and the timestamp intervals are irregular and sparse, causing traditional equal-interval time series analysis algorithms to fail.

[0113] To address the issue of varying time intervals, this invention employs the following steps: All scoring results are sorted according to timestamps to establish a non-equidistant time series. A cubic spline interpolation method is introduced to perform continuous curve fitting on the above series within a set interval (e.g., 7 days, 14 days). This method ensures the fitted curve is continuous at each sampling point and has continuous first and second derivatives at nodes, avoiding oscillation artifacts. A continuous function curve f(t) is output, where t is the time axis and the function value is the scoring result.

[0114] Based on the aforementioned trend curve, the following indicators are calculated to construct the capability volatility (FVR): This represents the severity of trend changes, derived by calculating the variance of the derivative value (i.e., slope) change per unit time, measured in "scores / day²", used to quantify the rate of fluctuation. Local bias magnitude: By setting a sliding window (e.g., 3 days), the residuals of local scores are compared with the fitted values, reflecting the patient's fluctuations within a short period, used to identify early signs of instability. Maximum rate of decline and duration (optional indicator): Records the most significant slope of score decline over a period and the number of days this trend lasts; this information is used for inference modeling in conjunction with subsequent risk evolution paths.

[0115] The capability volatility is ultimately expressed as: FVR = w1 slope variance + w2 local bias + w3 maximum downside rate (optional), where w1, w2, and w3 are weight adjustment parameters;

[0116] As a dynamic indicator, ability volatility reflects the fluctuations in the patient's recovery process and is a core parameter for determining whether there is a "non-linear degradation" trend. This method is particularly well-suited to the data characteristics in a home environment and has a high sensitivity to identify mild fluctuation trends over long periods outside the clinical setting.

[0117] Step S6: Perform a combined analysis of the cognitive interference index and ability volatility, and output anomaly alerts;

[0118] The combined analysis includes:

[0119] A two-dimensional risk state vector based on the cognitive interference index and ability volatility is constructed to represent the patient's current position in the cognitive-motor risk space. The trajectory trend of this vector is analyzed in continuous assessment cycles, and the trajectory direction change rate and risk area invasion angle are calculated. The trend synergy index of the cognitive interference index and ability volatility in multiple continuous cycles is calculated to identify risk synergy fluctuation patterns. The risk synergy fluctuation patterns, trajectory direction change rate, and risk area invasion angle constitute evolutionary features and are input into the risk evolution model to determine whether the patient conforms to the risk evolution path and output abnormal prompts.

[0120] The trend synergy indicators include the dynamic time-warped similarity of the rates of change of the two, the sliding window correlation coefficient, or the trend direction consistency score.

[0121] The risk evolution model includes:

[0122] Cluster analysis was performed on longitudinal assessment data of large-scale stroke rehabilitation patients to construct multiple typical risk evolution paths, and corresponding abnormal prompts were set for each path to form a risk evolution model. In the risk evolution model, the initial evolution path was obtained by initial matching of risk synergistic fluctuation patterns, and the final risk evolution path was obtained by similarity matching of the initial evolution path by trajectory direction change rate and risk area invasion angle.

[0123] In this application, the goal of the current steps is to jointly model the cognitive interference index (CII) and ability volatility (FVR) to form a dynamic process of patient risk state evolution, and to establish a risk evolution model based on real rehabilitation big data to identify the trend of patients evolving towards a high-risk state in advance.

[0124] Construct a two-dimensional vector space using the patient's CII and FVR generated in each assessment cycle as the x and y axes: RiskVecotr t =[CII t FVR t ];

[0125] The coordinate position of the vector represents the patient's current state in the two dimensions of "cognitive impact" and "motor stability". Combined with the set high-risk area boundary (such as the upper right quadrant corresponding to high interference + high volatility), the risk level of a single state point can be roughly determined. The vector sequence under multiple assessment cycles constitutes a "trajectory curve" for further trend judgment.

[0126] Trend analysis was performed on the RiskVector sequences to extract the following three types of evolutionary features:

[0127] Trajectory Direction Change Rate (DCR): Represents the continuity of trajectory direction change across different assessment periods; frequent directional fluctuations usually reflect unstable patient recovery or interference from multiple factors; calculated using the root mean square value of vector angle change.

[0128] Risk Area Intrusion Angle (REA): Measures the angle between the trajectory and the boundary vertical line when the trajectory first enters the high-risk quadrant; the closer to vertical entry, the faster and more sudden the risk accumulation; used to distinguish between the two types of evolutionary trends of "slow landslide" and "sudden drop".

[0129] Trend Coherence Index (TCL): This indicates whether the cognitive interference index and ability volatility show a consistent trend. The higher the trend coherence, the more clinically alarming the risk of synchronous deterioration of cognitive burden and motor instability. TCL includes the following three calculation methods: Dynamic Time Warping Similarity (DTW Similarity), used to capture non-linear similarity trends; Sliding Window Correlation Coefficient, to identify the degree of synchronous change within a short period; Directional Consistency Score, which counts the proportion of synchronous increase or decrease of the two.

[0130] The DCR, REA, and TCI are integrated into a three-dimensional evolutionary feature vector, which is then fed into the risk evolution model as input.

[0131] The risk evolution model is the core intelligent analysis component of this invention. It is built based on large-scale rehabilitation patient data and includes: extracting trajectory sequences with significant statistical characteristics from historical data, constructing typical evolution paths using time-series clustering algorithms (such as K-SpectralClustering); and binding specific abnormal warning labels to each path, such as: high risk of falls; dominant cognitive decline; volatility imbalance; and delayed rehabilitation.

[0132] Two-stage path matching: Model matching is triggered based on the presence of a risk-coordinated volatility pattern (TCI exceeding a threshold); paths are ranked for similarity using the trajectory direction change rate (DCR) and risk area intrusion angle (REA), selecting the most likely evolutionary direction. If the matched path is of a high-risk type and its similarity exceeds a set threshold, an anomaly alert is output.

[0133] Risk-coordinated volatility pattern: This refers to the simultaneous synchronous changes (e.g., simultaneous increases) of the cognitive interference index and ability volatility across multiple consecutive assessment periods, indicating a disorder in the cognitive-motor regulation mechanism and having a higher clinical risk orientation.

[0134] Evolutionary Feature Vector: Includes three-dimensional indicators [TCI, DCR, REA], used to quantify the risk status trend in the current assessment stage, and is the core input for path identification.

[0135] Risk evolution path: A historical experience model reflecting how risk indicators evolve in a typical recovery population at a specific stage, providing a basis for systematically predicting the future risk trend of patients.

[0136] Step S7: Output the motor ability score, cognitive interference index, ability volatility, and abnormal prompts as the evaluation results.

[0137] In this application, the current step aims to uniformly summarize and structure the results of the preceding analysis, and output them in a standardized format for feedback to user terminals, healthcare worker platforms, or clinical assessment systems, in order to support personalized rehabilitation management.

[0138] The evaluation results are output in structured JSON or HL7 data format, which facilitates integration with third-party medical information systems (such as EMR).

[0139] Example 2, a lower limb motor assessment system for stroke patients, see [link to example]. Figure 1 As shown, it includes the following modules:

[0140] Task guidance module: Used to guide patients to perform lower limb assessment tasks in a home environment;

[0141] Motion acquisition module: Used to acquire raw data during the patient's performance of a motion task and to perform quality control;

[0142] The exercise scoring module is used to extract features from the raw data, obtain motor function features, and calculate the exercise ability score through the scoring model.

[0143] Interference assessment module: used to analyze the impact of cognitive task performance on motor function parameters and generate a cognitive interference index;

[0144] Trend Analysis Module: Used to assemble time-series data of time-stamped athletic ability scores within a continuous assessment period, and to calculate ability volatility using a non-uniform sampling adaptive analysis method;

[0145] Combined Analysis Module: Used to perform combined analysis of cognitive interference index and ability volatility, and output anomaly alerts;

[0146] Results output module: Used to output motor ability score, cognitive interference index, ability volatility and abnormal prompts as evaluation results.

[0147] Example 3: This example provides a scheme for assessing the movement of the lower limbs of stroke patients by combining a foot rehabilitation device. Based on the assessment method proposed in this invention, this scheme expands upon the existing "simple, dynamic, controllable, multi-area independent weight-bearing test standing posture training rehabilitation device" through hardware and software synergy, and realizes a comprehensive assessment of the lower limb functional status of stroke patients in a home or unstructured environment.

[0148] In actual deployment, the patient stands on the foot pedals set by the device, and the system guides them to complete a sequence of single tasks (such as stepping in place and static weight transfer) and dual tasks (such as stepping + reverse counting). Gait cycle and foot force distribution data are acquired through the device's built-in multi-region plantar pressure sensor, and combined with external inertial measurement unit (IMU) and surface electromyography (sEMG) sensor, the patient's lower limb joint angles, accelerations, and electromyographic activities are collected.

[0149] The collected raw data, after undergoing dynamic filtering and multi-channel cross-validation by this system, enters the feature extraction module to obtain motor function features such as gait symmetry and duration of the stance phase. The system automatically matches a scoring model based on the patient's basic characteristics (such as age and motor ability level) and outputs a motor ability score. Simultaneously, the system processes the cognitive response time and accuracy during the dual-task process to generate a cognitive interference index.

[0150] Ultimately, continuous cycle motor ability scores and cognitive interference indices were used for risk combination analysis to construct a two-dimensional risk state vector, extract evolutionary features such as trajectory trends and trend synergy, and input them into a preset risk evolution model to determine whether the patient is in a high-risk evolutionary path, and output fall warnings or rehabilitation strategy adjustment suggestions when necessary.

[0151] This embodiment demonstrates the feasibility and scalability of the present invention based on existing rehabilitation devices, proving its practical application feasibility and accuracy assurance capabilities in home settings.

[0152] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the movement of both lower limbs of a stroke patient, characterized by, Comprise: Guiding the patient to perform an assessment task of lower extremity in a home environment, the assessment task being a dual task containing a simultaneous motor task and a cognitive task; Acquiring raw data including joint angles of lower extremity, gait cycle, walking acceleration, electromyographic activity and plantar pressure during the patient performing the motor task, and performing quality control on the raw data; Extracting features from the raw data to obtain motor function features including gait symmetry, support phase duration, ankle range of motion, stride length and walking speed, and matching a scoring model corresponding to the basic features of the population category to calculate the motor ability score; Simultaneously collecting the response time and accuracy of the patient in the cognitive task, analyzing the influence of cognitive task execution on motor function parameters, and generating a cognitive interference index; The cognitive interference index comprises: During the patient performing the dual task, synchronously collecting the motor function features of the motor task and the response time and accuracy of the cognitive task, quantifying the motor performance disturbance caused by the cognitive task, obtaining the cognitive interference index, including the weighted combination of cognitive response delay rate and motor parameter deviation rate and being corrected by accuracy, for representing the influence degree of the cognitive task on the lower extremity motor ability; The response delay rate is the percentage change of the dual task response time relative to the baseline task response time; the motor parameter deviation rate is the comprehensive change rate of the motor function features under the dual task condition compared with the motor function features under the baseline task condition; Grouping the motor ability scores with time stamps in the continuous assessment period into time series data, and calculating the ability fluctuation rate using a analysis method adapted to non-uniform sampling; The ability fluctuation rate comprises: Modeling the time series data using non-uniform sampling based on the time stamp information of the motor ability score, constructing a continuous trend curve using cubic spline interpolation, and taking the first derivative change rate and local deviation amplitude of the trend curve as the quantitative indicators of the ability fluctuation rate; Combination analysis of the cognitive interference index and the ability fluctuation rate, and output of an abnormal prompt; The combination analysis comprises: Constructing a two-dimensional risk state vector based on the cognitive interference index and the ability fluctuation rate, for representing the patient's current position in the cognitive motor risk space; analyzing the trajectory trend of the vector in the continuous assessment period, calculating the trajectory direction change rate and the risk region invasion angle; calculating the trend consistency index of the cognitive interference index and the ability fluctuation rate in multiple continuous periods, for identifying the risk collaborative fluctuation mode; constructing the evolution features from the risk collaborative fluctuation mode, the trajectory direction change rate and the risk region invasion angle, and inputting them into a risk evolution model to determine whether the patient meets the risk evolution path, and outputting an abnormal prompt; The trend consistency index includes the dynamic time warping similarity of the change rates, the sliding window correlation coefficient or the trend direction consistency score; Outputting the motor ability score, the cognitive interference index, the ability fluctuation rate and the abnormal prompt as the evaluation results. 2.The stroke patient lower limb motor function assessment method according to claim 1, characterized in that, The quality control on the raw data comprises: Dynamic filtering based on environmental perception parameters, including light intensity and ground flatness; signal drift correction based on historical data, including constructing a reference baseline based on stable sampling segments and performing data recalibration operations; consistency comparison based on multi-source cross-validation processing, including real-time error comparison between acquisition channels to identify and eliminate abnormal data points. 3.The stroke patient lower limb motor function assessment method of claim 2, wherein, The dynamic filtering and multi-source cross-validation include: Dynamic filtering, when the light intensity changes exceed a set threshold or the ground flatness fluctuates beyond a set range, a high-pass filter or a weighted moving average filter is enabled to filter the original data to suppress noise interference caused by environmental changes; Multi-source cross-validation processing, real-time matching comparison between acceleration data acquired from both lower limbs, and complementary verification combined with the synchronization signals of the inertial measurement unit and the electromyographic sensor, when the value deviation of a channel exceeds the set error tolerance, it is automatically determined as abnormal data and is eliminated or corrected. 4.The stroke patient lower limb motor function assessment method of claim 1, wherein, The scoring model is constructed based on a machine learning algorithm or a fixed weighted combination; When the number of samples corresponding to the population category does not meet the preset threshold, a scoring model is constructed based on the fixed weighted combination associated with the basic features of the population category; when the sample quantity meets the threshold condition, a machine learning algorithm is used to construct the scoring model, the machine learning algorithm is selected according to the sample features of the population category, including random forest, support vector machine or logistic regression, which is used to train the scoring model based on the motor function features and output the corresponding motor ability score; the basic features include the motor ability level, age, gender and disease stage of the patient.

5. The stroke patient lower extremity motor assessment method of claim 4, wherein, The fixed weighted combination for constructing the scoring model includes: A corresponding fixed weighted combination is preset for each population category, which is used to replace the machine learning model to score the motor ability when the number of samples does not meet the preset threshold; when the basic features of the target patient match a certain population category, the fixed weighted combination associated with the population category is selected to weight the motor function feature parameters of the patient. 6.The stroke patient lower limb motor function assessment method of claim 1, wherein, The risk evolution model includes: Cluster analysis based on longitudinal evaluation data of a large number of stroke rehabilitation patients to construct multiple typical risk evolution paths, and set corresponding abnormal prompts for each path to form a risk evolution model; in the risk evolution model, the initial evolution path is obtained by initial matching of the risk collaborative fluctuation mode, and the final risk evolution path is obtained by similarity matching of the initial evolution path through the trajectory direction change rate and the risk region invasion angle.

7. A stroke patient lower extremity motor assessment system, characterized in that, The system applies any one of the stroke patient lower limb motor evaluation methods of claims 1-6, including: Task guidance module: used to guide the patient to perform lower limb evaluation tasks in a home environment; Motion acquisition module: used to acquire raw data during the patient's performance of the motion task and perform quality control; Motion scoring module: used to extract features from the raw data, obtain motor function features, and calculate motor ability scores through the scoring model; Interference evaluation module: for analyzing the impact of cognitive task execution on motor function parameters, generating a cognitive interference index; Trend analysis module: for grouping time-stamped motor ability scores in consecutive evaluation cycles into time series data, and calculating the ability volatility rate using a non-uniform sampling adaptive analysis method; Combined analysis module: for combined analysis of the cognitive interference index and the ability volatility rate, and outputting an abnormality prompt; Result output module: for outputting the motor ability score, the cognitive interference index, the ability volatility rate, and the abnormality prompt as evaluation results.

Images