An artificial intelligence-based community elderly frailty screening and intervention training method

Abstract

The application discloses a community old person frailty screening and intervention training method based on artificial intelligence, and belongs to the technical field of artificial intelligence, which comprises the following steps: acquiring dynamic physiological parameters and medical treatment data, generating a frailty evaluation parameter set; determining a target frailty level and combining a user tolerance parameter set to generate a personalized training scheme; generating a user action quality evaluation result based on real-time action monitoring, adjusting a training intensity parameter according to a baseline error threshold; returning the evaluation result, updating the frailty evaluation parameter set, and generating a community-level frailty early warning signal according to a group risk threshold. The application adopts a time sequence physiological data fusion of a wearable device and a decision tree-migration learning hybrid model, can realize real-time optimization of a personalized training scheme and dynamic early warning of group frailty risk, and significantly improves screening accuracy and intervention effectiveness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a community elderly frailty screening and intervention training method based on artificial intelligence. BACKGROUND

[0002] With the aging of society, the number of patients in the intermediate stage between the healthy state and the nursing state is increasing, and the above-mentioned patients are difficult to find their own frailty, and often drag to the state of needing care.

[0003] The existing frailty screening and intervention of the elderly mainly relies on static assessment tools, such as FRAIL scale, Fried frailty phenotype; single physical test, such as grip strength, step speed; and questionnaire survey to determine the frailty level, and adopts a fixed period of universal training scheme, such as balance pad exercise, resistance band training.

[0004] This frailty screening and intervention training method relies on artificial periodic retesting, resulting in a lag in frailty assessment. The training scheme also lacks a personalized adjustment mechanism and cannot customize the corresponding training scheme according to different frailty degrees, which weakens the training effect. At the same time, it also lacks the assessment of the risk of group frailty, and cannot make early warning for community-level group frailty. SUMMARY

[0005] To solve the above problems, the present application provides a community elderly frailty screening and intervention training method based on artificial intelligence, which adopts the fusion of wearable device dynamic physiological parameters and historical medical data, real-time action monitoring triggered by personalized training scheme, can realize dynamic grading evaluation of the frailty state of the elderly, generation of personalized training scheme and real-time optimization of intensity, and synchronously triggers community-level risk early warning signal, significantly improves the community elderly frailty screening ability and the effect of intervention training.

[0006] The above-mentioned target can be realized by the following scheme:

[0007] The application discloses a community old-age frailty screening and intervention training method based on artificial intelligence, and relates to the technical field of health care.

[0008] Optionally, the generating the frailty evaluation parameter set comprises: performing first feature screening on the dynamic data set to obtain an initial feature subset; extracting a key frailty index set from the initial feature subset through a preset decision tree feature selection model; and calculating a frailty score value in the frailty evaluation parameter set according to the key frailty index set and a preset weight distribution rule.

[0009] Optionally, the generating the personalized training scheme comprises: determining a candidate intervention mode set based on the target frailty level matching a preset intervention mode library; screening a target intervention mode from the candidate intervention mode set through a mixed integer programming algorithm according to a cardiopulmonary function index in the user's preset tolerance parameter set; and combining the target intervention mode and a gait stability index in the frailty evaluation parameter set to generate a training cycle parameter and an action type sequence of the personalized training scheme.

[0010] Optionally, the generating the training cycle parameter and the action type sequence of the personalized training scheme comprises: if the gait stability index is lower than a preset balance imbalance threshold, inserting a balance compensation training unit in the action type sequence; and when the execution frequency of the balance compensation training unit reaches a preset correction frequency, updating the weight value of the gait stability index and regenerating the action type sequence.

[0011] Optionally, the generating the user action quality evaluation result comprises: collecting a joint angle parameter set of the user performing a training action in real time through a preset fuzzy logic controller, to generate an action deviation coefficient; comparing the action deviation coefficient with a preset standard action template in the personalized training scheme, to generate a completion score value in the user action quality evaluation result; and if the completion score value is lower than a preset action qualified line, triggering an intensity adjustment instruction of the personalized training scheme.

[0012] Optionally, the dynamically adjusting the training intensity parameter of the personalized training scheme comprises: calculating a real-time metabolic load value of the user according to a heart rate recovery rate parameter in the user action quality evaluation result; if the real-time metabolic load value exceeds a preset metabolic safety threshold in the personalized training scheme, reducing a resistance coefficient in the training intensity parameter; and after adjusting the resistance coefficient, synchronously updating an upper limit value of cardiopulmonary function in the tolerance parameter set.

[0013] Optionally, the generating the updated frailty evaluation parameter set comprises: extracting an action completion time parameter and a joint pressure distribution parameter in the user action quality evaluation result, to generate a dynamic feedback data set; performing correlation analysis on the dynamic feedback data set and the frailty evaluation parameter set through a preset transfer learning model, to update a weight distribution proportion of the frailty score value; and according to the updated frailty score value, re-dividing a classification boundary of the target frailty grade.

[0014] Optionally, the updating the weight distribution proportion of the frailty score value comprises: when it is detected that the joint pressure distribution parameter continuously deviates from a preset baseline distribution range, generating a muscle compensation risk signal; based on the muscle compensation risk signal, triggering a model calibration process of the intervention strategy generation model; and based on the joint pressure distribution parameter and the metabolic load value, constraining the calibrated personalized training scheme.

[0015] Optionally, the generating the community-level frailty early warning signal comprises: obtaining an updated frailty evaluation parameter set of all users in a community, to generate a group frailty index distribution graph; performing spatio-temporal clustering analysis on the group frailty index distribution graph, to identify a frailty index mutation area; and if a user proportion of the frailty index mutation area exceeds a group risk threshold, triggering a directional health education instruction and adjusting a warning level of the community-level frailty early warning signal.

[0016] Optionally, the triggering of the health education instruction and the adjustment of the warning level of the community-level frailty warning signal comprise: generating a set of nutritional intervention suggestion parameters according to the user's nutritional intake data in the mutation area of the frailty index; synchronously pushing the set of nutritional intervention suggestion parameters and the updated personalized training scheme to generate a joint intervention instruction; and when the execution cycle of the joint intervention instruction is detected to be completed, re-collecting the set of dynamic physiological parameters of the user and updating the group frailty index distribution map.

[0017] Compared with the prior art, the present application has the following advantages:

[0018] 1. The present application realizes dynamic closed-loop evaluation and optimization; dynamic physiological parameters are collected in real time by a wearable device to dynamically extract time sequence features, and the frailty score weight is updated in real time based on a preset historical medical data set, solving the problems of high static evaluation missed detection rate and insufficient dynamic physiological feature capture, and breaking through the limitations of static screening;

[0019] 2. Personalized intervention and safety control are adopted; based on the target frailty level and the user's preset tolerance parameter set, a personalized training scheme is generated by optimizing the balance of the training target through a preset intervention strategy generation model; real-time action monitoring is triggered according to the scheme to generate a user action quality evaluation result; and the strength parameters of the training scheme are dynamically adjusted in real time in combination with the evaluation result and a preset baseline error threshold, to realize the cooperative optimization of safety and effectiveness, solving the problems of homogenization of traditional intervention schemes and lag of safety control, and reducing the risk of sports injury;

[0020] 3. Group risk warning is realized; the traditional method lacks community-level risk identification capability, the present application generates an updated frailty evaluation parameter set by feeding back the user action quality evaluation result to the update module of the frailty evaluation parameter set, and generates a community-level frailty warning signal according to the updated frailty evaluation parameter set and a preset group risk threshold, solving the problem of low group management efficiency of the traditional scheme and optimizing the resource allocation efficiency.

[0021] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and achieved by the structures indicated in the specification, claims and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0023] Figure 1 is a process schematic diagram of the community frailty screening and intervention training method based on artificial intelligence of the embodiments of the present application.

[0024] Figure 2 is a real-time action monitoring and standard action comparison schematic diagram of the embodiments of the present application.

[0025] Figure 3 is a community frailty risk space release schematic diagram of the embodiments of the present application.

[0026] Figure 4 is a spatiotemporal clustering risk analysis schematic diagram of the embodiments of the present application. DETAILED DESCRIPTION

[0027] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0028] With reference to Figure 1 , one embodiment of the present application proposes a community elderly frailty screening and intervention training method based on artificial intelligence, which adopts the fusion of dynamic physiological parameters and historical medical data of wearable devices, real-time action monitoring triggered by personalized training schemes, can realize dynamic grading evaluation of the frailty state of the elderly, generation of personalized training schemes and real-time optimization of the intensity, and synchronously triggers community-level risk warning signals, significantly improves the community elderly frailty screening ability and the effect of intervention training.

[0029] The method of the embodiment specifically includes:

[0030] Obtaining a set of dynamic physiological parameters collected by a wearable device of a user and a set of preset historical medical data, and generating a set of dynamic data;

[0031] Specifically, the dynamic physiological parameter set is the real-time acquisition of the user's heart rate variability, three-dimensional gait data and activity energy consumption by the intelligent wearable device, reflecting the user's real-time physical state. The dynamic data set is a multi-dimensional dynamic data set generated by fusing the physiological real-time state, pathological history and behavior pattern time series data matrix, providing a holographic analysis basis for frailty assessment and personalized intervention.

[0032] Feature extraction is performed on the dynamic data set to generate a frailty assessment parameter set;

[0033] Specifically, the frailty assessment parameter set is a multi-dimensional index set integrated by physiological function parameters, metabolic and cardiovascular parameters, and clinical correlation parameters obtained by feature extraction, which is used to quantify the user's frailty state.

[0034] According to the frailty assessment parameter set and the preset frailty grade division rule, a target frailty grade is determined;

[0035] Specifically, the frailty grade is a frailty score generated by a multi-class SVM model according to the user's frailty assessment parameter set, combined with the preset frailty grade division rule to obtain the frailty degree division, which provides a clear basis for personalized intervention.

[0036] Based on the target frailty grade and the user's preset tolerance parameter set, a personalized training plan is generated by an intervention strategy generation model;

[0037] Specifically, the intervention strategy generation model generates a personalized training plan based on the target frailty grade and the tolerance parameter by a mixed integer programming combined with a multi-objective genetic algorithm, which fits the Fried frailty phenotype to ensure maximum training effect while minimizing risk.

[0038] According to the personalized training plan, real-time action monitoring is triggered to generate a user action quality evaluation result;

[0039] Specifically, real-time action monitoring captures limb vectors, target centers of gravity and movement rhythms through wearable IMU sensors, calculates action completion scores, and then compares them with preset user action quality evaluation standards to obtain the user action quality evaluation result. Figure 2

[0040] Based on the user action quality evaluation result and the preset baseline error threshold, the training intensity parameter of the personalized training plan is dynamically adjusted;

[0041] ​Specifically, the preset action quality score is divided into a threshold: if the action quality score is greater than or equal to 85, the user is in a frailty low-risk zone, and the training intensity is allowed to increase by 5% per week or the training duration is increased by 10%; if the action quality score is between 75 and 85, the user is in a frailty alert zone, and the current training intensity needs to be maintained, and the current parameters are kept; after the cumulative number of times of meeting the standard is greater than or equal to 3, the training intensity is upgraded; and if the action quality score is less than 75, the user is in a high-risk zone, and needs to be forced to degrade the training.

[0042] The user action quality evaluation result is fed back to an update module of the frailty evaluation parameter set to generate an updated frailty evaluation parameter set.

[0043] Specifically, the action completion time deviation rate and joint pressure distribution parameters in the user action quality evaluation are extracted to construct a standardized dynamic feedback data set; based on a transfer learning model, the data distribution difference is analyzed and the loss function is optimized, and the weight distribution ratio of the frailty score features is dynamically adjusted; finally, the fuzzy clustering algorithm is used to update the class center position, and the support vector machine hyperplane translation strategy is combined to reconstruct the frailty risk classification boundary, so that the dynamic adaptation and accurate classification of the evaluation parameters are realized.

[0044] According to the updated frailty evaluation parameter set and a preset group risk threshold, a community-level frailty early warning signal is generated.

[0045] Specifically, the proportion of high-risk users is obtained from the updated frailty evaluation parameter set, and the average value of the community frailty evaluation parameters is compared with the preset group risk threshold to obtain a community frailty risk space release, such as Figure 3 , to determine the community-level frailty early warning signal, as shown in Figure 4 The position of the target circle is the position of the generated community-level frailty early warning signal.

[0046] Optionally, the step of determining the target frailty level according to the frailty evaluation parameter set and a preset frailty level division rule comprises:

[0047] The first feature screening is performed on the dynamic data set to obtain an initial feature subset.

[0048] Specifically, the feature screening is performed on the dynamic data set containing the wearable device dynamic physiological parameters and the medical history data. The Pearson correlation coefficient method is used to calculate the correlation between each feature parameter and the preset frailty evaluation standard, and the parameters with a correlation coefficient greater than 0.3 are reserved to form the initial feature subset.

[0049] The key frailty index set is extracted from the initial feature subset by using a preset decision tree feature selection model.

[0050] Specifically, the initial feature subset is input into a preset screening model, which is composed of two layers of neural networks, the first layer is a fully connected layer of 64 nodes, and the second layer is a feature compression layer of 6 nodes. The model learns the nonlinear relationship between the features through the back propagation algorithm, and outputs a set of three key frailty indicators including gait stability, resting heart rate oscillation amplitude and electromyographic signal entropy value.

[0051] According to the key frailty indicator set and the preset weight distribution rule, the frailty score value in the frailty evaluation parameter set is calculated.

[0052] Specifically, according to the weight distribution rule determined by the doctor consensus meeting, the frailty score value is calculated by weighted summation according to the proportion of 40% for gait indicators, 35% for heart rate indicators, and 25% for electromyographic indicators. Wherein, GSI is the ratio of gait cycle standard deviation to average value, HRS is the range of adjacent R-R interval difference in resting state, and MSE is the sample entropy of electromyographic signal.

[0053] Exemplarily, the dynamic data of a 65-year-old user includes gait sequence recorded by accelerometer and electrocardiogram monitoring data. The system first calculates the initial feature subset including stride dispersion 0.15 (correlation coefficient 0.4 related to clinical frailty standard), heart rate variability 28 (correlation coefficient 0.35), and electromyographic signal periodicity deviation 0.22 (correlation coefficient 0.25). After training, the screening model identifies gait cycle variation GSI=0.18, resting heart rate oscillation amplitude HRS=32, and electromyographic signal entropy MSE=0.25. According to the weight formula, the frailty score FS=0.169 is calculated, and when the threshold is set to 0.15, it corresponds to the mild frailty level. When the user is trained and intervened, GSI decreases to 0.12, the system automatically updates FS=0.129, and realizes the dynamic degradation of frailty level.

[0054] Optionally, based on the target frailty level and the user's preset tolerance parameter set, the individualized training scheme is generated through the intervention strategy, and the step includes:

[0055] Based on the target frailty level, a candidate intervention mode set is determined from the preset intervention mode library, and according to the cardiopulmonary function indicators in the user's tolerance parameter set, the target intervention mode is selected from the candidate intervention mode set through a mixed integer programming algorithm.

[0056] Specifically, pattern matching is performed based on the target frailty level within a pre-defined intervention pattern library. This library is divided into three levels of storage according to frailty level: mild frailty corresponds to pattern codes L1-L3, moderate frailty to M1-M5, and severe frailty to S1-S4. Each pattern code includes a baseline energy expenditure value (EB) and a movement complexity coefficient (CA) (a normalized value between 0 and 1). After inputting the user's target frailty level into the pattern matching module, all pattern codes matching that level are extracted to form a candidate intervention pattern set. Cardiopulmonary function indicators, including peak oxygen uptake, are loaded from the user's pre-defined tolerance parameter set. And resting heart rate (HRrest). Set the screening criteria as follows: And HRrest is less than or equal to the baseline heart rate threshold HRb. The expected oxygen uptake (HRb) is calculated based on age, sex, and weight, with HRb set to the 75th percentile of the user's age group. For each pattern code in the candidate intervention pattern set, a conditional check is performed, retaining only those patterns that simultaneously meet certain criteria. Entries with CA ≤ 0.5 × (1 - HRrest / HRb) were used as the target intervention pattern.

[0057] By combining the target intervention model with the gait stability index in the set of frailty assessment parameters, training cycle parameters and movement type sequences for personalized training programs are generated.

[0058] Specifically, by combining the muscle activation sequence parameters recorded in the target intervention mode with the gait stability index GS in the set of weakness assessment parameters, training parameters are generated according to the following rules: training cycle length. ,in The maximum complexity coefficient in the selected pattern; the action type sequence is arranged in a three-day cycle, alternating between strength training and balance training action types in the target pattern, and additional posture control training is inserted into the sequence when GS exceeds 0.25.

[0059] For example, a moderately frail user had a VO2peak of 23 mL / kg / min and an HRrest of 78 bpm, resulting in a gait stability index (GS) of 0.28 according to frailty assessment. Pattern matching yielded five candidate patterns (M1-M5) with EB values ​​of 2.1, 2.5, 3.0, 3.2, and 3.8 MET, respectively. These patterns were first screened and retained. The pattern is such that all candidate sets meet the conditions; retain pattern and Combining If the threshold of 0.25 is exceeded, the training cycle parameter TC = 2.32 weeks is generated and rounded down to 2 weeks. Posture control training is inserted into the action sequence every two days to form a three-day cycle unit that includes strength training, balance training and posture control.

[0060] Optionally, the training cycle parameters and the action type sequence for generating the personalized training program further comprise:

[0061] If the gait stability index is lower than a preset balance imbalance threshold, a balance compensation training unit is inserted in the action type sequence.

[0062] Specifically, when the gait stability index GS is calculated from gait cycle data collected by an inertial sensor, specifically, the ratio of the standard deviation of the duration of the single support phase in the last 10 gait cycles to the average value. Set the balance imbalance threshold to 0.3, when GS exceeds the threshold, the system automatically starts the balance compensation training unit insertion mechanism. The unit insertion mechanism rule is to determine the arrangement position of the strength training module in the original action type sequence, and insert a balance compensation training unit after each strength training module, and the insertion frequency is determined by the correction coefficient F, where floor is the floor function. The balance compensation training unit includes lateral step training and heel-to-toe walking training, and each unit lasts minutes.

[0063] When the number of times of executing the balance compensation training unit reaches a preset correction number, the weight value of the gait stability index is updated and the action type sequence is regenerated.

[0064] Specifically, when the number N of cumulative execution of the balance compensation training unit in the action type sequence reaches a preset correction number (default value is 8 times), the system automatically starts the weight value updating process. When updating the weight, first calculate the historical GS fluctuation value ΔG as the difference between the maximum and minimum values of GS in the last N records. Adjust the weight of the original gait stability index in the deterioration score (initial value 0.4) according to the comparison result of ΔG and 0.2, and the new weight where tanh is the hyperbolic tangent function. The updated weight value cannot exceed the range constraint of 0.2-0.6, and finally the action type sequence is regenerated according to the new weight value.

[0065] Exemplarily, the first evaluation GS value of a user is 0.31, triggering F=4 times of balance compensation training insertion. The recorded GS value fluctuates between 0.28-0.33 in the last 8 training processes, and ΔG=0.05. The calculation result is , and the weight is reset to 0.2 due to the lower limit constraint. The system reduces the proportion of gait stability in the overall deterioration evaluation according to the new weight, and generates a new action sequence to reduce the balance compensation training frequency to F=3 times.

[0066] Optionally, the step of triggering real-time action monitoring according to the personalized training scheme to generate the user action quality evaluation result comprises:

[0067] By means of the preset fuzzy logic controller, the joint angle parameter set of the user when performing the training action is collected in real time to generate an action deviation coefficient;

[0068] Specifically, nine-axis inertial measurement data of the user when performing the training action is collected in real time by means of a wearable motion capture device, and the device is attached to the shoulder joint, elbow joint, hip joint and knee joint according to the standard human anatomy position. After the original signal is denoised by means of Kalman filtering, the real-time angle parameter set of each joint in the sagittal plane, coronal plane and horizontal plane is obtained by decomposition , wherein i=1~6 respectively correspond to the left shoulder, right shoulder, left elbow, right elbow, left knee and right knee three-axis angle. The ideal angle at the corresponding time point in the standard action template is taken As a reference, the actual angle is collected every 0.1 seconds in a single action cycle . For each joint, the root mean square value of the three-plane angle deviation is calculated , and the overall action deviation coefficient is , and the weights are respectively 0.25 for the shoulder, 0.2 for the elbow, 0.3 for the hip and 0.25 for the knee.

[0069] The action deviation coefficient is compared with the preset standard action template in the personalized training scheme to generate a completion score value in the user action quality evaluation result. If the completion score value is lower than a preset action qualified line, a strength adjustment instruction of the personalized training scheme is triggered.

[0070] Specifically, the calculation formula of the completion score value Q is , and the action qualified line is set to =75. When the arithmetic mean value of Q values of three consecutive training cycles is lower than 75, the system starts the strength adjustment algorithm: the current training intensity level is determined, the target intensity is set to , wherein is the preset minimum intensity level, and max is the maximum function of the two. At the same time, the number of standard action video demonstration links in the next cycle is increased to twice the original plan.

[0071] Optionally, the dynamic adjustment of the training intensity parameter of the personalized training scheme comprises:

[0072] According to the heart rate recovery rate parameter in the user action quality evaluation result, the real-time metabolic load value of the user is calculated; if the real-time metabolic load value exceeds the preset metabolic safety threshold in the personalized training scheme, the resistance coefficient in the training intensity parameter is reduced; after adjusting the resistance coefficient, the upper limit value of the cardiopulmonary function in the tolerance parameter set is updated synchronously.

[0073] Specifically, the heart rate recovery rate parameter is calculated by using the heart rate decline curve of the user after continuous monitoring of the motion , wherein is defined as the ratio between the time from the peak exercise heart rate to the resting heart rate and the corresponding heart rate drop difference. For the real-time metabolic load value , there is: , wherein is a standard recovery rate reference value based on the age of the user, is an age correction factor obtained by querying a preset age segmentation table, with 1.0 as the reference for users under the age of 40, and a decrease of 0.08 for every additional 5 years.

[0074] Set the metabolic safety threshold When the real-time calculated ML exceeds the threshold, adjust the intensity, and the new training intensity , there is: , wherein , and the adjusted should not be lower than the preset minimum intensity level = 3.

[0075] Optionally, the user action quality evaluation result is returned to the update module of the frailty assessment parameter set to generate an updated frailty assessment parameter set, including:

[0076] Extract the action completion time parameter and joint pressure distribution parameter in the user action quality evaluation result to generate a dynamic feedback data set;

[0077] Specifically, the plantar pressure distribution matrix of the user's training action is collected by the pressure sensing insole and the wearable timer (i=1-8 corresponds to the front palm to the heel partition, and j is the pressure value) and the action completion time . The dynamic feedback data set is constructed as a five-dimensional vector , wherein is the standard completion time ratio ( is the healthy reference value), is the maximum pressure area value, is the pressure distribution deviation sum, is the left and right foot pressure symmetry index, is the time variation coefficient of the last five training times.

[0078] The dynamic feedback dataset is analyzed in association with the set of frailty assessment parameters by a preset migration learning model to update a weight distribution proportion of the frailty score value;

[0079] Specifically, the weight distribution proportion update adopts a sliding window mechanism, and the window size is the last N=5 training records. The Pearson correlation coefficients of each dimension in the dynamic feedback dataset D and the baseline frailty score FS are calculated (k=1-5), the new weight , wherein is the original weight, is the Pearson correlation coefficient of the mth dimension in the dynamic feedback dataset D and the baseline frailty score FS, is the original weight of the mth dimension.

[0080] According to the updated frailty score value, the classification boundary of the target frailty level is re-divided.

[0081] Specifically, for the frailty score , there is: , wherein is the value after min-max normalization of each dimension of the dynamic feedback dataset D. The classification boundary update adopts a bisection method relocation, and the original three level boundary points B1, B2 are dynamically adjusted according to the 25% and 75% quantile values of the current user group distribution to ensure that the proportion of users in each level is balanced.

[0082] Exemplarily, the maximum foot pressure of a certain user in five training records is 280kPa (standard 250kPa), the time ratio =1.15, the pressure deviation =85, the symmetry index =0.72, and the time variation coefficient T_var=0.18. The correlation between each dimension and FS is R=[0.55, 0.68, 0.42, 0.37, 0.61], and the original weight W=[0.3, 0.25, 0.2, 0.15, 0.1]. After calculation, W'=[0.237, 0.213, 0.105, 0.069, 0.076], and normalization [0.33, 0.30, 0.15, 0.10, 0.11]. The new frailty score improves the time and pressure weight, and the user FS_new is upgraded from 62 to 68, which causes the level to be upgraded from Class2 to Class3, triggering higher intensity intervention.

[0083] Optionally, updating the weight distribution proportion of the frailty score value comprises:

[0084] When it is detected that the joint pressure distribution parameter continuously deviates from the preset baseline distribution range, a muscle compensation risk signal is generated.

[0085] Specifically, pressure values ​​in the metatarsal, arch, and calcaneal regions are collected using a plantar pressure sensor array to generate a three-dimensional pressure distribution curve. The baseline distribution range is defined as the spatial envelope of pressure distribution under similar movements in healthy elderly individuals. When the pressure value of a certain region exceeds the boundary of the baseline envelope for five consecutive movement cycles, and the duration exceeds 30% of the total training time, the system determines it to be a continuous deviation state. The overlap (DS) value between the actual pressure distribution and the baseline distribution is calculated using the probability density function comparison method. When DS < 0.6 for 3 minutes, a Level I muscle compensation risk signal is triggered; when DS < 0.4 for 5 minutes, it is upgraded to a Level II muscle compensation risk signal.

[0086] Based on muscle compensation risk signals, the model calibration process for the intervention strategy generation model is triggered.

[0087] Specifically, once the muscle compensation risk signal is triggered, the intensity escalation module of the current exercise mode is immediately stopped, and historical peak metabolic load data is retrieved. The dynamic relationship between joint pressure P and ML was established through linear regression: Where k is the proportional coefficient and b is the compensation value, the allowable load range of each action unit is recalculated using a constrained optimization algorithm to ensure... and .

[0088] The calibrated personalized training program is constrained based on the joint pressure distribution parameters and the metabolic load value.

[0089] Specifically, establish joint constraint equations ,in The maximum plantar pressure value is the joint pressure distribution parameter, ML is the metabolic load value, and the weighting coefficient is... =0.6、 =0.4, This is the safety threshold.

[0090] Optionally, generating community-level weakness early warning signals also includes:

[0091] Obtain the updated set of weakness assessment parameters for all users in the community and generate a population weakness index distribution map.

[0092] Specifically, the user's geographical location With evaluation timestamp By connecting and constructing a spacetime cube, communities are divided into grid units. and time window Population decay index for each unit Defined as: ,in The number of users within a unit is represented by the number of missing units. Missing units are filled by cubic spline interpolation of adjacent spatiotemporal grids, forming a continuous exponential distribution covering the entire community.

[0093] Spatiotemporal clustering analysis was performed on the population weakness index distribution map to identify regions of abrupt change in the weakness index;

[0094] Specifically, the density peak algorithm is used to identify high-density clusters. For each cluster... Calculate its average decay index and compared with the community's historical baseline. Calculate mutation strength ,like Then it is marked as a mutation region. The historical standard deviation of the community, threshold Determined by the significance level.

[0095] If the proportion of users in the area of ​​sudden change in the frailty index exceeds the risk threshold of the group, a targeted health education instruction will be triggered and the warning level of the community-level frailty warning signal will be adjusted.

[0096] Specifically, the percentage of at-risk users is obtained by comparing the total number of users covered by the mutation area set with the total number of users in the community. Compare with the preset group risk threshold If less than or equal to No warning will be issued; if it is greater than Less than or equal to If it triggers a low-risk warning, regular health reminders will be sent; if it is greater than This triggers a high-risk warning and coordinates the intervention of medical resources.

[0097] Optionally, triggering targeted health education instructions and adjusting the warning level of the community-level frailty warning signal further includes:

[0098] Based on the user's nutritional intake data within the frailty index mutation region, a set of recommended parameters for nutritional intervention is generated.

[0099] Specifically, the nutritional intake over the past 7 days is obtained through the user's food tracking app and compared with the recommended values ​​in the "Dietary Guidelines for the Elderly in China." The degree of missing information is defined as follows: ,in denoted as the ratio of the deficiency of the j-th nutrient; Recommended daily intake; This represents the user's actual daily intake. Weights are assigned based on the correlation between nutrients and frailty. Calculate the comprehensive intervention score: Nutrients are prioritized. Based on nutrient deficiency and priority, a daily increment of 15% is applied.

[0100] The set of nutritional intervention recommendations and the updated personalized training plan are pushed synchronously to generate joint intervention instructions;

[0101] Specifically, by extracting the optimal intake time of nutrients from the knowledge base and combining it with the user's personalized training plan, joint timing instructions are generated: ,in This is the start time of the k-th training session; The nutritional supplementation window is 30 minutes to 2 hours after training.

[0102] Once the execution cycle of the joint intervention instruction is completed, the user's dynamic physiological parameter set is re-collected and the group frailty index distribution map is updated.

[0103] Specifically, after the execution cycle ends, dynamic physiological parameters such as the user's resting heart rate variability (HRV) and modified SPPB fitness score are collected through wearable devices. The new dataset is then input into the transfer learning model to update the individual's frailty index. ,in The extent of improvement in the HRV indicator. This is the individual baseline value. The population decay index of the grid cells is recalculated based on the updated decay index. The warning for mutated regions can be lifted or the intervention can be maintained.

[0104] For example, in a certain community, area D, 12 users were screened during the early warning period and found to have a combined deficiency of vitamin D and protein. The system obtained their average daily vitamin D intake of 3.2 μg (10 μg lower than the recommended value) and protein intake of 48g (60g recommended value). The deficiency ratio was calculated to be 68% for vitamin D and 20% for protein. The overall score was... The system recommended a daily intake of 400 IU (equivalent to 10 μg) of vitamin D supplement and a high-protein snack (providing 12g of protein). The combined instruction was set to perform 30 minutes of balance training at 10:00 AM daily, followed by the nutritional supplement at 11:00 AM. After 28 days of implementation, the user's average HRV value increased from 52ms to 68ms, and the system lowered the frailty index for the D-zone group from 0.71 to 0.63, removing the red alert. This method addresses the limitation of single rehabilitation methods by precisely matching nutrients with training time. A dynamic monitoring data-driven scoring update mechanism ensures that the group risk assessment reflects the intervention's effectiveness in real time. The combined instruction push not only improves the convenience of implementation for the elderly but also enhances nutrient absorption through temporal correlation, forming a quantifiable frailty reversal closed loop.

[0105] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for frailty screening and intervention training for elderly people in the community based on artificial intelligence, characterized in that, The method includes: The system acquires a set of dynamic physiological parameters collected by the user's wearable device and a preset set of historical medical data to generate a dynamic data set. Feature extraction is performed on the dynamic data set to generate a set of attrition assessment parameters. The step of generating the set of attrition assessment parameters includes: feature filtering of the dynamic data set to obtain an initial feature subset; extraction of a set of key attrition indicators from the initial feature subset using a preset filtering model; and calculation of attrition score values ​​in the set of attrition assessment parameters based on the set of key attrition indicators and a preset weight allocation rule. The set of attrition assessment parameters includes gait stability indicators. The target attenuation level is determined based on the set of attenuation assessment parameters and the preset attenuation level classification rules. Based on the target weakness level and the user-preset tolerance parameter set, a personalized training plan is generated. Based on real-time motion monitoring of the personalized training program, a user motion quality evaluation result is generated. The generation of the user motion quality evaluation result includes: real-time collection of joint angle parameter sets when the user performs training movements, generating a motion deviation coefficient; comparing the motion deviation coefficient with a preset standard motion template in the personalized training program, generating a completion score in the user motion quality evaluation result; if the completion score is lower than a preset motion qualification line, adjusting the intensity of the personalized training program. Based on the user action quality evaluation results and the preset baseline error threshold, the training intensity parameters of the personalized training scheme are adjusted. The user action quality evaluation results are fed back and the frailty assessment parameter set is updated to generate an updated frailty assessment parameter set. The step of generating the updated frailty assessment parameter set includes: extracting the action completion time parameter and joint pressure distribution parameter from the user action quality evaluation results to generate a dynamic feedback dataset; the dynamic feedback dataset is constructed as a five-dimensional vector. ， Where Tnorm is the standard completion time ratio, Pmax is the maximum pressure region value, PΔ is the sum of pressure distribution deviations, Asym is the left and right foot pressure symmetry index, and Tvar is the coefficient of variation of training time over five consecutive cycles; the dynamic feedback dataset is correlated with the set of frailty assessment parameters to update the weight allocation ratio of the frailty score; the update of the weight allocation ratio of the frailty score includes: calculating the Pearson correlation coefficient Rk between each dimension of the dynamic feedback dataset and the baseline frailty score FS, and calculating the new weights according to the formula. ， Where Wk is the original weight, Rm is the Pearson correlation coefficient between the m-th dimension and the baseline frailty score FS, and Wm is the original weight of the m-th dimension; based on the updated frailty score, the classification boundary of the target frailty level is redefined; the redefined classification boundary of the target frailty level includes: using a binary repositioning method, dynamically adjusting the original level boundary points according to the 25th and 75th percentile values ​​of the updated frailty score distribution of the current user group; Based on the updated set of fragility assessment parameters and the preset group risk threshold, a community-level fragility early warning signal is generated.

2. The method for screening and intervening in frailty among elderly people in a community based on artificial intelligence, as described in claim 1, is characterized in that... The steps for generating personalized training schemes include: Based on the target weakening level, a preset intervention mode library is matched to determine a set of candidate intervention modes; Based on the cardiopulmonary function indicators in the user-preset tolerance parameter set, the target intervention mode is selected; By combining the target intervention mode with the gait stability index in the set of weakness assessment parameters, the training cycle parameters and movement type sequence of the personalized training program are generated.

3. The method for screening and intervening in frailty among elderly people in a community based on artificial intelligence, as described in claim 2, is characterized in that... The training cycle parameters and action type sequence for generating the personalized training plan include: If the gait stability index is lower than the preset balance imbalance threshold, then a balance compensation training unit is inserted into the action type sequence. When the number of times the balance compensation training unit is executed reaches the preset number of corrections, the weight value of the gait stability index is updated and the action type sequence is regenerated.

4. The method for screening and intervening in frailty among elderly people in a community based on artificial intelligence, as described in claim 1, is characterized in that... The steps for adjusting the training intensity parameters of the personalized training program include: Based on the heart rate recovery rate parameter in the user action quality evaluation results, generate the user's real-time metabolic load value; If the real-time metabolic load value exceeds the preset metabolic safety threshold in the personalized training program, then the training intensity parameter is reduced. After the training intensity parameters are adjusted, the cardiopulmonary function indicators in the tolerance parameter set are updated synchronously.

5. The method for screening and intervening in frailty among elderly people in a community based on artificial intelligence, as described in claim 4, is characterized in that... The method further includes: When the joint pressure distribution parameters are detected to deviate continuously from the preset baseline distribution range, a muscle compensation risk signal is generated; Based on the muscle compensation risk signals, a personalized training plan is calibrated and generated. The calibrated personalized training program is constrained based on the joint pressure distribution parameters and the real-time metabolic load value.

6. The method for frailty screening and intervention training of elderly people in the community based on artificial intelligence according to claim 1, characterized in that, The generation of the community-level decay early warning signal includes: Obtain the updated set of weakness assessment parameters for all users in the community and generate a population weakness index distribution map. The population weakness index distribution map was analyzed to identify regions of abrupt change in the weakness index. If the proportion of users in the area of ​​sudden change in the frailty index exceeds the risk threshold of the group, targeted health education will be conducted and the warning level of the community-level frailty warning signal will be adjusted.

7. The method for frailty screening and intervention training of elderly people in the community based on artificial intelligence according to claim 6, characterized in that, The method further includes: Based on the user's nutritional intake data within the frailty index mutation region, a set of recommended parameters for nutritional intervention is generated. The set of nutritional intervention recommendations and the updated personalized training plan are pushed synchronously to generate joint intervention instructions; Once the execution cycle of the joint intervention instruction is completed, the user's dynamic physiological parameter set is re-collected and the group frailty index distribution map is updated.

Images