Method and system for intelligent guidance of cardiac rehabilitation training based on motion data analysis
By integrating cross-modal coupling analysis of electrocardiogram signals and motion data, and utilizing spatiotemporal convolutional long short-term memory networks and cross-modal attention mechanisms, the problem of insufficient risk assessment accuracy in cardiac rehabilitation training was solved, enabling personalized dynamic exercise intensity control and improving the safety and intelligence of training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE SECOND AFFILIATED HOSPITAL OF ANHUI UNIVERSITY OF TRADITIONAL CHINESE MEDICINE (ACUPUNCTURE AND MOXIBUSTION HOSPITAL OF ANHUI PROVINCE)
- Filing Date
- 2026-04-18
- Publication Date
- 2026-07-14
AI Technical Summary
Existing cardiac rehabilitation training methods neglect the dynamic coupling relationship between cardiac electrophysiological response and exercise load, and lack multimodal data fusion, resulting in insufficient accuracy of risk assessment and difficulty in achieving personalized and dynamic rehabilitation guidance.
By fusing electrocardiogram signals and motion data, a cross-modal coupling analysis mechanism is constructed. A spatiotemporal convolutional long short-term memory network is used to extract physiological and motion feature vectors. A coupling index is generated by combining a cross-modal attention mechanism, and the exercise intensity is dynamically adjusted to achieve risk identification and personalized control.
It enables precise risk identification and personalized dynamic control of exercise intensity during cardiac rehabilitation training, significantly improving the safety and intelligence of training.
Smart Images
Figure CN122392802A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of rehabilitation data analysis, and in particular to a method and system for intelligent guidance of cardiac rehabilitation training based on exercise data analysis. Background Technology
[0002] With the development of cardiovascular disease rehabilitation concepts, exercise-based cardiac rehabilitation training has become an important means of clinical and home-based rehabilitation. In practice, patients usually need to conduct regular training under the guidance of an exercise prescription formulated by a doctor, and use wearable devices to monitor heart rate or electrocardiogram signals to assess exercise safety and training effectiveness.
[0003] In existing technologies, most cardiac rehabilitation monitoring programs rely on a single physiological indicator for judgment, such as adjusting exercise intensity based on whether the heart rate exceeds the target heart rate range, or identifying potential risks through simple electrocardiogram abnormality detection.
[0004] However, these methods neglect the dynamic coupling between cardiac electrophysiological responses and exercise load, making it difficult to promptly identify non-pathological deviations such as heart rate drift or delayed physiological responses. Furthermore, multimodal data fusion is insufficient; most methods only perform simple data splicing or rule-based judgments, failing to deeply characterize the intrinsic relationship between ECG signals and exercise behavior, resulting in insufficient accuracy in risk assessment. For high-risk conditions such as occult myocardial ischemia, traditional methods typically rely on static indicators or single thresholds, lacking dynamic constraint mechanisms that incorporate exercise status, easily leading to misjudgments or omissions. Existing systems often use fixed thresholds, failing to consider individual differences and physiological adaptation changes during long-term training, making it difficult to achieve personalized and dynamic rehabilitation guidance. Therefore, an intelligent guidance method is needed that can integrate ECG signals and exercise data, characterize the dynamic coupling between physiological responses and exercise load, and achieve accurate risk identification and adaptive adjustment of exercise intensity. Summary of the Invention
[0005] This application aims to at least partially address one of the technical problems in the related art.
[0006] Therefore, one objective of this application is to propose an intelligent guidance method and system for cardiac rehabilitation training based on exercise data analysis. This invention achieves accurate risk identification and personalized dynamic control of exercise intensity during cardiac rehabilitation training by integrating electrocardiogram signals and exercise data and constructing a cross-modal coupling analysis mechanism, thereby significantly improving the safety and intelligence level of training.
[0007] To achieve the above objectives, the first aspect of this application proposes a method for intelligent guidance of cardiac rehabilitation training based on exercise data analysis, comprising the following steps:
[0008] Step S10: Real-time acquisition of dynamic electrocardiogram waveform data sequence of the patient during rehabilitation training, and motion characteristic time-series data output by a multi-axis inertial sensor synchronized with the dynamic electrocardiogram waveform data sequence;
[0009] Step S20: Input the dynamic electrocardiogram waveform data sequence into the first spatiotemporal convolutional long short-term memory network to extract physiological feature vectors characterizing cardiac electrophysiological morphology; at the same time, input the motion feature time series data into the second spatiotemporal convolutional long short-term memory network to extract motion feature vectors characterizing motion dynamics.
[0010] Step S30: Input the physiological feature vector and the motion feature vector into the cross-modal attention mechanism module, and generate a numerical coupling index to characterize the real-time coupling state between the current physiological response and the exercise load by calculating the intermodal correlation weight between the physiological feature vector and the motion feature vector.
[0011] Step S40: Based on the comparison between the coupling index and the preset dynamic stability threshold, determine whether the coupling index is less than the dynamic stability threshold. If it is less, determine that there is a non-pathological deviation state in which the cardiac electrophysiological changes lag behind the changes in exercise intensity.
[0012] Step S50: When the non-pathological deviation state is determined to occur, a cardiac function compensation risk warning signal is generated, and the current exercise prescription intensity value is dynamically reduced according to the warning signal.
[0013] The intelligent guidance method and system for cardiac rehabilitation training based on exercise data analysis in this application realizes accurate risk identification and personalized dynamic control of exercise intensity during cardiac rehabilitation training by integrating electrocardiogram signals and exercise data and constructing a cross-modal coupling analysis mechanism, thereby significantly improving the safety and intelligence level of training.
[0014] In addition, the intelligent guidance method and system for cardiac rehabilitation training based on exercise data analysis proposed in this application may also have the following additional technical features:
[0015] In one embodiment of this application, the first spatiotemporal convolutional long short-term memory network includes a first spatiotemporal convolutional neural network branch and a first long short-term memory network branch connected in sequence;
[0016] Step S20 specifically includes:
[0017] The first spatiotemporal convolutional neural network branch performs spatiotemporal convolution operation on the dynamic electrocardiogram waveform data sequence to extract local electrophysiological feature maps reflecting ST segment morphology and QRS complex temporal changes.
[0018] The local electrophysiological feature map is then input into the first long short-term memory network branch to capture the long-term dependencies of the local electrophysiological feature map and output the physiological feature vector.
[0019] In one embodiment of this application, the second spatiotemporal convolutional long short-term memory network includes a second spatiotemporal convolutional neural network branch and a second long short-term memory network branch connected in sequence;
[0020] The motion characteristic time-series data includes step frequency data and acceleration rate of change data;
[0021] Step S20 further includes:
[0022] The second spatiotemporal convolutional neural network branch performs convolution operations on the step frequency data and acceleration rate of change data to extract local motion feature maps that reflect changes in motion intensity and body posture.
[0023] The local motion feature map is then input into the second long short-term memory network branch to capture the temporal evolution pattern of the local motion feature map and output the motion feature vector; and the motion intensity is calculated based on the motion feature vector.
[0024] In one embodiment of this application, step S30 specifically includes:
[0025] Using the physiological feature vector as the query term and the motion feature vector as the key and value terms, the attention weight distribution is calculated.
[0026] Based on the attention weight distribution, a coupling index is calculated to characterize the degree of matching between the physiological feature vector and the motion feature vector;
[0027] The coupling index is a statistical metric calculated based on the concentration of the attention weight distribution, and the statistical metric includes information entropy or its monotonic transformation value.
[0028] In one embodiment of this application, step S40 further includes:
[0029] The ST segment offset in the dynamic electrocardiogram waveform data sequence is obtained, and differential calculation is performed based on the ST segment baseline value of the patient at rest.
[0030] When the exercise intensity is lower than a preset intensity threshold, it is determined whether the ST segment offset exceeds a preset ischemia threshold;
[0031] If the threshold is exceeded, it is determined to be a state of myocardial ischemia risk, and a movement termination command or a manual intervention command is output.
[0032] In one embodiment of this application, the motion feature time-series data further includes attitude angle data;
[0033] The motion feature vector carries a pattern label for distinguishing different motion patterns, including walking, jogging, or upper limb activity.
[0034] In one embodiment of this application, the dynamic stability threshold is obtained by dynamically updating the coupling index in historical rehabilitation training data through statistical analysis, and the dynamic stability threshold is the lower bound of the statistical confidence interval based on the distribution of the coupling index.
[0035] To achieve the above objectives, a second aspect of this application proposes an intelligent guidance system for cardiac rehabilitation training based on exercise data analysis. The system includes: a data acquisition module for real-time acquisition of dynamic electrocardiogram waveform data sequences and time-series data of exercise characteristics; a feature extraction module for extracting physiological feature vectors and exercise feature vectors; a coupling analysis module for generating a numerical coupling index characterizing the coupling state between physiological response and exercise load; a risk determination module for determining whether a non-pathological deviation or myocardial ischemia risk state has occurred based on the coupling index and a dynamic stability threshold; and a feedback execution module for generating a risk warning signal and dynamically adjusting the intensity of the exercise prescription based on the determination result.
[0036] Based on the exercise data analysis of this application, a smart guidance method and system for cardiac rehabilitation training is presented.
[0037] 1. By fusing dynamic electrocardiogram and multi-axis inertial sensor data through a cross-modal attention mechanism, real-time coupling quantification of cardiac electrophysiological response and exercise load is achieved. The coupling index can accurately identify heart rate drift, non-pathological deviations, and the risk of occult myocardial ischemia, improving the accuracy and reliability of risk assessment.
[0038] 2. Based on dynamically updated thresholds and patients' historical training data, personalized exercise prescriptions can be adjusted. In the event of a risky situation, the exercise intensity can be reduced in real time or a manual intervention command can be triggered to ensure the safety and effectiveness of rehabilitation training.
[0039] 3. Multimodal data fusion and time-series analysis enhance the system's ability to recognize different movement patterns (walking, jogging, upper limb activities), supporting wearable devices or cloud deployment to achieve end-to-end intelligent rehabilitation management. Compared with existing technologies, this invention significantly improves the safety, dynamic adaptability, and clinical application value of the training process.
[0040] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0041] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0042] Figure 1 A flowchart of the intelligent guidance method for cardiac rehabilitation training based on exercise data analysis, as described in this application;
[0043] Figure 2 This is a block diagram of the intelligent guidance system for cardiac rehabilitation training based on exercise data analysis, as described in this application. Detailed Implementation
[0044] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.
[0045] This embodiment provides an intelligent guidance method for cardiac rehabilitation training based on exercise data analysis. It can monitor the dynamic coupling relationship between the patient's cardiac electrophysiological response and exercise load in real time during rehabilitation training. When a decline in cardiac reserve function or a lag in the electrophysiological response compared to changes in exercise intensity occurs, the exercise prescription intensity is adjusted promptly. This achieves safe and personalized cardiac rehabilitation training while avoiding pathological events such as myocardial ischemia. The intelligent guidance method and system for cardiac rehabilitation training based on exercise data analysis according to this application embodiment are described below with reference to the accompanying drawings.
[0046] like Figure 1 As shown, the intelligent guidance method for cardiac rehabilitation training based on exercise data analysis in this application embodiment may include the following steps:
[0047] Step S10: Real-time acquisition of dynamic electrocardiogram waveform data sequence of the patient during rehabilitation training, as well as motion characteristic time-series data output by a multi-axis inertial sensor synchronized with the dynamic electrocardiogram waveform data sequence;
[0048] Specifically, before a patient begins rehabilitation training, the rehabilitation physician sets an initial exercise prescription based on the patient's medical history and cardiopulmonary exercise test results, including the target heart rate zone, exercise mode (such as walking, jogging) and initial exercise intensity (such as pace or power).
[0049] The patient wears a multimodal wearable device that integrates a single-lead or multi-lead dynamic electrocardiogram (ECG) electrode module and a six-axis or nine-axis inertial measurement unit. The device acquires real-time dynamic ECG waveform data sequences at a sampling rate of 250Hz to 500Hz, and simultaneously acquires motion characteristic timing data synchronized with the ECG waveform data sequences at a sampling rate of 100Hz to 200Hz. The motion characteristic timing data includes: three-axis acceleration data, three-axis angular velocity data, and cadence, rate of change of acceleration, and posture angle data calculated from the above data. The device transmits the data in real-time to a rehabilitation training terminal (such as a tablet, mobile phone, or dedicated computing box) via Bluetooth or near-field communication.
[0050] To ensure time alignment between the two modalities, the system uses a unified timestamp reference during data acquisition, with the ECG R-wave detection time as the alignment anchor point. Linear interpolation resampling is performed on the motion data to ensure point-to-point temporal correspondence between the motion and ECG data. Through high-precision, multimodal synchronous acquisition, a foundation for time alignment between cardiac electrophysiology and motion dynamics is established, providing accurate data support for subsequent cross-modal correlation analysis.
[0051] Step S20: Input the dynamic electrocardiogram waveform data sequence into the first spatiotemporal convolutional long short-term memory network to extract the physiological feature vectors representing the electrophysiological morphology of the heart; at the same time, input the motion feature time series data into the second spatiotemporal convolutional long short-term memory network to extract the motion feature vectors representing the motion dynamics characteristics.
[0052] Understandably, this step extracts physiological feature vectors and motion feature vectors through two independent spatiotemporal convolutional long short-term memory networks.
[0053] In one embodiment of this application, the first spatiotemporal convolutional long short-term memory network includes a first spatiotemporal convolutional neural network branch and a first long short-term memory network branch connected in sequence.
[0054] Step S20 specifically includes: performing spatiotemporal convolution operations on the dynamic electrocardiogram waveform data sequence through the first spatiotemporal convolutional neural network branch to extract local electrophysiological feature maps that reflect the ST segment morphology and QRS complex temporal changes; then inputting the local electrophysiological feature maps into the first long short-term memory network branch to capture the long-term dependencies of the local electrophysiological feature maps and outputting physiological feature vectors.
[0055] Specifically, the first spatiotemporal convolutional neural network branch employs multiple one-dimensional convolutional kernels. The first layer of the convolutional kernel is set to have 64 sampling points (approximately 128ms to 256ms) and a stride of 4, used to capture the QRS complex morphology and ST segment morphology within a single cardiac cycle. Subsequently, pooling layers are used to reduce the dimensionality, and multiple layers of small-sized convolutional kernels (e.g., size 16) are used to further extract local spatiotemporal features. The final output is a local electrophysiological feature map reflecting the ST segment morphology, T wave changes, and QRS complex temporal variation patterns.
[0056] Subsequently, the local electrophysiological feature map is output to the first Long Short-Term Memory (LSTM) network branch. This branch employs a two-layer stacked LSTM structure, with each layer containing 128 hidden units, to capture the long-term dependencies of the local electrophysiological feature map across multiple consecutive cardiac cycles. The first LSTM branch outputs a 256-dimensional physiological feature vector, denoted as... Meanwhile, the temporal data of motion features are input into the second spatiotemporal convolutional long short-term memory network.
[0057] In one embodiment of this application, the second spatiotemporal convolutional long short-term memory network includes a second spatiotemporal convolutional neural network branch and a second long short-term memory network branch connected in sequence.
[0058] Motion characteristic time-series data includes step frequency data and acceleration rate of change data;
[0059] Step S20 further includes: performing convolution operations on the step frequency data and acceleration change rate data through the second spatiotemporal convolutional neural network branch to extract local motion feature maps that reflect the changes in motion intensity and body posture; then inputting the local motion feature maps into the second long short-term memory network branch to capture the temporal evolution pattern of the local motion feature maps and outputting motion feature vectors; and calculating the motion intensity based on the motion feature vectors.
[0060] In one embodiment of this application, the motion feature time series data further includes posture angle data; the motion feature vector carries a mode label for distinguishing different motion modes, including walking, jogging, or upper limb activity.
[0061] Specifically, simultaneously, the temporal data of motion features are input into a second spatiotemporal convolutional long short-term memory network. This network structure is symmetrical to the first network, including a second spatiotemporal convolutional neural network branch and a second long short-term memory network branch. The second spatiotemporal convolutional neural network branch performs multi-channel convolution operations on the cadence data, acceleration rate of change data, and posture angle data to extract local motion feature maps reflecting changes in motion intensity and body posture. The second long short-term memory network branch further captures the evolution of the local motion feature maps along the time axis, outputting a 256-dimensional motion feature vector, denoted as... Meanwhile, this embodiment calculates the current motion intensity scalar value based on the acceleration amplitude and step frequency information in the motion feature vector through a pre-calibrated regression model. This value is quantified in metabolic equivalents (METs) or watts (W).
[0062] By employing the aforementioned dual-tower spatiotemporal convolutional long short-term memory network, cardiac electrophysiological features and motion dynamics features are extracted separately, achieving decoupled representation of heterogeneous modal data. Compared to traditional single-threshold monitoring or manual feature engineering methods, this step utilizes deep learning to automatically extract deep features, capturing high-dimensional features that are difficult to quantify manually, such as subtle ST segment changes, heart rate variability, and gait stability, significantly improving the richness and robustness of feature representation. Simultaneously, the independent feature extraction architecture avoids feature interference between modalities, providing clean feature input for subsequent cross-modal attention analysis.
[0063] Step S30: Input the physiological feature vector and the motion feature vector into the cross-modal attention mechanism module. By calculating the intermodal correlation weight between the physiological feature vector and the motion feature vector, a numerical coupling index is generated to characterize the real-time coupling state between the current physiological response and the exercise load.
[0064] In one embodiment of this application, step S30 specifically includes:
[0065] Using physiological feature vectors as query terms and motion feature vectors as key and value terms, the attention weight distribution is calculated. Based on the attention weight distribution, a coupling index is calculated to characterize the degree of matching between physiological feature vectors and motion feature vectors. The coupling index is a statistical measure calculated based on the concentration of the attention weight distribution, and the statistical measure includes information entropy or its monotonic transformation value.
[0066] Specifically, step S20 above will convert the physiological feature vector obtained in step S20 into... As a query item , to transform the motion feature vector Simultaneously serving as a key item Sum of values Specifically, through three learnable linear transformation matrices , , Mapping the original features to the same high-dimensional space, the dimension of the mapped space... Set to 128. After mapping, the result is:
[0067] , , ;
[0068] Where Q, K, and V all have dimensions R128. Then, the dot product of the query term and the key term is calculated, and scaling is applied to stabilize the gradient: The score reflects the similarity between cardiac electrophysiological characteristics and motion characteristics at the current moment—the higher the score, the better the match between cardiac response and exercise load.
[0069] To obtain a probability distribution that can be used to quantify the coupling strength, the score is transformed into an attention weight distribution using a softmax function. : In the case of a single time step, It is a scalar of 1; however, in actual sequence processing, the system simultaneously inputs the physiological feature vectors and motion feature vectors from the most recent 8 time windows (each window being 2 seconds), at which point... and Both are 8×128 matrices; their dot product yields an 8×8 weight matrix. Each line represents the level of attention paid to various exercise time points at a given physiological time point.
[0070] The concentration of attention weight distribution A directly reflects the tightness of coupling between physiological response and exercise load. To transform this into an intuitive numerical indicator, this embodiment calculates the information entropy of A. :
[0071] ;in Let be the i-th element in the attention weight distribution A (when processing a sequence, iterate through all attention weight values). Information entropy. The range of values is ; for The number of elements. When attention weights are highly concentrated on a few elements, the entropy value is small; when the weights are evenly distributed, the entropy value is large.
[0072] To obtain an index consistent with the physical meaning (the higher the coupling degree, the larger the exponent), the coupling exponent C is defined as: As can be seen from the definition, When C approaches 1, it indicates that attention is highly focused, the physiological response closely follows the exercise load, and the heart-motor system is in a good coupling state. When C approaches 0, it indicates that attention is evenly dispersed, the physiological response lags behind the changes in exercise load, that is, a non-pathological deviation state of "cardiac electrophysiological changes lag behind changes in exercise intensity" has occurred.
[0073] This embodiment quantifies the dynamic coupling degree between cardiac response and exercise load by calculating the attention weight distribution between physiological and motor characteristics. Compared to traditional methods that rely solely on the linear relationship between heart rate and exercise intensity, this step can capture nonlinear, temporal coupling relationships, exhibiting higher sensitivity and specificity. In particular, converting information entropy into a coupling index transforms the previously abstract "physiological-motor matching degree" into an intuitive [0,1] interval index, facilitating clinical understanding and threshold setting. This coupling index can detect early signals of declining cardiac reserve function several minutes before pathological events such as myocardial ischemia occur, providing a valuable time window for timely intervention.
[0074] Step S40: Based on the comparison between the coupling index and the preset dynamic stability threshold, determine whether the coupling index is less than the dynamic stability threshold. If it is less, determine that there is a non-pathological deviation state in which the changes in cardiac electrophysiology lag behind the changes in exercise intensity. This step is based on the comparison results of the coupling index and the dynamic stability threshold, as well as the results of traditional ST segment deviation monitoring, to comprehensively determine the patient's current state.
[0075] In one embodiment of this application, the dynamic stability threshold is obtained by dynamically updating the coupling index in historical rehabilitation training data through statistical analysis. The dynamic stability threshold is the lower bound of the statistical confidence interval based on the coupling index distribution.
[0076] The system maintains a dynamically stable threshold. This threshold is obtained through statistical analysis of the coupling index in the patient's rehabilitation training data over the past 7 days, specifically taken as the 5th percentile of the coupling index distribution (i.e., the lower bound of the 95% confidence interval). This threshold is dynamically updated as the training progresses, reflecting individual differences and changes in training adaptation.
[0077] If the coupling index C is less than the dynamic stability threshold If the ST segment shift is detected, indicating a non-pathological deviation where cardiac electrophysiological changes lag behind changes in exercise intensity, this state is clinically correlated with decreased cardiac reserve, myocardial fatigue, or deadaptation, serving as an early warning signal before pathological events such as myocardial ischemia. During this process, the system monitors the ST segment shift in the dynamic electrocardiogram waveform data sequence in real time.
[0078] In one embodiment of this application, step S40 further includes:
[0079] The ST segment offset in the dynamic electrocardiogram waveform data sequence is obtained, and differential calculation is performed based on the ST segment baseline value of the patient at rest.
[0080] When the exercise intensity is lower than the preset intensity threshold, it is determined whether the ST segment offset exceeds the preset ischemia threshold; if it does, it is determined to be a myocardial ischemia risk state, and an exercise termination command or manual intervention command is output.
[0081] Specifically, using the PR or TP segment as a baseline in the patient's resting state, the ST segment shift 60 to 80 ms after the J point in each cardiac cycle is calculated. This is based on the current exercise intensity. The intensity is below a preset intensity threshold (e.g., 3 METs), while the ST segment offset exceeds a preset ischemia threshold (e.g., horizontal or downsloping depression). If the condition is positive, it is considered a state of myocardial ischemia risk.
[0082] When both the non-pathological deviation state and the myocardial ischemia risk state are met simultaneously, the system prioritizes the myocardial ischemia risk state and outputs a termination of exercise command or a manual intervention command.
[0083] Understandably, dynamic stability thresholds are updated individually based on patients' historical data. Compared to fixed thresholds, this allows for adaptive tracking of patients' recovery progress and physiological changes, avoiding false positives or false negatives caused by individual differences. Simultaneously, clear priority settings (myocardial ischemia > non-pathological deviations) ensure the most decisive intervention is taken in the most dangerous situations.
[0084] Step S50: When a non-pathological deviation is detected, a cardiac function compensation risk warning signal is generated, and the current exercise prescription intensity value is dynamically reduced based on the warning signal.
[0085] For example, this signal can manifest as a yellow warning on the terminal interface, a vibration alert, or a voice announcement stating "Heart response is slow, intensity will be reduced soon." The system dynamically reduces the current exercise prescription intensity value according to a preset step-down strategy, such as reducing the target METs by 0.5 or reducing the treadmill pace by 0.3 km / h. After the reduction, the system continues to execute steps S10 to S40, forming a closed-loop feedback control until the coupling index recovers to above the dynamic stability threshold.
[0086] If a state of myocardial ischemia risk is determined, a red high-level warning signal will be generated immediately, and a movement termination command will be output. At the same time, a rehabilitation physician will be notified or an emergency plan will be triggered.
[0087] like Figure 2 As shown, a smart guidance system for cardiac rehabilitation training based on exercise data analysis is implemented using a smart guidance method for cardiac rehabilitation training based on exercise data analysis. The system includes:
[0088] The data acquisition module is used to acquire dynamic electrocardiogram waveform data sequences and motion characteristic time series data in real time;
[0089] The feature extraction module is used to extract physiological feature vectors and motion feature vectors;
[0090] The coupling analysis module is used to generate numerical coupling indices to characterize the coupling state between physiological response and exercise load;
[0091] The risk assessment module is used to determine whether a non-pathological deviation or myocardial ischemia risk state has occurred based on the coupling index and dynamic stability threshold.
[0092] The feedback execution module is used to generate risk warning signals based on the judgment results and dynamically adjust the intensity of the exercise prescription.
[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent guidance of cardiac rehabilitation training based on exercise data analysis, characterized in that, Includes the following steps: Step S10: Real-time acquisition of dynamic electrocardiogram waveform data sequence of the patient during rehabilitation training, and motion characteristic time-series data output by a multi-axis inertial sensor synchronized with the dynamic electrocardiogram waveform data sequence; Step S20: Input the dynamic electrocardiogram waveform data sequence into the first spatiotemporal convolutional long short-term memory network to extract physiological feature vectors characterizing cardiac electrophysiological morphology; at the same time, input the motion feature time series data into the second spatiotemporal convolutional long short-term memory network to extract motion feature vectors characterizing motion dynamics. Step S30: Input the physiological feature vector and the motion feature vector into the cross-modal attention mechanism module, and generate a numerical coupling index to characterize the real-time coupling state between the current physiological response and the exercise load by calculating the intermodal correlation weight between the physiological feature vector and the motion feature vector. Step S40: Based on the comparison between the coupling index and the preset dynamic stability threshold, determine whether the coupling index is less than the dynamic stability threshold. If it is less, determine that there is a non-pathological deviation state in which the cardiac electrophysiological changes lag behind the changes in exercise intensity. Step S50: When the non-pathological deviation state is determined to occur, a cardiac function compensation risk warning signal is generated, and the current exercise prescription intensity value is dynamically reduced according to the warning signal.
2. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, The first spatiotemporal convolutional long short-term memory network includes a first spatiotemporal convolutional neural network branch and a first long short-term memory network branch connected in sequence; Step S20 specifically includes: The first spatiotemporal convolutional neural network branch performs spatiotemporal convolution operation on the dynamic electrocardiogram waveform data sequence to extract local electrophysiological feature maps reflecting ST segment morphology and QRS complex temporal changes. The local electrophysiological feature map is then input into the first long short-term memory network branch to capture the long-term dependencies of the local electrophysiological feature map and output the physiological feature vector.
3. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, The second spatiotemporal convolutional long short-term memory network includes a second spatiotemporal convolutional neural network branch and a second long short-term memory network branch connected in sequence; The motion characteristic time-series data includes step frequency data and acceleration rate of change data; Step S20 further includes: The second spatiotemporal convolutional neural network branch performs convolution operations on the step frequency data and acceleration rate of change data to extract local motion feature maps that reflect changes in motion intensity and body posture. The local motion feature map is then input into the second long short-term memory network branch to capture the temporal evolution pattern of the local motion feature map and output the motion feature vector; and the motion intensity is calculated based on the motion feature vector.
4. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, Step S30 specifically includes: Using the physiological feature vector as the query term and the motion feature vector as the key and value terms, the attention weight distribution is calculated. Based on the attention weight distribution, a coupling index is calculated to characterize the degree of matching between the physiological feature vector and the motion feature vector; The coupling index is a statistical metric calculated based on the concentration of the attention weight distribution, and the statistical metric includes information entropy or its monotonic transformation value.
5. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, Step S40 further includes: The ST segment offset in the dynamic electrocardiogram waveform data sequence is obtained, and differential calculation is performed based on the ST segment baseline value of the patient at rest. When the exercise intensity is lower than a preset intensity threshold, it is determined whether the ST segment offset exceeds a preset ischemia threshold; If the threshold is exceeded, it is determined to be a state of myocardial ischemia risk, and a movement termination command or a manual intervention command is output.
6. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, The motion feature time-series data also includes attitude angle data; The motion feature vector carries a pattern label for distinguishing different motion patterns, including walking, jogging, or upper limb activity.
7. The intelligent guidance method for cardiac rehabilitation training based on exercise data analysis according to claim 1, characterized in that, The dynamic stability threshold is obtained by dynamically updating the coupling index in historical rehabilitation training data through statistical analysis. The dynamic stability threshold is the lower bound of the statistical confidence interval based on the distribution of the coupling index.
8. A smart guidance system for cardiac rehabilitation training based on exercise data analysis, implemented based on the smart guidance method for cardiac rehabilitation training based on exercise data analysis as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to acquire dynamic electrocardiogram waveform data sequences and motion characteristic time series data in real time; The feature extraction module is used to extract physiological feature vectors and motion feature vectors; The coupling analysis module is used to generate numerical coupling indices to characterize the coupling state between physiological response and exercise load; The risk assessment module is used to determine whether a non-pathological deviation state or a myocardial ischemia risk state has occurred based on the coupling index and the dynamic stability threshold. The feedback execution module is used to generate risk warning signals based on the judgment results and dynamically adjust the intensity of the exercise prescription.