Digital health management system and method based on cardiopulmonary fitness analysis

Abstract

The invention discloses a digital health management system and method based on cardiopulmonary fitness analysis, and belongs to the technical field of artificial intelligence and digital health. The method comprises the following steps: collecting and preprocessing physiological features and motion data of a user; constructing a data set containing the actually measured maximum oxygen uptake; introducing a sparrow search algorithm to optimize hyper-parameters of the multi-layer perceptron model, and training to obtain a high-precision maximum oxygen uptake prediction model; predicting by using the model; generating a correlation analysis heat map based on the physiological features of the user and the maximum oxygen uptake predicted value; generating a personalized exercise prescription based on the prediction result and an FITT-VP rule base; the maximum oxygen uptake is re-predicted regularly, the change trend is tracked, and the exercise prescription is automatically adjusted; and finally, all information is fed back through a visual interface. The method has the technical advantages of accuracy, robustness and closed loop, breaks through the limitation of a traditional health management mode, and provides a one-stop active health management service for a user.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and digital health technology, and in particular to a digital health management system and method based on cardiopulmonary fitness analysis. Background Technology

[0002] Cardiorespiratory fitness is a core indicator for assessing human health, listed by the American Heart Association as the "fifth vital sign" after respiration, pulse, blood pressure, and body temperature. It plays a crucial role in predicting cardiovascular disease risk, all-cause mortality, and post-surgical risks. Currently, the cardiopulmonary exercise test (CPET) is considered the "gold standard" for assessing cardiorespiratory fitness. This test comprehensively monitors key indicators such as maximum oxygen uptake and anaerobic threshold under exercise load, providing accurate information for exercise prescriptions. However, the CPET testing process is complex, requires highly skilled equipment and operators, is expensive, and inherently carries certain exercise risks, thus hindering its widespread adoption in public health management.

[0003] Existing digital health management platforms attempt to collect data conveniently through wearable devices and other means, and use algorithmic models for cardiopulmonary function assessment and exercise prescription recommendations. However, they generally face two major challenges: first, the predictive accuracy, robustness, and reliability of the assessment models are insufficient; second, the generated prescriptions are mostly general suggestions, lacking deep integration with specific smart fitness equipment, resulting in a disconnect between the "assessment" and "intervention" stages, and failing to form an effective, personalized, and monitorable closed-loop management system. Furthermore, health management platforms still have limitations in data integration, system compatibility, and the depth of intelligent analysis.

[0004] Therefore, there is an urgent need in this field for a digital health management solution based on cardiopulmonary fitness analysis to overcome the problems of low accuracy, poor robustness, and lack of closed-loop management in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a digital health management system and method based on cardiopulmonary fitness analysis, which has the characteristics of high accuracy, strong robustness and closed-loop capability, in order to overcome the shortcomings of existing technologies and promote the transformation from passive medical care to proactive health.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A digital health management method based on cardiorespiratory fitness analysis includes the following steps: S1: Data acquisition and preprocessing, collecting users' physiological characteristic data and exercise data, cleaning the collected data, handling outliers, and smoothing and filtering the heart rate time series data; S2: Feature extraction and dataset construction: Extract feature vectors for model training from the preprocessed data and construct a structured dataset containing the physiological feature data and the corresponding measured maximum oxygen uptake; S3: Model training and optimization. Using the dataset, the sparrow search algorithm is introduced to automatically and globally optimize the key hyperparameters of the multilayer perceptron model. The model is trained with minimizing the root mean square error of the validation set as the optimization objective to obtain the optimized maximum oxygen uptake prediction model. S4: Maximum oxygen uptake prediction. The standardized physiological characteristic data of the user to be evaluated is input into the trained prediction model, and the predicted value of the user's maximum oxygen uptake is output. S5: Interpretability analysis, generating a correlation analysis heatmap based on user physiological characteristics and predicted maximum oxygen uptake; S6: Personalized exercise prescription generation: Based on the predicted maximum oxygen uptake, user physiological characteristics, and the preset FITT-VP exercise prescription rule base, a personalized exercise prescription is automatically generated. S7: By collecting data after user training, it periodically re-predicts their maximum oxygen uptake and tracks the trend of change, and automatically adjusts the exercise prescription based on the trend analysis results; S8: Results visualization and interaction. The predicted values, personalized exercise prescriptions, and correlation analysis heatmaps are presented to the user through a visualization interface.

[0007] Furthermore, the sparrow search algorithm described in S3 optimizes hyperparameters including the number of hidden layers, the number of neurons in each layer, the learning rate, the activation function type, and the batch size.

[0008] Furthermore, the model training described in S3 also includes employing Dropout regularization and a cosine annealing hot restart learning rate scheduling strategy.

[0009] Furthermore, the correlation analysis heatmap described in S5 is transformed into a visual heatmap based on the Pearson correlation coefficient matrix between user physiological characteristics and predicted maximum oxygen uptake.

[0010] Furthermore, the personalized exercise prescription described in S6 includes parameters such as exercise frequency, exercise intensity, duration of each exercise session, and type of exercise.

[0011] Furthermore, the personalized exercise prescription undergoes safety verification based on the user's age and injury history to ensure the safety and effectiveness of the prescription.

[0012] Furthermore, the tracking of trends described in S7 includes: If the trend shows an increase in maximum oxygen uptake, increase the intensity, volume, or complexity of the prescription; If the trend shows that the maximum oxygen uptake remains unchanged or decreases, reduce the prescription strength, volume, or complexity, and analyze and investigate the cause.

[0013] Another objective of this invention is to provide a digital health management system based on cardiopulmonary fitness analysis. When executed, this system implements the aforementioned digital health management method based on cardiopulmonary fitness analysis, comprising a data acquisition module, a data preprocessing module, a machine learning model module, an interpretability analysis module, an exercise prescription generation engine, and a user interface module connected in sequence, wherein: The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps. The data acquisition module is used to acquire the user's physiological characteristics and motion data, and transmit them to the data preprocessing module; The data preprocessing module is used to clean, filter and standardize the received data, and input the processed data into the machine learning model module. The machine learning model module is used to run a prediction model optimized by the sparrow search algorithm to obtain a predicted value of maximum oxygen uptake, and send the predicted value to the interpretability analysis module and the exercise prescription generation engine respectively. The interpretability analysis module is used to analyze the decision basis of the predicted value and submit it to the user interface module for display. An exercise prescription generation engine is used to generate personalized exercise prescriptions based on the predicted values ​​and user characteristics, and to send the prescriptions to the user interaction interface module for display. The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps.

[0014] Another object of the present invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned digital health management method based on cardiopulmonary fitness analysis.

[0015] Another object of the present invention is to provide an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the aforementioned digital health management method based on cardiopulmonary fitness analysis.

[0016] In summary, the present invention has at least one of the following beneficial technical effects: This invention achieves non-invasive, precise, and robust assessment of cardiopulmonary fitness. Its core lies in employing the SSA-MLP model, which can accurately predict VO2max using only easily accessible physiological parameters. Furthermore, it integrates correlation analysis heatmaps, providing transparent and reliable decision-making support for the prediction results. Based on the prediction data, the system can automatically generate FITT-VP exercise prescriptions highly tailored to individual conditions. Through user feedback training data, cardiopulmonary function is reassessed. Based on the reassessment results and trend analysis, the system dynamically optimizes the user's exercise prescription, constructing a closed-loop management mechanism of "assessment-prescription-feedback," improving intervention effectiveness while ensuring safety. In summary, this invention, with its accurate, robust, and closed-loop technical advantages, breaks through the limitations of traditional health management models, providing users with one-stop proactive health management services and strongly supporting the transformation from passive medical care to proactive health management. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the correlation analysis heatmap of the present invention; Figure 3 This is a flowchart of the personalized exercise prescription generation process of the present invention; Figure 4 This is a schematic diagram of the interface for predicting and assessing the maximum oxygen uptake of the population. Figure 5 This is a schematic diagram of the personalized exercise prescription of the present invention; Figure 6 This is a diagram of the overall system architecture of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0019] This embodiment provides a digital health management method based on cardiopulmonary fitness analysis, such as... Figure 1 As shown, it includes the following steps: S1: Data acquisition and preprocessing, collecting users' physiological characteristic data and exercise data, cleaning the collected data, handling outliers, and smoothing and filtering the heart rate time series data; S2: Feature extraction and dataset construction: Extract feature vectors for model training from the preprocessed data and construct a structured dataset containing the physiological feature data and the corresponding measured maximum oxygen uptake; S3: Model training and optimization. Using the dataset, the sparrow search algorithm is introduced to automatically and globally optimize the key hyperparameters of the multilayer perceptron model. The model is trained with minimizing the root mean square error of the validation set as the optimization objective to obtain the optimized maximum oxygen uptake prediction model. S4: Max prediction takes the standardized physiological characteristic data of the user to be evaluated and inputs it into the trained prediction model, and outputs the predicted value of the user's maximum oxygen uptake. S5: Interpretability analysis, generating a correlation analysis heatmap based on user physiological characteristics and predicted maximum oxygen uptake; S6: Personalized exercise prescription generation: Based on the predicted maximum oxygen uptake, user physiological characteristics, and the preset FITT-VP exercise prescription rule base, a personalized exercise prescription is automatically generated. S7: By collecting data after user training, it periodically re-predicts their maximum oxygen uptake and tracks the trend of change, and automatically adjusts the exercise prescription based on the trend analysis results; S8: Results visualization and interaction. The predicted values, personalized exercise prescriptions, and correlation analysis heatmaps are presented to the user through a visualization interface.

[0020] The following is a detailed description with reference to specific embodiments.

[0021] S1: Data acquisition and preprocessing. Collect users' physiological characteristic data and exercise data, and perform data cleaning, outlier handling, and smoothing filtering on heart rate time series data.

[0022] Initial user data is collected through multiple channels: First, basic physiological characteristics, including age, gender, height, weight, and resting heart rate, are obtained through user registration forms or API interfaces with smart wearable devices (such as heart rate bracelets and smartwatches). Simultaneously, peak heart rate, a key exercise data point, is obtained through user self-reporting or analysis of historical heart rate data from wearable devices. After data collection, a rigorous preprocessing procedure is implemented: During the data cleaning phase, the system automatically identifies and handles missing values. For missing key features (such as maximum heart rate), the system prompts the user to re-enter the data or estimates it using a preset algorithm. Simultaneously, reasonable threshold ranges for each parameter are set based on generally accepted physiological principles (e.g., resting heart rate 30-120 beats / min), and outliers significantly exceeding these ranges are identified, marked, corrected, or removed. For time-series data like heart rate, to eliminate unavoidable instantaneous noise and fluctuations during signal acquisition, a sliding window averaging method from digital signal processing is used for smoothing filtering. Typically, a window with a width of 5 data points is set, sliding along the time series and calculating the average value of the data within the window to generate a smooth heart rate curve. Finally, to eliminate the dimensional differences between features and accelerate the convergence of subsequent machine learning models, all continuous features are Z-score standardized, transforming them into a standard normal distribution with a mean of 0 and a standard deviation of 1. After these processes, a clean, standardized, high-quality dataset suitable for model training or prediction is obtained.

[0023] During the model training phase, this invention uses a self-collected Clinical Cardiopulmonary Exercise Test (CPET) dataset, which contains 311 subjects who completed the standard power vehicle test. The input features include age, sex, height, weight, resting heart rate, and maximum heart rate. The target label is obtained directly from measurements taken using a CPET gas analyzer. The maximum value is determined by the system's actual application phase. Basic physiological characteristics (age, gender, height, weight, resting heart rate) and exercise data (maximum heart rate) of the user are collected through user registration forms or the API interface of smart wearable devices.

[0024] After data collection, a unified and rigorous preprocessing workflow is implemented: First, data cleaning is performed, handling missing values ​​(removing samples with missing key features in clinical studies; prompting users to supplement or estimating values ​​using algorithms in system applications); then, outlier handling is performed, setting reasonable ranges based on physiological common sense (e.g., resting heart rate 30-120 beats / min), and correcting or removing outliers that significantly exceed these ranges. For heart rate time-series data, a moving average with a window size of 5 is used for smoothing filtering to eliminate transient noise interference. Finally, to eliminate dimensional differences and accelerate model convergence, all continuous features are Z-score standardized, transforming them into a distribution with a mean of 0 and a standard deviation of 1. After these processes, a high-quality, standardized dataset suitable for model training or prediction is obtained.

[0025] S2: Feature extraction and dataset construction: Extract feature vectors for model training from the preprocessed data and construct a structured dataset containing the physiological feature data and the corresponding measured maximum oxygen uptake.

[0026] 1. Extract feature vectors.

[0027] The system directly extracts key physiological and motor features such as age, gender, height, weight, resting heart rate, and maximum heart rate from preprocessed clean data and encodes them into standardized numerical feature vectors. From preprocessed user data, the system extracts k-dimensional numerical feature vectors that are both physiologically meaningful and easy for the model to process.

[0028] The system directly extracts the following six core physiological and motor characteristics: x1: Age (in years) x2: Gender (numerical encoding: male=0, female=1) x3: Height (cm) x4: Weight (Weight, kg) x5: Resting Heart Rate (bpm) x6: Maximum Heart Rate (bpm) For a user i, its feature vector x i It can be represented as: x i =[x i1 ,x i2 ,x i3 ,x i4 ,x i5 ,x i6 ] Where each component xij Corresponding to one of the features mentioned above, the vector x... i ∈R6 is a 6-dimensional real vector.

[0029] 2. Construct a structured dataset.

[0030] Based on this, the feature vector of each sample is compared with its true maximum oxygen uptake obtained directly through the gold standard of medicine—the cardiopulmonary exercise test (CPET). The feature vectors of all samples are then precisely correlated and labeled with their corresponding (max) values. The maximum label values ​​are structurally integrated to construct a complete structured dataset that can be used for model training and validation. This dataset forms the foundation for subsequent model training and achieving accurate predictions. Dataset construction is a process of systematically associating and integrating feature vectors with real labels.

[0031] For each feature vector x i This is linked to a measured maximum oxygen uptake (VO2 max) directly measured using the gold standard in medicine—the cardiopulmonary exercise test (CPET). The maximum value is denoted as y. i (Unit: mL / kg / min). This process ensures the accuracy and authority of the data labels.

[0032] A complete training sample i consists of its feature vector and its label, and can be represented as an ordered pair: S i =(x i ,y i ), where x i y is the feature vector of the i-th user. i Is this the real user? The maximum value.

[0033] If valid data from N users is collected, N samples {S1, S2, ..., S} are obtained through the above process. N By systematically organizing these samples, the structured dataset D required by this invention can be constructed.

[0034] The dataset D can be precisely represented by an N×(k+1) matrix: In this matrix, each row represents an independent user sample S. i The first k columns are the user's feature vector x. i The components of each dimension (basic feature k=6 in this example), the last column is the true value corresponding to this sample. max tag yi .

[0035] The matrix is ​​usually randomly shuffled and divided into training, validation and test sets according to a certain ratio, which are used for model training, hyperparameter optimization and final performance evaluation, respectively.

[0036] The final output of this step is a high-quality, structured dataset D, which will be used for model training and optimization in subsequent steps to achieve high accuracy. The foundation of max prediction. This dataset meets the requirements of machine learning for data format: numerical features, accurate labels, and normalized structure.

[0037] S3: Model training and optimization. Using the dataset, the sparrow search algorithm is introduced to automatically and globally optimize the key hyperparameters of the multilayer perceptron model. The model is trained with the goal of minimizing the root mean square error of the validation set, resulting in an optimized maximum oxygen uptake prediction model.

[0038] The core objective of model training and optimization is to construct a high-precision, highly robust maximum oxygen uptake (VO2 max) model. The prediction model is based on a benchmark model. The entire process is divided into three key stages: benchmark model comparison, intelligent hyperparameter optimization, and final model training.

[0039] Phase 1: Performance comparison of benchmark models. To determine the most suitable benchmark model for this research task, a systematic performance comparison of three mainstream machine learning models was first conducted using the same structured dataset. Specifically, three machine learning and deep learning methods were used to predict maximal oxygen uptake, including Random Forest (RF), XGBoost, and Multilayer Perceptron (MLP). Mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) were selected as evaluation metrics for model detection. Maximum oxygen uptake was used as the detection label. This experiment employed a 10-fold cross-validation method, and the average of the 10 validation results was selected as the final detection result. The detection results of the three models are shown in Table 1.

[0040] Table 1. Detection results of three algorithm models As shown in Table 1, all models achieved above-average prediction accuracy (R² > 0.65). Among them, MLP performed best with an R² of 0.736, which was higher than RF and XGboost. However, its MAE and RMSE were 5.85% and 3.64, respectively, both lower than RF and XGboost. Therefore, MLP has a better predictive effect in predicting VO2 max during cardiopulmonary exercise testing and should be used as the benchmark model.

[0041] Phase 2: Hyperparameter optimization based on the sparrow search algorithm. After determining the MLP as the base model, an advanced metaheuristic optimization algorithm—the Sparrow Search Algorithm (SSA)—is introduced to automatically and globally optimize the key hyperparameters of the MLP model. SSA simulates the cooperative behavior of sparrow populations during foraging and predation (including the role division of finders, joiners, and watchdogs), demonstrating powerful global search capabilities and a fast convergence speed. Its optimization process is as follows: Optimization objective: Minimize the RMSE of the MLP model on the independent validation set as the fitness function of SSA.

[0042] In the process of optimizing MLP models using the Sparrow Search Algorithm (SSA), the fitness function is the sole criterion for evaluating the quality of each hyperparameter combination (i.e., each individual "sparrow").

[0043] For the set of hyperparameters λ currently evaluated by SSA, first train an MLP model using the training set, and then let it predict on the independent validation set. The validation set contains M samples {(x1,y1),...,(x...y1)}. M ,y M The model will output the corresponding predicted values ​​{y1,...,y}. M}

[0044] The root mean square error of the validation set, i.e., the root mean square error between the predicted and actual values, is calculated using the following formula: This value intuitively reflects the model's average prediction bias on unknown data; the smaller the value, the stronger the model's accuracy and generalization ability. This RMSE value is directly defined as the fitness value F(λ) of this set of hyperparameters: F(λ) = RMSE(λ).

[0045] Therefore, the optimization objective of SSA is to find the combination of hyperparameters that minimizes the fitness function F(λ). This process is equivalent to having SSA act as an intelligent search engine, efficiently exploring a vast hyperparameter space and continuously filtering out the elite configurations that enable the model to achieve the highest score (lowest error) in the "mock test" (validation set).

[0046] Hyperparameter space: Key hyperparameters for optimization include the number of hidden layers, the number of neurons in each layer, the learning rate, the type of activation function, and the batch size. These parameters are encoded as position vectors of individual sparrows in the search space.

[0047] In the sparrow search algorithm, the position of each individual sparrow in the D-dimensional search space represents a complete hyperparameter combination scheme. This position is represented as a D-dimensional vector: X i =[x i1 ,x i2 ,...,x iD ] In this context, the subscript i identifies the i-th sparrow individual in the population, and each dimension x of the vector represents a different sparrow. id (d=1,2,...,D) corresponds to a hyperparameter that needs to be optimized, where D is the total number of hyperparameters.

[0048] Decoding the position vector, position vector X i Each component x in id Internally, this is a continuous real value. To convert it into the actual hyperparameter value λ required for model training... d Decoding is required: λ d =Decode(x id ) The decoding function Decode(.) depends on the type of the hyperparameters: For integer hyperparameters (such as the number of layers or neurons): rounding is usually used, for example, λ. d =round(x id ).

[0049] For continuous hyperparameters (such as the learning rate): linear scaling or exponential transformation may be involved to map to the target interval. For categorical hyperparameters (such as activation functions): by using continuous values ​​x id The domain is divided into several non-overlapping intervals, and each interval maps to a specific category.

[0050] The position matrix P of the entire sparrow population can be represented as: Where N is the number of sparrows in the population.

[0051] The sparrow search algorithm iteratively updates the position matrix P, allowing the sparrows to move and explore in the hyperparameter space, ultimately finding the position that satisfies the fitness function F(X). i The optimal position vector that minimizes the RMSE (i.e., the validation set RMSE) This allows us to determine the optimal combination of hyperparameters.

[0052] Optimization Process: SSA initializes a sparrow population, with each individual representing a set of hyperparameter combinations. For each individual (i.e., each set of hyperparameters), a corresponding MLP model is constructed and trained on the training set. Subsequently, its RMSE (Recovery Mean Squared Error) is calculated on the validation set as the individual's fitness value. The SSA algorithm iteratively updates the population position according to the sparrows' swarm intelligence behavior rules, continuously exploring the hyperparameter space, aiming to find the optimal hyperparameter combination that minimizes the validation set RMSE, effectively avoiding getting trapped in local optima. Finally, an optimal set of hyperparameters is obtained through SSA optimization, forming an SSA-MLP hybrid optimization model.

[0053] The core of this process is to combine the swarm intelligence search capability of SSA with the training process of MLP models. In essence, it is a nested optimization problem: the inner layer is the optimization of model weights under fixed hyperparameters, and the outer layer is the hyperparameter space search driven by SSA.

[0054] 1. Initialization SSA initializes a population of N sparrows, where the position vector Xi of each sparrow is randomly generated within a predefined D-dimensional search space Λ. P(0)={X1(0),X2(0),...,X N (0)},Xi(0)∈Λ where (0) represents the 0th generation (initial generation).

[0055] 2. Fitness assessment For each individual X in the t-th generation population i (t): a) Decoding: Decoding the position vector into a specific hyperparameter combination λ i =Decode(X i (t)).

[0056] b) Model building and training (inner layer optimization): using hyperparameter λ i Building an MLP model f θ (·;λi). On the training set Dtrain, the optimal model weight parameters θλ are solved by minimizing the training loss Ltrain (e.g., mean squared error) through backpropagation and gradient descent algorithms (e.g., Adam). i : c) Calculate fitness: Perform predictions on the validation set Dval using the trained model, and calculate the root mean square error as the fitness value F(X) for that individual sparrow. i (t)): This F is the target function that drives the SSA to perform a global search.

[0057] 3. Position Update The SSA sorts according to the fitness value, divides the population into discoverers, joiners, and vigilants, and updates their positions respectively. The core update formula is as follows: Position update of discoverers (individuals with excellent fitness): Where: R2 ∈ [0, 1] is the early warning value, and ST ∈ [0.5, 1.0] is the safety threshold.

[0058] T max is the maximum number of iterations, and α is a random number.

[0059] Q is a random number obeying the normal distribution, and L is a matrix of all 1s.

[0060] When it is safe (R2 < ST), the discoverers conduct extensive searches; when there is danger (R2 ≥ ST), they fly to the safe area quickly.

[0061] 4. Termination and Output Repeat steps 2 and 3 until the maximum number of iterations Tmax or the fitness converges. Finally, output the global optimal position vector , and decode it to obtain the optimal hyperparameter combination . <00002​​​​​​​​​​​​​​​​​​​​​​​​​Therefore, the SSA-MLP hybrid optimization model is a final predictor refined through a two-layer optimization process (swarm intelligence search + gradient descent), possessing both high accuracy and high generalization ability. It not only has the powerful nonlinear fitting capability of MLP, but also the specific capabilities granted by SSA. Optimal architecture and hyperparameter configuration for the max prediction task.

[0066] Phase 3: Final model training and advanced regularization strategies.

[0067] The optimal hyperparameter configuration obtained through SSA optimization is used to reconstruct and train the final MLP model. At this stage, two important techniques are introduced to further improve the model's generalization ability and convergence performance: Dropout regularization: To effectively suppress overfitting, a common phenomenon in deep neural networks, a Dropout layer is introduced after the hidden layers of an MLP. Specifically, the Dropout technique randomly "drops" (i.e. temporarily ignores) the output of a portion of neurons during training. This forces the network to avoid over-reliance on any single neuron or feature combination, thereby learning more robust feature representations and enhancing the model's generalization ability.

[0068] Cosine annealing hot-restart learning rate scheduling: A cosine annealing hot-restart strategy is used to dynamically adjust the learning rate during training. This strategy causes the learning rate to decrease from its initial value according to a cosine function curve, and then "hot-restarts" at the end of a preset epoch, resetting the learning rate to a higher value (not starting from zero). This periodic restart mechanism helps the optimization process escape potentially trapped local optima, increases the likelihood of finding a better global solution, and accelerates model convergence.

[0069] Finally, using the Adam optimizer, the MLP model integrating Dropout regularization and cosine annealing learning rate scheduling strategies was fully trained on all available training data, resulting in a high-performance, highly robust final model. The maximum prediction model (SSA-MLP) is serialized and saved for subsequent predictions.

[0070] 1. Model Architecture and Regularization: The Introduction of Dropout The optimal hyperparameters were found using SSA. To construct the final MLP model, a Dropout regularization layer is introduced after the hidden layers of the MLP to further enhance the model's generalization ability and prevent overfitting on the training data.

[0071] Dropout operation: During the training phase of the model, for the output vector h of each hidden layer, Dropout will randomly set the output values ​​of some neurons to zero (i.e. "discard") with a preset probability p, while amplifying the values ​​of the retained neurons by a factor of 1 / (1-p) to keep the overall expected value unchanged.

[0072] Mathematical expression: For the output h(l) of a layer, after Dropout it becomes h1(l): in: ⊙ indicates element-wise multiplication (Hadamard product).

[0073] m is a mask vector, where each element independently has a probability p of 0 and a probability p of 0. The value is 1.

[0074] denominator It is a scaling factor used to approximately maintain the total strength of the layer's output during training.

[0075] Mechanism of action: This random "deactivation" forces the network to not rely on any specific combination of neurons and to learn more redundant and robust feature representations, thereby effectively suppressing overfitting.

[0076] 2. Optimization process and cosine annealing hot restart The Adam optimizer is used to minimize the loss function on the training set, and a cosine annealing hot restart strategy is used to dynamically adjust the learning rate during this process. Loss function: Where θ is the model weight parameter.

[0077] Cosine annealing hot restart: Periodization: Divide the entire training process into multiple periods, each period containing T i One training round.

[0078] Learning rate variation: The learning rate η during each period t Based on the cosine function, from the initial value η max Descending to η min : Where Tcur is the number of rounds executed in the current cycle.

[0079] Hot restart: When a cycle ends (Tcur=T i The learning rate will immediately reset to its initial value η. max(or a value close to it), instead of starting from zero and then starting the next cycle.

[0080] Periodic increases in the learning rate help the optimization process escape the current local optima, increasing the chances of finding a better global optimum or a flatter region of minima. At the beginning of each cycle, a higher learning rate can accelerate the model's convergence in the initial stages.

[0081] 3. Determination of the final model After completing all the preset training cycles, the model parameters from the last training step are usually not selected; instead, the set of parameters that performs best on the independent validation set is chosen. This serves as the final model weight. This further ensures the model's generalization performance.

[0082] final The expression for the max prediction model After the complete training process described above, the final result was obtained. The max prediction model is an integrated model that incorporates optimal hyperparameters. Dropout regularization strategy, and optimized by cosine annealing hot restart, resulting in a weight of The MLP model.

[0083] When making predictions, with Dropout disabled (meaning all neurons participate without scaling), the model degenerates into a deterministic, powerful nonlinear function. This is the version delivered to the end user.

[0084] final The expression for the max prediction model is: in: x: The input feature vector, i.e., the user's standardized physiological data; The optimal combination of hyperparameters obtained by SSA optimization defines the network structure; :exist The optimal weight parameters obtained under advanced training strategies (Dropout, cosine annealing); f: represents the complex, nonlinear function mapping relationship implemented by the entire multilayer perceptron; y: The output predicted value, i.e. the user's estimated maximum oxygen uptake; The final (x; The model is the result of automated hyperparameter search and advanced regularization and optimization techniques. It not only captures input features but also... It possesses the ability to understand complex relationships between max values, and boasts excellent generalization performance and robustness, enabling high-precision, non-invasive assessment of users' cardiorespiratory fitness based on readily available physiological parameters.

[0085] S4: Max prediction takes the standardized physiological characteristics data of the user to be evaluated and inputs them into a trained prediction model, and outputs the predicted value of the user's maximum oxygen uptake.

[0086] The system uses standardized physiological characteristic data of the user to be evaluated (such as age, gender, height, weight, resting heart rate, and maximum heart rate) as input feature vectors, which are then loaded into a pre-trained SSA-MLP prediction model. The model performs forward propagation calculations, weighted summation of each hidden layer, and activation function transformations, ultimately generating a continuous value at the output layer. This value represents the system's prediction of the user's maximum oxygen uptake. The maximum value is used to complete a non-invasive and convenient quantitative assessment of cardiopulmonary function.

[0087] S5: Interpretability analysis, generating a correlation analysis heatmap based on user physiological characteristics and predicted maximum oxygen uptake.

[0088] Correlation analysis heatmaps are used to analyze the feature usage patterns of the SSA-MLP model, enhancing the transparency and reliability of prediction results. Specifically, the system calculates the correlation between input features (user physiological characteristics) and model predictions. The Pearson correlation coefficient matrix between the maximum values ​​is then visualized as a heatmap. This method first collects all input features and corresponding predicted outputs of the model on the validation set, and then calculates the correlations at the following two levels: The interrelationships between the input features; Each input feature and the final The association strength of the maximum predicted values.

[0089] The strength of the linear relationship between features and the predicted target is visually represented by color intensity and numerical labels: warm colors (such as red) indicate a positive correlation, cool colors (such as blue) indicate a negative correlation, and darker colors indicate a stronger correlation. For example, a heatmap can clearly show the relationship between maximum heart rate and... The maximum predicted value showed a significant positive correlation, while age showed a significant negative correlation.

[0090] like Figure 2As shown, this heatmap, based on the Pearson correlation coefficient, quantitatively analyzes the linear correlation between seven physiological and exercise characteristics: age, gender, height, weight, resting heart rate, maximum heart rate, and maximum oxygen uptake. The Pearson correlation coefficient ranges from -1 to 1; the closer the absolute value is to 1, the stronger the linear correlation; the closer it is to 0, the weaker the linear correlation. Positive and negative signs correspond to positive and negative correlation directions, respectively. The heatmap uses color gradients to visually represent the magnitude of the coefficients, with red indicating positive correlation and blue indicating negative correlation. Color saturation is positively correlated with the strength of the correlation. From the feature pairs, height and weight show a significant positive correlation (r = 0.66), consistent with common physiological knowledge and reflecting the intrinsic relationship of human morphological characteristics; modeling should pay attention to collinearity. Height, weight, and gender (r = 0.63 and 0.60, respectively) show a moderate positive correlation, suggesting that the average height and weight of men in the dataset are higher than that of women; this difference can be verified by grouping. Maximum heart rate and maximum oxygen uptake (r = 0.55) show a moderate positive correlation, consistent with the logic of exercise physiology, and can be used as a joint reference for exercise prescription formulation. Regarding strong negative correlations, age and maximum heart rate (r = -0.85) strictly conform to the physiological decline pattern, providing a basis for exercise risk assessment. The correlation coefficients between gender and resting heart rate (r = 0.07) and between height and resting heart rate (r = 0.04) are close to 0, indicating that resting heart rate is only weakly affected by these correlations and is more related to the cardiovascular health baseline. In terms of research and application value, the feature correlations are consistent with the consensus of physiological / exercise science, verifying the rationality of the dataset and facilitating subsequent modeling.

[0091] S6: Personalized exercise prescription generation: Based on the predicted maximum oxygen uptake, user physiological characteristics, and the preset FITT-VP exercise prescription rule base, a personalized exercise prescription is automatically generated.

[0092] This step is the core link connecting cardiopulmonary function assessment and specific exercise intervention. It aims to automatically generate scientific, personalized, and actionable personalized exercise prescriptions (plans) based on predicted VO2max values, individual user physiological characteristics, and health goals. This process is driven by a pre-defined exercise prescription rule base built on the FITT-VP scientific principles. This rule base integrates authoritative guidelines from the American College of Sports Medicine with association rules obtained through historical data mining using the Apriori algorithm.

[0093] The system first predicts The maximum value is mapped to the user's fitness level, and then the prescription generation engine performs rule-based reasoning and AI fine-tuning. Based on the FITT-VP principle, the engine calculates exercise prescription parameters that precisely match the user's current state, specifically including: exercise frequency, exercise intensity (quantified as a percentage of maximum heart rate or oxygen reserve), duration of each exercise session, and exercise type (e.g., recommended aerobic exercise, strength training, or flexibility exercises). This process performs safety checks based on the user's age and injury history, ensuring that the prescription parameters are within physiologically safe ranges (if the user's injury history includes cardiovascular disease risk, the exercise intensity parameters are automatically reduced), guaranteeing the safety and effectiveness of the prescription. Ultimately, this forms a structured preliminary plan that can be dynamically adjusted as the user's fitness progresses. Through the above processing, the system finally outputs a personalized, quantifiable, and executable exercise training plan that conforms to the principles of sports science and can directly guide specific practices. Figure 3 As shown in the figure, this is the process of generating a personalized exercise prescription.

[0094] 1. Input parameters Collect all personalized information needed to generate a prescription. The input data is divided into four categories: Max predicted value: Represents the user's cardiopulmonary function level and is a key indicator for measuring physical fitness.

[0095] User physiological characteristics: such as gender, height, weight, resting heart rate and other basic physiological data.

[0096] Health goals: Specific goals that users hope to achieve, such as losing fat, gaining muscle, improving endurance, or improving health indicators.

[0097] Constraints: The most important safety considerations include the user's age and medical history to ensure the safety and feasibility of the prescription.

[0098] 2. FITT-VP Exercise Prescription Rule Base Processing Transforming input parameters into preliminary prescription plans, the FITT-VP rule base is a professional sports science knowledge base that provides a basic framework for prescriptions. It contains six core elements: Frequency: Number of times you exercise per week.

[0099] Intensity: The degree of effort exerted during exercise.

[0100] Time: The duration of each movement.

[0101] Type: The type of exercise (such as running, swimming, strength training, etc.).

[0102] Total load: The overall motion load consisting of frequency, intensity, and time.

[0103] Advanced: As physical fitness improves, how to gradually increase exercise load.

[0104] 3. Basic prescription generation: The system will generate the user's prescription. Data such as `max` is mapped to a specific "fitness level". Based on this level, rule reasoning is performed in the FITT-VP rule base to generate a basic prescription. Next, AI algorithms are used to fine-tune this basic prescription to make it more accurate. Finally, a crucial safety check is performed to ensure the plan meets constraints such as the user's age and medical history.

[0105] 4. Output personalized exercise prescriptions After processing, a customized and executable exercise prescription (plan) is generated for the user. The prescription will clearly provide specific recommendations under the FITT-VP framework, such as: Exercise frequency: 3-5 times / week.

[0106] Exercise intensity: Keep it within 60%-70% of your maximum heart rate (HRmax).

[0107] Exercise duration: 30-45 minutes each time.

[0108] Exercise type: It is recommended to combine aerobic exercise with strength training.

[0109] S7: By collecting user training data, it periodically re-predicts their maximum oxygen uptake and tracks the trend of change, and automatically adjusts the exercise prescription based on the trend analysis results.

[0110] Data-driven periodic reassessment and program optimization: This step is the core of achieving a closed loop in personalized health management, forming a continuous optimization chain of "assessment—prescription—implementation—feedback—reassessment". After a user completes a phase of training on smart fitness equipment (such as a smart Pilates bed), the system initiates the data feedback and program optimization process.

[0111] 1. Training data collection and feedback During the training regimen, the sensor array integrated into the smart fitness equipment (including inertial measurement unit, force sensor, and heart rate monitoring device) continuously records the user's kinematic data (such as movement trajectory, peak strength, and number of sets completed) and physiological data (such as average heart rate and heart rate recovery rate). After a single training session or a training cycle, this detailed training performance data is automatically uploaded and integrated into the user's personal health record.

[0112] 2. Cardiorespiratory fitness reassessment and trend analysis The system uses new data accumulated over a period of time (e.g., every 2 to 4 weeks) to trigger a reassessment of cardiorespiratory fitness. Specifically: The user's latest basic physiological data (such as weight and resting heart rate) and periodic training performance data are used as new input features.

[0113] The feature vector is input into a pre-trained SSA-MLP VO2 max prediction model to re-predict the user's current VO2 max. (max) value.

[0114] The system compares the current predicted value with historical predicted values ​​longitudinally, and generates... The maximum change trend chart is used, supplemented by statistical analysis (such as calculating the rate of change and conducting significance tests) to objectively quantify the degree of improvement, stability or decline in the user's cardiorespiratory fitness.

[0115] 3. Adaptive adjustments to exercise prescriptions are based on reassessment results and trend analysis; the system dynamically optimizes users' exercise prescriptions: If the trend shows Significant improvement in max: This indicates that the current prescription is effective. The system will automatically increase the prescription strength, capacity or complexity according to the advanced principles in the FITT-VP rule base to provide new training stimuli and promote continuous improvement in cardiorespiratory fitness.

[0116] If the trend shows If the maximum value does not change significantly or decreases: The system will analyze the training data, investigate possible causes (such as poor training compliance, insufficient intensity, etc.), and adjust the prescription accordingly, reducing the intensity, volume, or complexity of the prescription, such as modifying the type of exercise, adjusting the intensity range, or increasing the training frequency, in order to break through the plateau.

[0117] At the same time, prescription adjustments will take into full account the user's latest physiological state to ensure that any modifications are within a safe range.

[0118] Through this closed-loop feedback mechanism, the system has evolved from a single assessment intervention to a long-term, adaptive, and sustainable health management model, ensuring that exercise prescriptions are always highly matched with the user's dynamically changing physiological state and health goals, thereby maximizing the long-term intervention effect.

[0119] S8: Results visualization and interaction. The predicted values, personalized exercise prescriptions, and correlation analysis heatmaps are presented to the user through a visualization interface.

[0120] Through the visual interactive interface of the health management platform ( Figure 4 This interface presents the results of all the aforementioned steps to the user in an intuitive and easy-to-understand format, completing the final loop of information delivery. The core function of the interface is to transform complex professional data into multi-dimensional, interactive visual elements: 1. Evaluation Results Presentation: Clearly Presented The maximum predicted value and its percentile ranking among its peers allow users to clearly understand their cardiorespiratory fitness level.

[0121] 2. By having the user input data and click the prediction button, the system can generate corresponding predictions, including... The maximum predicted value and its percentile ranking among its peers allow users to clearly understand their cardiorespiratory fitness level.

[0122] 3. Transparent Decision-Making Interpretation: Physical condition is reflected by combining correlation analysis heatmaps and visualization charts with natural language descriptions.

[0123] This heatmap, based on the Pearson correlation coefficient, quantitatively analyzes the linear correlations among seven physiological and exercise characteristics: age, sex, height, weight, resting heart rate, maximum heart rate, and maximum oxygen uptake. The Pearson correlation coefficient ranges from -1 to 1; the closer the absolute value is to 1, the stronger the linear correlation; the closer it is to 0, the weaker the linear correlation. Positive and negative signs correspond to the positive and negative correlation directions, respectively. The heatmap uses color gradients to visually represent the coefficient magnitude, with red indicating positive correlation and blue indicating negative correlation. Color saturation is positively correlated with correlation strength. From the feature pairs, height and weight show a significant positive correlation (r = 0.66), consistent with physiological common sense, reflecting the inherent correlation of human morphological characteristics; collinearity needs to be considered in modeling. Height, weight, and sex (r = 0.63 and 0.60, respectively) show moderate positive correlations, suggesting that the average height and weight of men in the dataset are higher than that of women; this difference can be verified by further grouping. Maximum heart rate and maximum oxygen uptake show a moderate positive correlation (r = 0.55), consistent with exercise physiology logic, and can be used as a joint reference for exercise prescription formulation. Regarding strong negative correlations, the correlation coefficient between age and maximum heart rate (r = -0.85) strictly conforms to the physiological decline pattern, providing a basis for exercise risk assessment. However, the correlation coefficients between gender and resting heart rate (r = 0.07) and between height and resting heart rate (r = 0.04) are close to 0, indicating that resting heart rate is minimally affected by these correlations and is more correlated with baseline cardiovascular health.

[0124] Personalized exercise prescriptions are presented in various formats, including Gantt charts, structured tables, and concise text descriptions. Personalized exercise prescriptions are generated based on the user's predicted cardiorespiratory function level, such as... Figure 5 As shown, the interface displays motion schemes from four perspectives: frequency, intensity, time, and type, helping users to understand and implement them more easily. The visual interface also supports in-depth interactive queries (such as clicking on charts to view details) and one-click export of complete reports, greatly enhancing the user experience and system usability.

[0125] This embodiment also provides a digital health management system based on cardiopulmonary fitness analysis. When executed, this system implements the aforementioned digital health management method based on cardiopulmonary fitness analysis, such as... Figure 6As shown, it includes a data acquisition module, a data preprocessing module, a machine learning model module, an interpretability analysis module, an exercise prescription generation engine, and a user interface module connected in sequence, specifically: The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps.

[0126] The data acquisition module is used to acquire the user's physiological characteristics and motion data, and transmit them to the data preprocessing module.

[0127] The data preprocessing module is used to clean, filter, and standardize the received data, and then input the processed data into the machine learning model module.

[0128] The machine learning model module is used to run a prediction model optimized by the sparrow search algorithm to obtain a predicted value of maximum oxygen uptake, and then send the predicted value to the interpretability analysis module and the exercise prescription generation engine respectively.

[0129] The interpretability analysis module is used to analyze the decision-making basis of the predicted values ​​and submit them to the user interface module for display.

[0130] An exercise prescription generation engine is used to generate personalized exercise prescriptions based on the predicted values ​​and user characteristics, and then distribute the prescriptions to the user interaction interface module for display.

[0131] The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps.

[0132] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned digital health management method based on cardiopulmonary fitness analysis.

[0133] This embodiment provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform a digital health management method based on cardiopulmonary fitness analysis.

[0134] In summary, this invention, with its accurate, robust, and closed-loop technical advantages, breaks through the limitations of traditional health management models, provides users with one-stop proactive health management services, and strongly supports the transformation from passive medical care to proactive health management.

[0135] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A digital health management method based on cardiopulmonary fitness analysis, characterized in that, Includes the following steps: S1: Data acquisition and preprocessing, collecting users' physiological characteristic data and exercise data, cleaning the collected data, handling outliers, and smoothing and filtering the heart rate time series data; S2: Feature extraction and dataset construction: Extract feature vectors for model training from the preprocessed data and construct a structured dataset containing the physiological feature data and the corresponding measured maximum oxygen uptake; S3: Model training and optimization. Using the dataset, the sparrow search algorithm is introduced to automatically and globally optimize the key hyperparameters of the multilayer perceptron model. The model is trained with minimizing the root mean square error of the validation set as the optimization objective to obtain the optimized maximum oxygen uptake prediction model. S4: Maximum oxygen uptake prediction. The standardized physiological characteristic data of the user to be evaluated is input into the trained prediction model, and the predicted value of the user's maximum oxygen uptake is output. S5: Interpretability analysis, generating a correlation analysis heatmap based on user physiological characteristics and predicted maximum oxygen uptake; S6: Personalized exercise prescription generation: Based on the predicted maximum oxygen uptake, user physiological characteristics, and the preset FITT-VP exercise prescription rule base, a personalized exercise prescription is automatically generated. S7: By collecting data after user training, it periodically re-predicts their maximum oxygen uptake and tracks the trend of change, and automatically adjusts the exercise prescription based on the trend analysis results; S8: Results visualization and interaction. The predicted values, personalized exercise prescriptions, and correlation analysis heatmaps are presented to the user through a visualization interface.

2. The digital health management method based on cardiopulmonary fitness analysis according to claim 1, characterized in that: The sparrow search algorithm described in S3 optimizes hyperparameters including the number of hidden layers, the number of neurons in each layer, the learning rate, the activation function type, and the batch size.

3. The digital health management method based on cardiopulmonary fitness analysis according to claim 1, characterized in that, The model training described in S3 also includes the use of Dropout regularization and cosine annealing hot restart learning rate scheduling strategies.

4. The digital health management method based on cardiopulmonary fitness analysis according to claim 1, characterized in that, The correlation analysis heatmap described in S5 is a visual heatmap based on the Pearson correlation coefficient matrix between user physiological characteristics and predicted maximum oxygen uptake.

5. A digital health management method based on cardiopulmonary fitness analysis according to claim 1, characterized in that, The personalized exercise prescription described in S6 includes parameters such as exercise frequency, exercise intensity, duration of each exercise session, and type of exercise.

6. A digital health management method based on cardiopulmonary fitness analysis according to claim 5, characterized in that, The personalized exercise prescriptions are validated for safety based on the user's age and injury history to ensure the safety and effectiveness of the prescriptions.

7. A digital health management method based on cardiopulmonary fitness analysis according to claim 1, characterized in that, The tracking of changing trends described in S7 includes: If the trend shows an increase in maximum oxygen uptake, increase the intensity, volume, or complexity of the prescription; If the trend shows that the maximum oxygen uptake remains unchanged or decreases, reduce the prescription strength, volume, or complexity, and analyze and investigate the cause.

8. A digital health management system based on cardiopulmonary fitness analysis, characterized in that, When executed, this system implements a digital health management method based on cardiopulmonary fitness analysis as described in any one of claims 1-7, comprising a data acquisition module, a data preprocessing module, a machine learning model module, an interpretability analysis module, an exercise prescription generation engine, and a user interface module connected in sequence, wherein: The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps. The data acquisition module is used to acquire the user's physiological characteristics and motion data, and transmit them to the data preprocessing module; The data preprocessing module is used to clean, filter and standardize the received data, and input the processed data into the machine learning model module. The machine learning model module is used to run a prediction model optimized by the sparrow search algorithm to obtain a predicted value of maximum oxygen uptake, and send the predicted value to the interpretability analysis module and the exercise prescription generation engine respectively. The interpretability analysis module is used to analyze the decision basis of the predicted value and submit it to the user interface module for display. An exercise prescription generation engine is used to generate personalized exercise prescriptions based on the predicted values ​​and user characteristics, and to send the prescriptions to the user interaction interface module for display. The user interface module is used to receive user input and comprehensively display predicted values, personalized exercise prescriptions, and correlation analysis heatmaps.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements a digital health management method based on cardiopulmonary fitness analysis as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store computer programs, and the processor running the computer programs to cause the electronic device to perform a digital health management method based on cardiopulmonary fitness analysis as described in any one of claims 1-7.

Images