A method, device, and storage medium for predicting cardiopulmonary exercise test data
By employing a reservoir calculation model and a composite loss function training method, the problem of individual differences and dynamic changes in the prediction of low-intensity cardiopulmonary exercise test data was solved, enabling efficient and personalized cardiopulmonary function assessment under low-intensity exercise, which is applicable to primary healthcare institutions and home environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-09
AI Technical Summary
Existing methods for predicting low-intensity cardiopulmonary exercise test data are difficult to achieve personalized adaptive prediction, cannot effectively cope with individual differences and dynamic changes, and have insufficient prediction accuracy. They are especially difficult to replace traditional cardiopulmonary function assessments in people who cannot tolerate high-intensity exercise.
A reservoir calculation model is adopted, which dynamically updates the reservoir state by inputting physiological signals and control variables. The model is trained by combining ridge regression algorithm and composite loss function, and Bayesian optimization is used to find the hyperparameters to achieve individualized prediction.
It improves the accuracy and individualized adaptability of cardiopulmonary function assessment under low-intensity exercise conditions, reduces equipment and venue requirements, is suitable for primary healthcare institutions and home environments, and provides safe and efficient cardiopulmonary function prediction.
Smart Images
Figure CN122177457A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of cardiopulmonary exercise testing technology, specifically relating to a method, device, equipment, and storage medium for predicting cardiopulmonary exercise test data. Background Technology
[0002] Cardiopulmonary function assessment is a key aspect of modern medical diagnosis and health management. It can reveal the health status of the heart and lungs and is not only essential for the diagnosis and treatment of many cardiopulmonary diseases, but also an indispensable tool in health management and sports training.
[0003] The gold standard for assessing cardiopulmonary function is the cardiopulmonary exercise testing (CPET). This method involves monitoring multiple physiological parameters, including electrocardiogram (ECG), expiratory gas composition (such as oxygen uptake (VO2) and carbon dioxide excretion (VCO2), and ventilation volume), during progressively increasing exercise loads. This comprehensive assessment evaluates the response of the heart, lungs, skeletal muscles, and central nervous system to exercise load. However, the CPET test demands a high level of physical fitness from the subject, typically requiring progressively strenuous exercise on a treadmill or stationary bike to accurately capture physiological inflection points such as maximum oxygen uptake. This high-intensity exercise poses certain risks and performance challenges for individuals with poor cardiopulmonary reserve, and is particularly unsuitable for the elderly, patients with heart failure, those with respiratory dysfunction, patients recovering from major surgery, obese individuals with limited mobility, those with weak constitutions, or those in a chronically sub-healthy state. Furthermore, CPET has high requirements for equipment configuration, site conditions, and the professionalism of operators. The entire testing process is time-consuming and costly, making it difficult to widely deploy in resource-constrained primary healthcare institutions, community health service systems, and home environments. These factors mean that although CPET is the gold standard, it is insufficient to meet the larger-scale assessment needs of the general population, chronic disease management, and public health promotion. There is an urgent need for an alternative solution that can be completed under low-intensity exercise conditions while being safe, universal, and predictive.
[0004] In recent years, researchers and clinicians have proposed various simplified assessment protocols and prediction methods, attempting to replace traditional high-intensity exercise tests with lower intensity and easier operation. Typical methods include the 6-minute walk test (6MWT), the three-minute step test (Step Test), submaximal power heart rate analysis (such as PWC170), and multivariate scoring based on physiological parameters or artificial intelligence prediction models. These methods have lowered the assessment threshold and expanded the application scenarios to some extent, but they still have significant shortcomings. First, the assessment results of tests such as the 6MWT and Step Test are easily affected by subjective factors such as the subject's active cooperation, cadence control, and exercise habits, and the assessment error of cardiopulmonary function during high-intensity exercise is relatively large. Second, prediction models built based on statistical or machine learning methods often rely on a large number of training samples and often use CPET results as labels to build models, resulting in limited generalization ability and insufficient interpretability, and making it difficult to provide detailed cardiopulmonary metabolic parameters.
[0005] Therefore, while existing simplified methods reduce operational complexity, they still cannot replace the core functions of CPET in terms of accuracy and adaptability to individual differences. Especially in populations that cannot tolerate high-intensity exercise, there is an urgent need for a new cardiopulmonary function assessment and prediction technology that can be triggered at low intensity or in a non-exercise state, uses multi-parameter fusion modeling, and combines repeatability, predictability, and clinical interpretability, in order to achieve a shift from "high-intensity exercise dependence" to "safe, efficient, and personalized" approaches. Summary of the Invention
[0006] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, device, equipment and storage medium for predicting cardiopulmonary exercise test data, so as to solve the technical problem that the existing low-intensity cardiopulmonary exercise test data prediction methods are difficult to achieve personalized adaptive prediction to cope with individual differences and dynamic changes and high prediction accuracy.
[0007] To achieve the above objectives, the present invention employs the following technical solution: This invention provides a method for predicting cardiopulmonary exercise test data, comprising the following steps: 1) Acquiring input physiological signals x(t) and control variables c(t) The control variables include individual characteristic parameters and motion parameters; 2) The input physiological signal x(t) and control variables c(t) The data is input into the reservoir calculation model, which includes the reservoir. 3) Update the reservoir state r(t) using the reservoir state dynamic update equation, which is:
[0008] in, r（t） Let N be the current states of N neurons in the reservoir at time t. α For leakage parameters, W r For the recursive weight matrix, W in For the input weight matrix, Wc To control the mapping matrix, x(t) To input physiological signals, c(t) For control variables; 4) Generate prediction output based on the updated reservoir state.
[0009] Preferably, in step 1), the control variable c(t) The individual characteristic parameters include at least one of age, sex, height, and weight, and the exercise parameters include at least one of real-time exercise speed, exercise load, and exercise pattern. Preferably, the reservoir calculation model is trained in the following manner, wherein the training includes a data preparation step: acquiring time-series data of cardiopulmonary exercise tests from multiple subjects; The time-series data of cardiopulmonary exercise tests from multiple subjects were concatenated along the time dimension to construct a combined training dataset, with data labels set at the intersection of the time-series data of cardiopulmonary exercise tests from different subjects.
[0010] The following are the definitions of the input physiological signal x(t) and the time series data of cardiopulmonary exercise test in this invention, and the relationship between the two is explained.
[0011] I. Definition The input physiological signal x(t) is defined as: physiological measurement data obtained from a single individual subject when the reservoir computational model constructed in this application is input into the input layer of the reservoir computational model at time point t for individualized prediction.
[0012] Specifically, the input physiological signal x(t) is a time-series vector representing a set of physiological parameters measured at a specific time t. The input physiological signal x(t) includes signals such as oxygen uptake (VO2), carbon dioxide excretion (VCO2), minute ventilation (VE), heart rate (HR), and respiratory rate (RR). In the prediction process, it serves as a direct input to the reservoir state dynamic update equation.
[0013] Cardiopulmonary exercise test time series data definition: refers to a complete historical data set of cardiopulmonary exercise tests collected from multiple subjects during the model training phase.
[0014] Specifically, the cardiopulmonary exercise test time-series data is a collection of independent and complete physiological parameter records (i.e., multiple complete input physiological signal sequences) from multiple subjects. After preprocessing, the cardiopulmonary exercise test time-series data is used to construct a training dataset, particularly by concatenating multi-subject data along the time dimension using multi-subject data concatenation technology to form a combined training dataset.
[0015] II. Explanation of the relationship between the two Both are essentially the same type of data, originating from cardiopulmonary exercise tests, and representing the same physiological parameters (such as VO2, VCO2, VEz, HR, RR). Cardiopulmonary exercise test time series data are the data source and aggregate form of input physiological signals.
[0016] The application stage and purpose are different: Cardiopulmonary exercise test time-series data is used in the model training phase. As raw material, its purpose is to allow the model to learn general physiological dynamics and the influence of individual differences from a large amount of diverse historical data, thereby determining the model parameters (mainly the output weight matrix W). out Its processing targets are groups and batches of historical data.
[0017] The input physiological signal x(t) is used in the prediction (application) phase of the model. As a driving signal, its purpose is to input real-time or historical data of a specific individual after model training, driving the model's state evolution to generate predictions for that individual's future. It processes single-individual, real-time or continuous time-series data.
[0018] Logical flow relationship: Training flow: Cardiopulmonary exercise test time series data → (after preprocessing and concatenation) → combined training dataset → used for training → resulting in a reservoir computational model.
[0019] Prediction flow: For the target individual, obtain its input physiological signal x(t) and control variables → input them into the pre-trained reserve pool computational model → obtain the prediction output.
[0020] In summary, the time series data of cardiopulmonary exercise tests is the original historical data set used in the training phase, while the input physiological signal x(t) is the real-time or online data received by the model in the application (prediction) phase.
[0021] Preferably, the training further includes the following steps: Based on the combined training dataset, the output weight matrix W is calculated using the ridge regression algorithm. out The calculation formula is as follows:
[0022] in,W out To output the weight matrix, u Output the target matrix. R The reservoir state matrix, R T Reservoir state matrix R The transpose of the matrix, β For Tikhonov regularization parameters, I It is an identity matrix.
[0023] Preferably, during the training process, the loss function L used is a composite loss function, the expression of which is:
[0024] in, L For loss function, p These are the weighting coefficients. X predict For the dynamic physiological data sequence predicted by the model, X ture This is a real, dynamic physiological data sequence. n The total number of time steps. VO 2 max predict The maximum oxygen uptake predicted by the model. VO 2 max ture This represents the actual measured maximum oxygen uptake.
[0025] Preferably, during the training of the reservoir computational model, a Bayesian optimization combined with grid search method is used to automatically optimize the hyperparameters of the reservoir computational model; the hyperparameters include spectral radius and leakage parameters. α Recursive weight matrix W r Connection density, input weight matrix W in The scaling factor and at least one of the Tikhonov regularization parameter β.
[0026] Preferably, step 4) involves generating the prediction output based on the updated reservoir state, including: Initiation and warm-up phase: Real physiological data of the subject during the initial period after the start of exercise are used as input physiological signals. x(t) Combined with the corresponding control variables c(t) This drives the reservoir calculation model, synchronizing the reservoir state with the subject's physiological and dynamic state. Closed-loop prediction phase: After the warm-up phase, switch to closed-loop mode, using the prediction output of the previous time step of the reserve pool calculation model as part of the input for the current time step, and combining it with subsequent control variables.c(t) This enables the reserve pool calculation model to autonomously generate prediction data for subsequent time points.
[0027] The present invention also provides a cardiopulmonary exercise test data prediction device applied to the above-mentioned cardiopulmonary exercise test data prediction method, comprising: The data acquisition module is used to acquire input physiological signals. x(t) and control variables c(t) The control variables include individual characteristic parameters and motion parameters; The reservoir state update module is used to update the input physiological signal. x(t) and control variables c(t) The reservoir is input into the reservoir computational model, and the reservoir state r(t) is updated according to the following equation:
[0028] in, r（t） Let N be the current states of N neurons in the reservoir at time t. α For leakage parameters, W r For the recursive weight matrix, W in For the input weight matrix, Wc To control the mapping matrix, x(t) To input physiological signals, c(t) For control variables; The output prediction module is used to generate prediction outputs based on the updated reservoir state.
[0029] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to, when executing the computer program, implement the method for predicting cardiopulmonary exercise test data as described above.
[0030] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for predicting cardiopulmonary exercise test data as described above.
[0031] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for predicting cardiopulmonary exercise test (CPE) data. Its core lies in constructing a prediction framework based on a reserve pool calculation. By inputting a weight matrix, low-dimensional physiological signals are mapped to a high-dimensional reserve pool space, optimizing the problem that traditional statistical models struggle to handle complex nonlinear correlations. When preparing training data, a strategic concatenation of CPE time-series data from a large number of different subjects along the time axis overcomes the problem of scarce clinical individual data leading to difficulty in model training, resulting in deep neural networks failing to converge sufficiently or easily overfitting. By introducing a VO2max prediction error term into the loss function and setting a dedicated proportional weight coefficient ρ, the problem is solved where standard machine learning loss functions typically only focus on the fitting error of the overall dynamic trend, neglecting the crucial diagnostic significance of the prediction accuracy of physiological limit indicators such as VO2max in clinical applications. In principle, the reserve pool itself is a dynamic system with short-term memory characteristics, and its high-dimensional state space can effectively capture the complex spatiotemporal dynamic characteristics of physiological signals. The control variable c(t) (individual characteristics and exercise parameters) is weighted by a dedicated weight matrix W. c By directly embedding it into the reservoir state update equation, the model's intrinsic dynamics can be modulated and guided in real time by these external conditions, thus realizing the deep coupling between the prediction model and the individual state and motion scenario in principle.
[0032] Furthermore, the specific composition of the control variables was clarified. In principle, individual characteristics such as age, gender, height, and weight are important physiological constraints determining basal metabolic rate and baseline cardiopulmonary function; while exercise parameters such as real-time exercise speed, load, and mode are direct external stimuli driving the physiological system to respond. Using these key variables as explicit inputs to the model allows it to differentiate between the physiological characteristics of different individuals in principle and respond to dynamic changes in exercise intensity, which is a necessary condition for achieving accurate and personalized modeling.
[0033] Furthermore, a method for constructing training data by temporally concatenating multi-subject data is proposed. In principle, cardiopulmonary exercise test data typically exhibits strong individuality and limited data volume per sample. This method, through concatenation, splices together longer training sequences containing multiple sample patterns over time, effectively expanding the scale and diversity of the training data. This helps the reserve pool learn more general and robust physiological dynamic patterns, improving the model's generalization ability. Setting data labels helps the model identify the boundaries of different data segments in principle, preventing dynamic confusion between data from different individuals.
[0034] Furthermore, it specifically specifies the use of the Ridge Regression algorithm to train the output layer. In principle, one of the core advantages of pooled computation is that it only requires training the output weights W. outRidge regression, by introducing the Tikhonov regularization parameter β, constrains the norm of the solution vector when solving for weights. This effectively prevents overfitting when training data is limited or noisy, improves the stability of the model and the robustness of predictions on new data, and is a key technical link to ensure high prediction accuracy.
[0035] Furthermore, a composite loss function was introduced. Its innovation lies in expanding the optimization objective from a single time series fitting to simultaneously optimizing both dynamic sequence prediction error and physiological limit indices (such as VO2max) prediction error. This forces the model, during training, to not only learn the matching of local dynamics but also the global correlation between its internal state and overall physiological limits. This multi-objective optimization, in principle, guides the model to generate predictions that conform to both short-term dynamics and long-term physiological constraints, significantly improving the physiological rationality and overall accuracy of the predictions.
[0036] Furthermore, by employing an automated optimization strategy combining Bayesian optimization and grid search, the hyperparameter space can be intelligently explored. Bayesian optimization uses previous evaluation results to build a surrogate model to guide subsequent searches, focusing on promising regions, thus finding better parameter combinations with fewer evaluations; combined with grid search, it ensures basic coverage within a certain range. This hybrid strategy achieves an excellent balance between efficiency and effectiveness, automatically finding the hyperparameter configuration that enables the model to reach its optimal predictive state.
[0037] Furthermore, the pre-prediction and closed-loop prediction phases for individual prediction are specified in detail. From a kinetic perspective, the pre-prediction phase uses the subject's initial real data to synchronize the reservoir's internal state to a trajectory that matches the individual's current physiological dynamics, completing personalized state initialization. Subsequently, the system switches to closed-loop prediction, where the model autonomously evolves using the synchronized state and known movement plan, achieving a smooth transition from data-driven to model-driven approaches. This process, in principle, simulates the clinical logic of calibration-prediction, ensuring the stability and individual relevance of long-term predictions.
[0038] This invention also modularizes the function of the method and defines the corresponding apparatus. From the perspective of system implementation principles, the modular design clarifies the data flow and processing responsibilities, enabling the above method to be implemented in a clear and efficient software or hardware form, thereby improving the feasibility and interpretability of the technology.
[0039] This invention also extends the method to electronic devices and storage media. In principle, it completes the leap from "algorithm" to product, clarifies the specific application form and carrier of the invention's technical solution, and demonstrates its industrial practicality and scalability. Attached Figure Description
[0040] Figure 1 This is a schematic diagram of the cardiopulmonary exercise test data prediction method of the present invention; Figure 2 This is a schematic diagram of the reservoir computing network (RC network) structure of the present invention. Detailed Implementation
[0041] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0042] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0043] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 The core of this invention lies in using an improved reservoir calculation model to predict dynamic physiological data (such as VO2, VCO2, etc.) of an individual during a cardiopulmonary exercise test (CPET) through a controlled mechanism.
[0044] 1. Overall Method Flow This invention constructs a reservoir computing network architecture based on a controlled mechanism for processing high-dimensional, nonlinear physiological time series data. The specific implementation process of this method is as follows: Figure 1 As shown, it mainly includes three stages: data preprocessing, reservoir computing network (RC network) model construction and training, and real-time prediction (personalized prediction).
[0045] 2. Data Preparation First, the acquired raw CPET time series data is preprocessed. This raw CPET time series data includes any one or more signals from VO2, VCO2, VE, HR, and RR, as well as individual characteristic parameters (such as age and sex) and motion parameters (such as motion speed scheme) as control variables c(t). The preprocessing steps include: Outlier filtering: Remove test samples that deviate from the standard test plan or have more than 2% missing key data.
[0046] Missing value handling: Linear interpolation is used to fill in data points with minor missing values.
[0047] Standardization: Divide the raw CPET time series data by the subject's weight to convert it into energy expenditure per unit body weight (ml / kg / min).
[0048] Isochronous resampling: Since the sampling points of the original CPET time series data are not equidistant, linear interpolation is used to perform uniform resampling at a fixed interval of 2 seconds to form a regular time series.
[0049] To overcome the problem of insufficient individual data, this invention proposes a multi-subject data concatenation technique. Specifically, based on the assumption that the basic dynamics of gas exchange during CPET in healthy individuals are consistent, the preprocessed time-series data of M subjects (e.g., 581 subjects) are concatenated end-to-end along the time dimension to construct a large combined training dataset with a total time step length of N (e.g., approximately 316,726 steps). To prevent the model from learning irrelevant inter-individual transitional features, markers are set at the intersections of the time-series data of each subject, and the predicted targets for several time steps (e.g., 10 steps) after the marked points are removed during training.
[0050] 3. Model Architecture and State Update The storage pool calculation model architecture of the present invention is as follows: Figure 2 As shown, it mainly includes an input layer, a reservoir, and an output layer.
[0051] Input layer (left side, indicated by red square): Responsible for receiving the input physiological signal x(t) (such as VO2, VCO2) and control variable c(t) at time t. x(t) is input through the weight matrix W. in Mapped to the reservoir, c(t) is mapped through a dedicated control mapping matrix W. c Mapped to the reservoir. The input physiological signal x(t) includes one or more of oxygen uptake (VO2), carbon dioxide excretion (VCO2), minute ventilation (VE), heart rate (HR), and respiratory rate (RR).
[0052] The reservoir (in the middle, the green node network surrounded by dashed circles): consists of a large number (e.g., N) of randomly sparsely connected neurons, whose connections are determined by a fixed recursive weight matrix W. r Definition. Reservoirs possess short-time memory characteristics, enabling them to nonlinearly map low-dimensional inputs to a high-dimensional state space. The dynamic update equation for the reservoir state r(t) is the core of this scheme:
[0053] Where r(t) represents the current state of N neurons in the reservoir at time t, parameter α represents the leakage parameter (used to adjust the influence of the previous state on the current state), used to adjust the degree of influence of historical states, x(t) is the input physiological signal, and c(t) is the control variable (velocity, age, gender). W in W is the weight matrix that maps the input to the neuron. c It is the control mapping matrix, W r It is a recursive weight matrix.
[0054] The equation explicitly uses matrix W to represent the control variable c(t). c Integrating it into state evolution enables the model to dynamically respond to individual differences and changes in motion patterns.
[0055] Output layer (right side, indicated by blue squares): Employs a trainable output weight matrix W out The high-dimensional reservoir state r(t) is mapped to the target predicted value y(t) (such as VO2 at the next time step). In the closed-loop prediction mode, the prediction result of the output layer is fed back to the input layer as the input at the next time step.
[0056] 4. Model Training The model training only requires optimizing the output weight matrix W. out Reservoir parameters (W) r, W in W c The structure is fixed. The training process is as follows: Using the combined training dataset, the output weight matrix W is calculated using the Ridge Regression algorithm. out The calculation formula is:
[0057] in, W out Let R be the output weight matrix, and R be the reservoir state matrix, which is the reservoir state matrix obtained from all time steps in the training set. R T Reservoir state matrix RThe transpose of the matrix is given, r(t) is the stacked reservoir state matrix, u is the target output matrix, β is the Tikhonov regularization parameter, and I is the identity matrix. This method effectively prevents overfitting. Matrices R and u are constructed by stacking the reservoir states r(t) at all time steps t in the training dataset and the input data x(t).
[0058] The loss function L used for training is a customized composite loss function, designed to simultaneously optimize the accuracy of dynamic trajectory fitting and the accuracy of key clinical indicator prediction.
[0059] in, L The loss function is ρ, which is the weighting coefficient. This function not only constrains the accuracy of the instantaneous dynamic value, but also performs weighted optimization specifically for the physiological limit index of maximum oxygen uptake (VO2max). X predict For the dynamic physiological data sequence predicted by the model, X ture This represents a real, dynamic physiological data sequence, where n is the total number of time steps. VO 2 max predict The maximum oxygen uptake predicted by the model. VO 2 max ture This is the actual measured maximum oxygen uptake. The loss function explicitly includes... VO 2 max The prediction error of this core physiological limit indicator is one of the optimization objectives.
[0060] The model hyperparameters (such as spectral radius, leakage rate α, regularization coefficient β, loss function weight ρ, etc.) are automatically optimized using Bayesian optimization combined with grid search.
[0061] 5. Personalized forecasting process For a new subject, the individualized prediction process consists of two stages: Initiation and warm-up phase: The real physiological signal x(t) and its corresponding control variable c(t) of the subject during an initial period (e.g., the first 3 minutes) after the start of exercise are input into the trained model. During this phase, the model is in open-loop mode and driven by real data, with the aim of rapidly synchronizing the reservoir state r(t) with the initial physiological and dynamic state of the specific individual.
[0062] Closed-loop prediction phase: After the warm-up phase, the system switches to closed-loop mode. In this mode, the actual physiological signal x(t) is no longer input; instead, the predicted output y(t) of the model from the previous time step is fed back to the input layer as part of the input for the current time step. Simultaneously, subsequent preset control variables c(t) (such as increasing movement speed) are continuously input. Driven by this mechanism, the model can autonomously evolve to generate prediction curves for the subject's entire subsequent movement and recovery phases, thereby achieving predictions of key indicators such as VO2max.
[0063] Example 2 The present invention also provides a cardiopulmonary exercise test data prediction device, which can be implemented by software, hardware or a combination of software and hardware, and includes the following modules: Data acquisition module: used to acquire input physiological signal x(t) and control variable c(t).
[0064] Reservoir state update module: It has a built-in reservoir of the reservoir calculation model and calculates and updates the reservoir state according to the following equation. r（t） The equation is:
[0065] in, r（t） Let N be the current states of N neurons in the reservoir at time t. α For leakage parameters, W r For the recursive weight matrix, W in For the input weight matrix, Wc To control the mapping matrix, x(t) To input physiological signals, c(t) For control variables; Output prediction module: used to predict the reservoir state r(t) and output weight matrix W. out Generate the prediction result y(t).
[0066] The working process of this device is consistent with the aforementioned method embodiments, with each module working together to complete the function from data input to prediction output.
[0067] Example 3 The method of the present invention can be deployed in an electronic device. The electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the cardiopulmonary exercise test data prediction method as described above.
[0068] The computer program implementing the above method can be stored on a computer-readable storage medium, such as a USB flash drive, hard disk, optical disk, server, etc. When the program in the storage medium is executed by a processor, the method described in this invention can be implemented.
[0069] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, devices, and storage media for predicting cardiopulmonary exercise test data. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented 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.
[0070] This invention is described with reference to flowchart illustrations and / or block diagrams of a method, apparatus, device, and storage medium for predicting cardiopulmonary exercise testing data 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 device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0071] 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.
[0072] 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.
[0073] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for predicting cardiopulmonary exercise test data, characterized in that, Includes the following steps: 1) Acquiring input physiological signals x(t) and control variables c(t) The control variables include individual characteristic parameters and motion parameters; 2) The input physiological signal x(t) and control variables c(t) The data is input into the reservoir calculation model, which includes the reservoir. 3) Update the reservoir state through the dynamic update equation of the reservoir state. r（t） The reservoir state dynamic update equation is: in, r（t） Let N be the current states of N neurons in the reservoir at time t. α For leakage parameters, W r For the recursive weight matrix, W in For the input weight matrix, W c To control the mapping matrix, x(t) To input physiological signals, c(t) For control variables; 4) Generate prediction output based on the updated reservoir state.
2. The method for predicting cardiopulmonary exercise test data according to claim 1, characterized in that, In step 1), the individual characteristic parameters include at least one of age, gender, height and weight, and the exercise parameters include at least one of real-time exercise speed, exercise load and exercise mode.
3. The method for predicting cardiopulmonary exercise test data according to claim 1, characterized in that, The reservoir calculation model is trained in the following way, including a data preparation step before training: Obtain time-series data of cardiopulmonary exercise tests from multiple subjects; The time-series data of cardiopulmonary exercise tests from multiple subjects were concatenated along the time dimension to construct a combined training dataset, with data labels set at the intersection of the time-series data of cardiopulmonary exercise tests from different subjects.
4. The method for predicting cardiopulmonary exercise test data according to claim 3, characterized in that, The training also includes the following steps: Based on the aforementioned combined training dataset, the output weight matrix W is calculated using the ridge regression algorithm. out The calculation formula is: in, W out To output the weight matrix, u Output the target matrix. R The reservoir state matrix, R T Reservoir state matrix R The transpose of the matrix, β For Tikhonov regularization parameters, I It is an identity matrix.
5. The method for predicting cardiopulmonary exercise test data according to claim 4, characterized in that, During the training process, the loss function L used is a composite loss function, and its expression is: in, L For loss function, ρ These are the weighting coefficients. X predict For the dynamic physiological data sequence predicted by the model, X ture This is a real, dynamic physiological data sequence. n The total number of time steps. VO 2 max predict The maximum oxygen uptake predicted by the model. VO 2 max ture This represents the actual measured maximum oxygen uptake.
6. A method for predicting cardiopulmonary exercise test data according to claim 4 or 5, characterized in that, During the training of the reservoir computational model, a method combining Bayesian optimization and grid search is used to automatically optimize the hyperparameters of the reservoir computational model; the hyperparameters include spectral radius and leakage parameters. α Recursive weight matrix W r Connection density, input weight matrix W in The scaling factor and at least one of the Tikhonov regularization parameter β.
7. The method for predicting cardiopulmonary exercise test data according to claim 1, characterized in that, Step 4) generates the prediction output based on the updated reservoir state, including: Initiation and warm-up phase: Real physiological data of the subject during the initial period after the start of exercise are used as input physiological signals. x(t) Combined with the corresponding control variables c(t) This drives the reservoir calculation model, synchronizing the reservoir state with the subject's physiological and dynamic state. Closed-loop prediction phase: After the warm-up phase, switch to closed-loop mode, using the prediction output of the previous time step of the reserve pool calculation model as part of the input for the current time step, and combining it with subsequent control variables. c(t) This enables the reserve pool calculation model to autonomously generate prediction data for subsequent time points.
8. A cardiopulmonary exercise test data prediction device applied to the cardiopulmonary exercise test data prediction method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire input physiological signals. x(t) and control variables c(t) The control variables include individual characteristic parameters and motion parameters; The reservoir state update module is used to update the input physiological signal. x(t) and control variables c(t) The reservoir is input into the reservoir computational model, and the reservoir state is updated according to the following equation. r（t） : in, r（t） Let N be the current states of N neurons in the reservoir at time t. α For leakage parameters, W r For the recursive weight matrix, W in For the input weight matrix, Wc To control the mapping matrix, x(t) To input physiological signals, c(t) For control variables; The output prediction module is used to generate prediction outputs based on the updated reservoir state.
9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the computer program to implement the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-7.