A method and system for predicting maximum oxygen uptake based on multi-source resting physiological parameters

Abstract

The application discloses a maximum oxygen uptake prediction method and system based on multi-source resting physiological parameters, and the method comprises the following steps: obtaining multi-source resting data of a target object and performing quality control and feature preprocessing to obtain multi-source preprocessed feature data; performing representation and fusion according to the multi-source preprocessed feature data to obtain multi-source resting feature data, and inputting the multi-source resting feature data into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object; and the structure of the deep learning prediction model comprises a first nonlinear mapping layer, a conditional coding layer, a modulation parameter generation layer, a conditional modulation layer, a second nonlinear mapping layer and an output layer. The application can realize fast, objective and low-burden estimation of the maximum oxygen uptake in a non-movement state, reduce the dependence of traditional cardiopulmonary exercise testing on movement ability, special equipment and test scenes, and improve the accessibility of cardiopulmonary function evaluation for people who cannot complete movement testing or are not suitable for movement testing.

Description

Technical Field

[0001] This application relates to the fields of medical artificial intelligence and cardiopulmonary function assessment technology, and in particular to a method and system for predicting maximum oxygen uptake based on multi-source resting physiological parameters. Background Technology

[0002] Maximum oxygen uptake (VO2 max) is an important physiological indicator reflecting a person's cardiopulmonary endurance, aerobic metabolic capacity, and overall exercise function. It has significant reference value in scenarios such as cardiovascular disease assessment, perioperative risk stratification, rehabilitation management, chronic disease follow-up, and health management. In current clinical or research settings, VO2 max is usually obtained through cardiopulmonary exercise testing. This method requires the subject to perform progressively increasing exercise loads on a stationary bike or treadmill, while simultaneously collecting respiratory gas exchange parameters, heart rate, ventilation, and other indicators to assess the level of cardiopulmonary function under exercise load. However, traditional cardiopulmonary exercise testing has high requirements for testing equipment, testing site, professional operators, and the exercise ability of the target subjects. For elderly patients, patients with heart failure, patients with poor preoperative physical fitness, patients with lower limb movement disorders, or patients who cannot tolerate exercise load, standard cardiopulmonary exercise testing may be difficult to complete, and may even induce discomfort or safety risks. Therefore, in actual clinical application, it is often limited by equipment accessibility, testing cost, patient compliance, and safety, making it difficult to use as a low-burden, high-frequency, and wide-coverage assessment method for routine screening, continuous follow-up, or rapid preoperative risk assessment.

[0003] To reduce reliance on the target individual's athletic ability and professional testing scenarios, some existing technologies attempt to estimate VO2 max using non-exercise information or simplified exercise test information. For example, some solutions use information such as age, gender, height, weight, body mass index, resting heart rate, and physical activity questionnaires, estimating VO2 max through empirical formulas, linear regression models, or traditional machine learning models; some solutions combine user attribute information, exercise data, and historical prediction results for multi-stage correction; some solutions still require the use of cardiopulmonary function instruments, masks, heart rate monitoring devices, or gas sampling devices to collect respiratory gas metabolism parameters during exercise, and establish predictive relationships through parameter screening, correlation analysis, or regression models; some solutions use simplified exercise test data such as the 20-meter shuttle run, combined with body mass index stratification or multilayer perceptron models to output VO2 max prediction results; in addition, in the fields of wearable devices and digital health management, VO2 max prediction results are also used for generating exercise training plans or recommending exercise prescriptions.

[0004] While the aforementioned methods can reduce the frequency of standard cardiopulmonary exercise testing to some extent, or achieve VO2 max estimation in scenarios such as fitness management, they still have significant shortcomings in rapid clinical assessment and non-exercise status assessment scenarios. Some methods still rely on exercise data, shuttle run test data, respiratory gas metabolism parameters during exercise, or historical exercise sequences, making them unsuitable for elderly individuals, those with poor preoperative physical fitness, those with limited cardiopulmonary function, those with exercise disorders, or those unsuitable for exercise load testing. Some non-exercise estimation methods require additional information such as echocardiography, imaging examinations, laboratory tests, exercise history questionnaires, or invasive hemodynamic parameters. Such data is not readily available in all clinical scenarios, increasing data collection costs and application barriers, and limiting their widespread application in routine outpatient screening, preoperative initial screening, and continuous follow-up.

[0005] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0006] The technical problem to be solved by this application is to provide a method and system for predicting maximum oxygen uptake based on multi-source resting physiological parameters, in order to address the above-mentioned deficiencies of the prior art. This aims to solve the technical problem that existing maximum oxygen uptake estimation methods are difficult to apply to elderly people, people with limited cardiopulmonary function, and other unsuitable exercise targets and clinical rapid assessment scenarios because they rely on exercise load test data or require additional examination information such as echocardiography and imaging examinations that are not easy to obtain in real time.

[0007] The technical solution adopted in this application to solve the problem is as follows: In a first aspect, embodiments of this application provide a method for predicting maximum oxygen uptake based on multi-source resting physiological parameters, the method comprising: Acquire multi-source static data of the target object; Quality control and feature preprocessing are performed on the multi-source resting data to obtain multi-source preprocessed feature data; The multi-source preprocessed feature data is characterized and fused to obtain multi-source resting feature data. The multi-source resting feature data is input into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object; The structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

[0008] In one implementation, the multi-source resting data includes demographic information, basic physical parameters, and physiological parameters at rest; the demographic information includes age and / or sex; the basic physical parameters include at least one of height, weight, and body mass index; and the physiological parameters at rest include at least one of resting heart rate, blood pressure, respiratory rate, ventilation, oxygen uptake, carbon dioxide excretion, respiratory exchange rate, and metabolic equivalent.

[0009] In one implementation, the quality control steps include: The multi-source static data is processed using at least one of the following methods: field integrity verification, data type verification, unit unification, outlier identification, duplicate record processing, non-final record exclusion, and abnormal data removal or marking.

[0010] In one implementation, the feature preprocessing steps include: For numerical features in multi-source resting data after quality control, at least one of the following imputation methods is used to handle missing values: median imputation, mean imputation, model imputation, and imputation based on clinical rules. Then, the data is standardized or normalized. For categorical features in multi-source static data after quality control, at least one of the following processing methods is used for feature processing: missing category imputation, one-hot encoding, and embedding encoding.

[0011] In one implementation, the step of characterizing and fusing the multi-source preprocessed feature data to obtain multi-source resting feature data includes: Numerical resting physiological parameters are input into the numerical feature representation branch to form a numerical feature vector, and categorical features are input into the categorical feature encoding branch to form a categorical feature vector. Then, a unified multi-source resting feature vector is formed by splicing, weighted fusion, or gated fusion.

[0012] In one implementation, the step of performing dimension-wise adaptive modulation of the latent feature representation based on the modulation parameters includes: The different dimensions in the latent feature representation are amplified, weakened, or shifted according to the scaling and translation parameters in the modulation parameters to achieve dimension-wise adaptive modulation; wherein, the scaling parameter is used to adjust the intensity of the latent feature, and the translation parameter is used to adjust the degree of latent feature shift.

[0013] In one implementation, the method further includes: Downstream auxiliary applications are based on the predicted maximum oxygen uptake; the downstream auxiliary applications include at least one of the following: high-risk identification, preoperative risk assessment, rehabilitation management, follow-up trend monitoring, and clinical stratification management.

[0014] Secondly, embodiments of this application also provide a maximum oxygen uptake prediction system based on multi-source resting physiological parameters, the system comprising: The data acquisition module is used to acquire multi-source resting data of the target object; The data preprocessing module is used to perform quality control and feature preprocessing based on the multi-source resting data to obtain multi-source preprocessed feature data; The feature fusion module is used to characterize and fuse the multi-source preprocessed feature data to obtain multi-source resting feature data. The maximum oxygen uptake prediction module is used to input the multi-source resting feature data into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object. The structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

[0015] Thirdly, embodiments of this application also provide a terminal, the terminal including a memory and one or more processors; the memory stores one or more programs; the programs include instructions for executing the maximum oxygen uptake prediction method based on multi-source resting physiological parameters as described above; the processor is used to execute the programs.

[0016] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a plurality of instructions adapted to be loaded and executed by a processor to implement the steps of the maximum oxygen uptake prediction method based on multi-source resting physiological parameters as described above.

[0017] The beneficial effects of this application are as follows: This application obtains multi-source resting data of the target object and performs quality control and feature preprocessing to obtain multi-source preprocessed feature data; based on the multi-source preprocessed feature data, it performs characterization and fusion to obtain multi-source resting feature data, which is then input into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object; the structure of the deep learning prediction model includes: a first nonlinear mapping layer, a conditional coding layer, a modulation parameter generation layer, a conditional modulation layer, a second nonlinear mapping layer, and an output layer. This application can achieve rapid, objective, and low-burden estimation of maximum oxygen uptake in a non-exercise state, reducing the dependence of traditional cardiopulmonary exercise testing on exercise capacity, specialized equipment, and testing scenarios, and improving the accessibility of cardiopulmonary function assessment for people who cannot complete exercise testing or are unsuitable for exercise testing. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the maximum oxygen uptake prediction method based on multi-source resting physiological parameters provided in the embodiments of this application.

[0020] Figure 2 This is a detailed schematic diagram illustrating the steps of the maximum oxygen uptake prediction method based on multi-source resting physiological parameters provided in the embodiments of this application.

[0021] Figure 3 This is a schematic diagram of the model structure provided in the embodiments of this application.

[0022] Figure 4 This is a schematic diagram of downstream auxiliary applications provided in the embodiments of this application.

[0023] Figure 5 This is a schematic diagram of the functional modules of the maximum oxygen uptake prediction system based on multi-source resting physiological parameters provided in the embodiments of this application.

[0024] Figure 6 This is a schematic block diagram of the terminal provided in the embodiments of this application. Detailed Implementation

[0025] This application discloses a method and system for predicting maximum oxygen uptake based on multi-source resting physiological parameters. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0026] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0027] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0028] To address the aforementioned deficiencies in existing technologies, this application provides a method for predicting maximum oxygen uptake (VO2 max) based on multi-source resting physiological parameters. The method includes: acquiring multi-source resting data of a target object; performing quality control and feature preprocessing on the multi-source resting data to obtain multi-source preprocessed feature data; characterizing and fusing the multi-source preprocessed feature data to obtain multi-source resting feature data; and inputting the multi-source resting feature data into a deep learning prediction model to obtain the predicted VO2 max of the target object. The deep learning prediction model comprises: a first nonlinear mapping layer for extracting latent feature representations from the multi-source resting feature data; a conditional encoding layer for generating conditional codes from the multi-source resting feature data; a modulation parameter generation layer for generating modulation parameters from the latent feature representations and the conditional codes; a conditional modulation layer for performing dimension-wise adaptive modulation of the latent feature representations based on the modulation parameters to obtain modulated latent feature representations; a second nonlinear mapping layer for generating high-level feature representations from the modulated latent feature representations; and an output layer for generating the predicted VO2 max of the target object from the high-level feature representations.

[0029] This application addresses the technical problems of existing methods for obtaining VO2 max, which mainly rely on cardiopulmonary exercise testing, have high requirements for the exercise capacity of the target subjects and equipment conditions, are incapable of some patients from completing standard exercise load tests, and have limited application in rapid clinical screening and risk assessment. It proposes a method for predicting VO2 max without requiring exercise load testing. This method collects multi-source resting data from the target subjects, preprocesses and fuses the multi-source resting physiological parameters, and inputs the obtained multi-source resting feature data into a specially designed deep learning prediction model. This model learns the nonlinear mapping relationship between resting parameters and VO2 max, and can accurately output the predicted VO2 max. This application is characterized by multi-source resting physiological parameter input, conditional modulation deep learning structure, and clinical auxiliary output. It can achieve rapid, objective, and low-burden estimation of VO2 max in a non-exercise state, reducing the dependence of traditional cardiopulmonary exercise testing on exercise capacity, specialized equipment, and testing scenarios, and improving the accessibility of cardiopulmonary function assessment for people who cannot complete exercise testing or are unsuitable for exercise testing. At the same time, by integrating multi-source resting physiological parameters and utilizing the individual conditional modulation mechanism within the model, it helps to improve the accuracy of VO2 max estimation and adaptability to different target groups, providing auxiliary decision-making basis for clinical applications such as high-risk identification, preoperative risk assessment, rehabilitation management, and long-term follow-up.

[0030] like Figure 1 As shown, the method specifically includes the following steps: Step S100: Obtain multi-source static data of the target object.

[0031] Furthermore, such as Figure 2As shown, the multi-source resting data includes demographic information, basic physical parameters, and physiological parameters at rest; the demographic information includes age and / or sex; the basic physical parameters include at least one of height, weight, and body mass index; the physiological parameters at rest include at least one of resting heart rate, blood pressure, respiratory rate, ventilation, oxygen uptake, carbon dioxide excretion, respiratory exchange rate, and metabolic equivalent.

[0032] Specifically, firstly, with the target subject in a resting, non-exercising state, a set of resting characteristic parameters covering multiple dimensions is obtained through at least one data acquisition method to construct the input data foundation for predicting maximum oxygen uptake (VO2max). Multi-source resting data includes three levels: demographic information, basic physical parameters, and physiological parameters at rest. As shown in Table 1, demographic information includes at least age and / or sex, used to characterize the basic demographic features of the target subject; basic physical parameters include at least one of height, weight, and body mass index (BMI), used to reflect the target subject's body structure and nutritional status; physiological parameters at rest include at least one of resting heart rate, blood pressure, respiratory rate, ventilation, oxygen uptake, carbon dioxide excretion, respiratory exchange rate, and metabolic equivalent, and may further include other resting physiological indicators obtainable by cardiopulmonary function testing equipment.

[0033] Regarding data sources, the multi-source resting data in this embodiment can come from various data acquisition sources, including but not limited to: cardiopulmonary function testing equipment, such as resting pulmonary function instruments or metabolic carts; electronic medical record systems (EMR); bedside monitoring equipment; outpatient vital sign collection equipment; wearable devices, such as smart bracelets or heart rate monitoring patches; and manual input terminals, such as the manual input interface of medical workstations, thereby ensuring the flexibility and clinical adaptability of data acquisition.

[0034] At the input feature construction level, this embodiment is not limited to using all resting parameters, but allows for the selection of different combinations of resting parameters based on actual clinical scenarios, such as pre-hospital screening, bedside rapid assessment, or home health management. This enables customization of the feature input for the VO2 max prediction model, balancing prediction accuracy with the feasibility of data collection. Through the integration and feature processing of the above-mentioned multi-source resting data, reliable data support can be provided for the cardiopulmonary endurance assessment of the target subjects without relying on exercise stress testing.

[0035] Table 1. Variables Required for Prediction

[0036] Step S200: Perform quality control and feature preprocessing based on the multi-source resting data to obtain multi-source preprocessed feature data.

[0037] Furthermore, the quality control steps include: The multi-source static data is processed using at least one of the following methods: field integrity verification, data type verification, unit unification, outlier identification, duplicate record processing, non-final record exclusion, and abnormal data removal or marking.

[0038] Furthermore, the feature preprocessing steps include: For numerical features in multi-source resting data after quality control, at least one of the following imputation methods is used to handle missing values: median imputation, mean imputation, model imputation, and imputation based on clinical rules. Then, the data is standardized or normalized. For categorical features in multi-source static data after quality control, at least one of the following processing methods is used for feature processing: missing category imputation, one-hot encoding, and embedding encoding.

[0039] Specifically, after acquiring multi-source static data of the target object, the multi-source static data is first subjected to systematic quality screening and cleaning. Then, feature scaling and encoding transformation are performed on the cleaned data to finally generate a set of structured and standardized feature vectors that can be directly input into the deep learning model.

[0040] The quality control steps include at least one of the following processing methods for multi-source resting data: field integrity verification, data type verification, unit standardization, outlier identification, duplicate record handling, non-final record exclusion, and removal or marking of data that clearly does not conform to the physiological range. Through this quality control process, the adverse effects of common problems in raw clinical data, such as missing fields, incorrect data formats, inconsistent units, abnormal physiological data entry, duplicate records, or outdated records, on the stability of model prediction can be effectively reduced, ensuring that the data entering the subsequent feature extraction stage has basic consistency and reliability.

[0041] After quality control is completed, feature preprocessing is further performed. For numerical features in the multi-source resting data after quality control, such as resting heart rate, blood pressure, and ventilation, at least one of the following methods is used to handle missing values: median imputation, mean imputation, model imputation, or imputation based on clinical rules. The imputed data is then standardized or normalized to eliminate interference from different units and numerical ranges on model training. For categorical features, such as gender and medical history classification, at least one of the following methods is used for feature transformation: missing class imputation, one-hot encoding, embedding encoding, or other categorical variable encoding methods. In addition, variables derived from basic parameters, such as body mass index, can be calculated from height and weight and included in the feature set during the preprocessing stage.

[0042] In actual clinical data acquisition, some resting parameters may be missing. To address this, the system can employ one or more combinations of missing value imputation, missing value labeling, feature dimensionality reduction, or model fault tolerance mechanisms to ensure the model's effective output even with incomplete data input. After the aforementioned quality control and feature preprocessing, clinical data from different sources, with different dimensions, and of different types are converted into standardized input suitable for deep learning model processing—i.e., multi-source preprocessed feature data.

[0043] It should be noted that this embodiment does not rely on the target subject completing an exercise load test, nor does it rely on shuttle runs, treadmills, incremental cycling, gas metabolism data during exercise, or historical exercise sequences. Instead, it only collects demographic information, basic physical parameters, and multi-source resting physiological parameters obtained from the target subject in a resting state. After data quality control, missing value processing, standardization, categorical variable encoding, and feature fusion, a multi-source resting feature vector is formed for model input. Therefore, it can significantly reduce the dependence of VO2 max assessment on exercise capacity, exercise scenario, specialized testing equipment, and professional testing personnel, and is particularly suitable for clinical assessment scenarios involving the elderly, those with poor preoperative physical fitness, those with limited cardiopulmonary function, or those who cannot tolerate exercise testing.

[0044] Step S300: Characterize and fuse the multi-source preprocessed feature data to obtain multi-source resting feature data.

[0045] Further, the steps of characterizing and fusing the multi-source preprocessed feature data to obtain multi-source resting feature data include: Numerical resting physiological parameters are input into the numerical feature representation branch to form a numerical feature vector, and categorical features are input into the categorical feature encoding branch to form a categorical feature vector. Then, a unified multi-source resting feature vector is formed by splicing, weighted fusion, or gated fusion.

[0046] Specifically, after completing quality control and feature preprocessing, the cleaned and encoded features of different data types are further sent to their respective independent feature extraction branches to fully explore the inherent representation of various types of data. Through a cross-branch fusion strategy, the heterogeneous features are integrated into a unified feature vector with comprehensive expressive power, which serves as the direct input to the subsequent maximum oxygen uptake prediction model.

[0047] In practical applications, the steps for representing and fusing preprocessed features specifically include: inputting numerical resting physiological parameters, such as resting heart rate, blood pressure, respiratory rate, ventilation, oxygen uptake, carbon dioxide excretion, respiratory exchange rate, metabolic equivalent, and body mass index, into the numerical feature representation branch to form a numerical feature vector; simultaneously, inputting categorical features, such as gender, disease classification, and missing data markers, into the categorical feature encoding branch to form a categorical feature vector. Based on this, the numerical feature vector and the categorical feature vector are then combined into a unified multi-source resting feature vector through concatenation, weighted fusion, gating fusion, or other learnable feature fusion methods.

[0048] This feature representation and fusion approach enables the model to simultaneously utilize demographic information, baseline physical parameters, and multi-source physiological parameters at rest, comprehensively analyzing the cardiopulmonary reserve function of the target subject from multiple dimensions. This avoids the technical limitations of relying on a single parameter or a single empirical formula to fully represent individual differences in cardiopulmonary function. Through this feature representation and fusion approach, resting data from different sources, with different dimensions, and of different types can complement and enhance each other within a unified feature space, providing subsequent predictive models with more complete and powerful input feature representations.

[0049] Step S400: Input the multi-source resting feature data into the deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object.

[0050] Furthermore, the structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

[0051] Specifically, the multi-source resting feature data, after quality control, feature preprocessing, feature representation, and fusion, is input into a pre-built and trained deep learning prediction model. This model then uses multi-layer nonlinear transformation and conditional adaptive modulation mechanisms to ultimately output the predicted maximum oxygen uptake of the target object. This prediction process does not require the target object to undergo any exercise load testing; it can achieve quantitative assessment of cardiorespiratory endurance indicators solely based on multi-source data collected in a resting state.

[0052] like Figure 3 As shown, the deep learning prediction model structure includes an input layer, a first nonlinear mapping layer (also known as a resting-state feature representation layer), a conditional coding layer, a modulation parameter generation layer, a conditional modulation layer, a second nonlinear mapping layer (also known as a high-level feature mapping layer), and an output layer. The fused multi-source resting-state feature data first enters the model through the input layer and is then fed into the first nonlinear mapping layer. This layer can consist of fully connected layers, activation functions, normalization layers, or other neural network structures. It is used to extract latent feature representations from demographic information, basic physical parameters, and resting-state physiological parameters, thereby learning the complex nonlinear interactions between multi-source resting-state features.

[0053] Simultaneously, the model incorporates a conditional encoding layer to generate conditional codes based on the input multi-source resting feature data, i.e., the individual conditional information of the target object. Individual conditional information may include gender, age group, body type, disease state, or a combination thereof. This conditional code is then input to a modulation parameter generation layer to generate modulation parameters that match the latent feature representation. Subsequently, the conditional modulation layer adaptively modulates the latent feature representation output from the first nonlinear mapping layer according to the modulation parameters, obtaining the modulated latent feature representation.

[0054] The conditionally modulated latent feature representation is further input into a second nonlinear mapping layer. This layer is used to learn the complex mapping relationship between resting-state physiological parameters and maximal oxygen uptake based on the modulated features, and then passes the learned high-level feature representation to the output layer. Finally, the output layer generates the predicted maximal oxygen uptake of the target object based on this high-level feature representation.

[0055] In terms of model structure, deep learning prediction models can adopt any one or a combination of multilayer perceptrons, residual networks, attention networks, gating networks, feature interaction networks, or other equivalent neural network structures.

[0056] It should be noted that, unlike the conventional approach of simply concatenating gender, age, or body type information as ordinary input features to the model input, this embodiment uses an independent conditional coding layer and modulation parameter generation mechanism to enable individual conditional information to participate in the dynamic adjustment of the model's implicit representation, rather than merely providing static features at the input layer. This mechanism allows the model to adaptively adjust the response mode of intermediate representations based on the physiological characteristics of different individuals within a unified network framework. This achieves personalized representation modulation for groups of different ages, genders, body types, and disease states, effectively improving the model's adaptability to physiological differences among different individual groups and its predictive generalization performance.

[0057] In one implementation, the conditional codes generated by the conditional coding layer can be generated based on the gender information of the target object, for example, encoding males and females into different conditional categories to characterize the gender attributes of an individual.

[0058] In other implementations, the conditional encoding can also be generated individually from age groups, such as youth, middle-aged, and elderly; body type groups, such as underweight, normal, overweight, and obese based on body mass index; disease states, such as the presence or absence of cardiovascular disease, metabolic disease, etc.; or other clinical features that can characterize an individual's physiological or pathological state.

[0059] Furthermore, the aforementioned multiple individual characteristics can also be jointly encoded, combining any two or more features from age groups, body types, disease states, etc., to generate high-dimensional conditional codes that integrate multidimensional information. These diverse conditional coding generation strategies allow for flexible adaptation to data availability in different clinical scenarios: single-feature coding can be used when only a single feature is available; when richer individual information is available, multi-feature conditional codes can be fused, thereby enabling the model to possess stronger personalized modulation capabilities.

[0060] In one implementation, the step of performing dimension-wise adaptive modulation of the latent feature representation based on the modulation parameters includes: The different dimensions in the latent feature representation are amplified, weakened, or shifted according to the scaling and translation parameters in the modulation parameters to achieve dimension-wise adaptive modulation; wherein, the scaling parameter is used to adjust the intensity of the latent feature, and the translation parameter is used to adjust the degree of latent feature shift.

[0061] Specifically, the conditional modulation layer independently applies corresponding scaling parameters to amplify or weaken each feature dimension in the latent feature representation, and simultaneously applies corresponding translation parameters to shift it positively or negatively, thus allowing different dimensions to receive their own independent adjustments during modulation. Through this dimension-wise adaptive modulation mechanism, the conditional modulation layer can amplify, weaken, or shift different dimensions of the latent features according to the individual conditions of the target object, such as gender, age group, body type, or disease state, enabling the same set of resting physiological parameters to form differentiated internal representations under different individual conditions. This embodiment does not require training independent prediction networks for different populations; instead, within a unified model framework, it dynamically adjusts the response intensity and shift direction of intermediate features in each dimension through modulation parameters driven by conditional encoding, thereby effectively reflecting individual differences.

[0062] In practical applications, conditional modulation methods are not limited to scaling and translation modulation; other equivalent modulation mechanisms such as gating modulation, attention modulation, or parameter generation network modulation can also be used. Regardless of the modulation form used, the goal is to enable the model to adaptively reshape the distribution and expression of latent features based on the individual characteristics of the target object within a unified network structure. This improves the model's adaptability to physiological differences among groups of different ages, genders, body types, and disease states, as well as its predictive generalization performance.

[0063] In one implementation, the maximum oxygen uptake or peak oxygen uptake obtained from a cardiopulmonary exercise test (CPET) report can be used as a supervised learning label during the model training phase to guide parameter optimization of the deep learning prediction model. For the loss function, any one of mean absolute error, mean squared error, Huber loss, or a combination thereof can be used. Preferably, Huber loss combined with L2 regularization is used to reduce the adverse effects of outlier labels or extreme samples on model training, while suppressing overfitting and thus improving generalization ability. For the optimizer, the Adam optimizer, stochastic gradient descent optimizer, or other neural network optimization algorithms can be used for parameter updates. Regarding training strategies, model training can employ methods such as cross-validation, training / validation / test set partitioning, or independent external validation set validation for model performance evaluation and hyperparameter tuning. To further improve training stability, an early stopping mechanism can be introduced, terminating training early when the validation set loss no longer decreases within a set number of rounds to avoid over-iteration.

[0064] Furthermore, the training method can employ a single regression task training approach, directly outputting continuous VO2 max prediction values; alternatively, a high-risk classification task can be added to the regression output, forming a multi-task learning structure. By sharing underlying feature representations, multi-task learning can leverage the additional supervision signals provided by the classification auxiliary task to guide the model in learning more discriminative and robust feature representations, thereby further improving the accuracy of regression predictions and clinical decision support capabilities. Overall, model training can use VO2 max or peak VO2 max from CPET sources as supervision labels, and comprehensively employ at least one or more combinations of Huber loss, L2 regularization, Adam optimizer, early stopping mechanism, and cross-validation strategies to effectively improve the stability of the training process and the generalization ability of the final model.

[0065] In one implementation, the method further includes: Downstream auxiliary applications are based on the predicted maximum oxygen uptake; the downstream auxiliary applications include at least one of the following: high-risk identification, preoperative risk assessment, rehabilitation management, follow-up trend monitoring, and clinical stratification management.

[0066] Specifically, after obtaining the predicted maximum oxygen uptake of the target subject through a deep learning prediction model, the prediction results are further used to perform at least one clinical decision support or health management task, thereby transforming a single quantitative prediction value into assessment information with greater clinical application value. For example... Figure 4 As shown, downstream auxiliary applications include at least one of the following: high-risk identification, preoperative risk assessment, rehabilitation management, follow-up trend monitoring, clinical stratification management (also known as patient stratification management), and comparison of clinical profiles under predicted stratification. For example, the system can use predicted maximal oxygen uptake to support auxiliary scenarios such as high-risk population identification, preoperative risk assessment, clinical stratification management, and follow-up trend monitoring.

[0067] In terms of output format, it can directly output continuous predicted VO2 max values, or generate risk alerts or classification labels based on preset thresholds. For example, individuals with predicted VO2 max values ​​below a preset threshold can be marked as high-risk groups, thereby achieving automated high-risk screening. The threshold can be flexibly set according to clinical application scenarios, guideline standards, training data distribution, or the actual needs of medical institutions to adapt to risk assessment standards in different clinical environments.

[0068] It is important to emphasize that the application results of this embodiment are only intended to provide auxiliary reference for clinicians and do not replace the final medical decision made by doctors based on medical history, physical signs, examination results, and clinical judgment. Through the aforementioned downstream auxiliary applications, not only can continuous predicted VO2 max values ​​be output, but high-risk identification results can also be output based on preset thresholds, further serving downstream auxiliary applications. In short, this embodiment can transform single numerical prediction results into auxiliary assessment information that is more convenient for clinical use, enabling VO2 max prediction results to truly serve practical application scenarios such as clinical screening, risk alerts, and follow-up management, thus improving its operability and practical value in clinical practice.

[0069] In summary, the key technical points and advantages of this application include: (1) By collecting multi-source physiological parameters under resting conditions and using them for VO2 max prediction, this application reduces the reliance on standard exercise load testing and can be applied to target subjects who cannot complete exercise testing, are not suitable for exercise load testing, or need rapid screening, thereby improving the accessibility and low burden of VO2 max assessment.

[0070] (2) By performing unified preprocessing and feature fusion on demographic information, basic physical parameters and resting physiological parameters, this application can reduce the impact of different data sources, different dimensions, missing values ​​and outliers on model input, and improve the consistency of input data and the stability of model prediction.

[0071] (3) By setting numerical feature representation branch and categorical feature encoding branch, this application can process continuous physiological parameters and categorical individual features respectively and convert them into a unified deep learning input representation, thereby improving the model's ability to adapt to heterogeneous clinical data.

[0072] (4) By setting a conditional coding layer, a modulation parameter generation layer, and a conditional modulation layer in the deep learning model, this application can generate feature scaling parameters and translation parameters according to individual conditions, and adaptively modulate the latent representation. This structure can enhance the model's ability to express the physiological differences of different individual groups and avoid the insufficient characterization of individual differences caused by relying entirely on a single shared mapping function.

[0073] (5) By using a nonlinear deep learning structure to establish the mapping relationship between multi-source resting parameters and maximum oxygen uptake, this application can more fully characterize the nonlinear relationship between complex physiological parameters than simple empirical formulas or linear models, thereby improving the expressive power of maximum oxygen uptake prediction.

[0074] (6) By further using the predicted maximum oxygen uptake for high-risk identification, preoperative risk assessment, follow-up trend monitoring and clinical stratification management, this application can transform the model output into auxiliary assessment information that is more convenient for clinical use, thereby improving the availability of the prediction results in practical application scenarios.

[0075] Based on the above embodiments, this application also provides a maximum oxygen uptake prediction system based on multi-source resting physiological parameters, such as... Figure 5 As shown, the system includes: The data acquisition module is used to acquire multi-source resting data of the target object; The data preprocessing module is used to perform quality control and feature preprocessing based on the multi-source resting data to obtain multi-source preprocessed feature data; The feature fusion module is used to characterize and fuse the multi-source preprocessed feature data to obtain multi-source resting feature data. The maximum oxygen uptake prediction module is used to input the multi-source resting feature data into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object. The structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

[0076] Specifically, the system in this embodiment may include a data acquisition module, a data preprocessing module, a feature fusion module, a maximum oxygen uptake prediction module, and a result output module. The data acquisition module is used to acquire multi-source resting data of the target object; the data preprocessing module is used to perform quality control and standardization processing on the raw multi-source resting data; the feature fusion module is used to convert different types of input features into a unified model input representation; the maximum oxygen uptake prediction module is used to generate prediction results based on a deep learning model; and the result output module is used to output the predicted maximum oxygen uptake, high-risk warnings, report information, or other auxiliary assessment results.

[0077] Furthermore, the system in this embodiment may also include a data storage and management module, a clinical application interface module, and a user terminal module. In terms of deployment, the system in this embodiment can be deployed on a local server, a hospital information system, a cloud server, a doctor's workstation, or a mobile terminal, and can interact with existing medical information systems through interfaces.

[0078] Based on the above embodiments, this application also provides a terminal, the principle block diagram of which can be as follows: Figure 6 As shown, the terminal includes a processor, memory, network interface, and display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for predicting maximum oxygen uptake based on multi-source resting physiological parameters. The display screen can be an LCD screen or an e-ink screen.

[0079] Those skilled in the art will understand that Figure 6 The block diagram shown is only a partial structural diagram related to the solution of this application and does not constitute a limitation on the terminal on which the solution of this application is applied. The specific terminal may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0080] In one implementation, the terminal's memory stores one or more programs, and these programs are configured to be executed by one or more processors, and the programs include instructions for performing a method for predicting maximum oxygen uptake based on multi-source resting physiological parameters.

[0081] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0082] In summary, this application discloses a method and system for predicting maximum oxygen uptake based on multi-source resting physiological parameters. The method includes: acquiring multi-source resting data of a target object; performing quality control and feature preprocessing on the multi-source resting data to obtain multi-source preprocessed feature data; characterizing and fusing the multi-source preprocessed feature data to obtain multi-source resting feature data; and inputting the multi-source resting feature data into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object. The deep learning prediction model includes: a first nonlinear mapping layer for extracting latent feature representations based on the multi-source resting feature data; a conditional encoding layer for generating conditional codes based on the multi-source resting feature data; a modulation parameter generation layer for generating modulation parameters based on the latent feature representations and the conditional codes; a conditional modulation layer for performing dimension-wise adaptive modulation on the latent feature representations based on the modulation parameters to obtain modulated latent feature representations; a second nonlinear mapping layer for generating high-level feature representations based on the modulated latent feature representations; and an output layer for generating the predicted maximum oxygen uptake of the target object based on the high-level feature representations.

[0083] It should be understood that the application of this application is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for predicting maximum oxygen uptake based on multi-source resting physiological parameters, characterized in that, The method includes: Acquire multi-source static data of the target object; Quality control and feature preprocessing are performed on the multi-source resting data to obtain multi-source preprocessed feature data; The multi-source preprocessed feature data is characterized and fused to obtain multi-source resting feature data. The multi-source resting feature data is input into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object; The structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

2. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The multi-source resting data includes demographic information, basic physical parameters, and physiological parameters at rest; The demographic information includes age and / or sex; the basic physical parameters include at least one of height, weight, and body mass index; the physiological parameters at rest include at least one of resting heart rate, blood pressure, respiratory rate, ventilation, oxygen uptake, carbon dioxide excretion, respiratory exchange rate, and metabolic equivalent.

3. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The quality control steps include: The multi-source static data is processed using at least one of the following methods: field integrity verification, data type verification, unit unification, outlier identification, duplicate record processing, non-final record exclusion, and abnormal data removal or marking.

4. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The feature preprocessing steps include: For numerical features in multi-source resting data after quality control, at least one of the following imputation methods—median imputation, mean imputation, model imputation, and clinical rule-based imputation—is used to handle missing values, and then standardized or normalized. For categorical features in multi-source static data after quality control, at least one of the following processing methods is used for feature processing: missing category imputation, one-hot encoding, and embedding encoding.

5. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The steps for characterizing and fusing the multi-source preprocessed feature data to obtain multi-source resting feature data include: Numerical resting physiological parameters are input into the numerical feature representation branch to form a numerical feature vector, and categorical features are input into the categorical feature encoding branch to form a categorical feature vector. Then, a unified multi-source resting feature vector is formed by splicing, weighted fusion, or gated fusion.

6. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The step of performing dimension-wise adaptive modulation of the latent feature representation based on the modulation parameters includes: The different dimensions in the latent feature representation are amplified, weakened, or shifted according to the scaling and translation parameters in the modulation parameters to achieve dimension-wise adaptive modulation; wherein, the scaling parameter is used to adjust the intensity of the latent feature, and the translation parameter is used to adjust the degree of latent feature shift.

7. The method for predicting maximum oxygen uptake based on multi-source resting physiological parameters according to claim 1, characterized in that, The method further includes: Downstream auxiliary applications are based on the predicted maximum oxygen uptake; the downstream auxiliary applications include at least one of the following: high-risk identification, preoperative risk assessment, rehabilitation management, follow-up trend monitoring, and clinical stratification management.

8. A system for predicting maximum oxygen uptake based on multi-source resting physiological parameters, characterized in that, The system includes: The data acquisition module is used to acquire multi-source resting data of the target object; The data preprocessing module is used to perform quality control and feature preprocessing based on the multi-source resting data to obtain multi-source preprocessed feature data; The feature fusion module is used to characterize and fuse the multi-source preprocessed feature data to obtain multi-source resting feature data. The maximum oxygen uptake prediction module is used to input the multi-source resting feature data into a deep learning prediction model to obtain the predicted maximum oxygen uptake of the target object. The structure of the deep learning prediction model includes: The first nonlinear mapping layer is used to extract latent feature representations based on the multi-source resting feature data; A conditional coding layer is used to generate conditional codes based on the multi-source resting feature data; A modulation parameter generation layer is used to generate modulation parameters based on the implicit feature representation and the conditional coding. A conditional modulation layer is used to perform dimension-wise adaptive modulation on the latent feature representation according to the modulation parameters to obtain the modulated latent feature representation; The second nonlinear mapping layer is used to generate a high-level feature representation based on the modulated implicit feature representation; An output layer is used to generate the predicted maximum oxygen uptake of the target object based on the high-level feature representation.

9. A terminal, characterized in that, The terminal includes a memory and one or more processors; the memory stores one or more programs; the programs contain instructions for executing the maximum oxygen uptake prediction method based on multi-source resting physiological parameters as described in any one of claims 1 to 7; the processor is used to execute the programs.

10. A computer-readable storage medium storing a plurality of instructions thereon, characterized in that, The instructions are applicable to be loaded and executed by a processor to implement the steps of the method for predicting maximum oxygen uptake based on multi-source resting physiological parameters as described in any one of claims 1 to 7.

Images