Oxygen uptake cardiopulmonary endurance test method based on load turn-back and human brain structure

Abstract

The invention discloses an oxygen uptake cardiopulmonary endurance testing method based on load turn-back walking and a human brain structure, and relates to the technical field of exercise physiology and biomedicine detection.The method comprises the steps that progressive load turn-back walking exercise data and a high-resolution brain structure image of a subject are obtained; extracting key brain region parameters such as the thickness of the anterior cinerary cortex, the grey matter density of island leaves and the signal intensity of brainstem respiratory center; normalizing the motion features and brain structure parameters and then constructing a 12-dimensional multi-modal fusion feature vector; and inputting a deep feedforward neural network model to predict the maximum oxygen uptake, and performing five-level cardiopulmonary endurance evaluation according to an age and gender correction result. According to the method, quantitative association between peripheral exercise performance and a central nervous anatomy basis can be realized, the heart and lung endurance evaluation precision of old people and individuals with abnormal neurological functions is remarkably improved, and rapid screening and personalized exercise prescription making are supported.

Description

A method for testing oxygen uptake and cardiorespiratory endurance based on load shuttle walk and human brain structure Technical Field

[0001] This invention relates to the fields of exercise physiology and biomedical testing technology, and in particular to a method for testing cardiopulmonary endurance based on load shuttle walk and human brain structure. Background Technology

[0002] With the continuous evolution of cardiorespiratory endurance assessment technology, non-invasive testing methods based on the dynamic correlation between exercise load and oxygen uptake have become important tools in clinical and health monitoring fields. Cardiorespiratory endurance, as a core indicator reflecting an individual's aerobic metabolic capacity, is typically quantified by measuring or predicting VO2max. Among various exercise modalities, shuttle walking is widely adopted as a standardized incremental load protocol due to its simplicity, high safety, and moderate physical requirements for subjects, used to construct VO2max prediction models suitable for large-scale population screening. This type of method records the walking distance completed by subjects within a specified time and combines it with basic anthropometric parameters such as height and weight, using statistical regression techniques to estimate their cardiorespiratory function level. It has demonstrated good feasibility and cost-effectiveness in public health and sports medicine practice.

[0003] Among them, the cardiorespiratory endurance test method based on the correlation between load shuttle walk and oxygen uptake focuses on converting exercise output behavior into a proxy indicator of physiological capacity. Its basic principle is that as the speed or duration of shuttle walk increases, the body's oxygen demand increases approximately linearly until it reaches the individual's upper limit of aerobic capacity; during this process, there is a significant correlation between the total walking distance and the measured VO2max, thus providing a theoretical basis for indirect assessment. However, the modeling logic of this method has long been limited to the combination of peripheral physiological variables and macroscopic exercise performance, failing to fully incorporate the intrinsic mechanisms of the central nervous system's regulation of cardiorespiratory function, leading to systematic bias in models when facing populations with significant neurophysiological heterogeneity.

[0004] In existing technologies, such as those disclosed in publications CN109620234A and CN109620234B, although VO2max prediction models based on back-and-forth walking distance and body surface parameters have been constructed, their data dimensions completely exclude the influence of human brain structural characteristics. Specifically, these methods do not consider the volume of gray matter in the cerebral cortex, the integrity of white matter fiber bundles, or the morphological or connectivity characteristics of key regulatory regions. These neuroanatomical elements directly participate in higher physiological processes such as motor intention generation, respiratory rhythm regulation, sympathetic-parasympathetic balance control, and fatigue perception integration. Due to the lack of quantitative characterization of the above-mentioned central regulatory factors, existing models cannot explain why, under the same height, weight, and walking distance, there are significant differences in the actual oxygen uptake response of different individuals, especially in the elderly, Parkinson's disease patients, stroke rehabilitation populations, or individuals with mild cognitive impairment, where the prediction error is significantly increased. In addition, traditional regression frameworks are difficult to integrate multimodal heterogeneous data and have not established a mapping relationship between brain structural characteristics and exercise-oxygen uptake coupling efficiency, resulting in assessment results lacking physiological mechanism support and individualized adaptability. Summary of the Invention

[0005] The purpose of this invention is to provide a method for testing cardiopulmonary endurance based on load-bearing back-and-forth walking and human brain structure, in order to solve the technical problem that existing cardiopulmonary endurance assessment models do not integrate the central nervous system regulatory mechanisms, resulting in decreased prediction accuracy in populations with neurophysiological heterogeneity.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] Methods for testing cardiopulmonary endurance based on load shuttle walk and human brain structure include:

[0008] Step S1: Obtain the subject's load shuttle walk exercise data. Perform a progressive speed shuttle walk test on a standardized track. Record the actual walking distance, total time, cadence sequence, and dynamic heart rate response curve for each stage of the prescribed number of round trips. The starting speed of the test is set to 1.0 m / s, increasing by 0.2 m / s in each stage. The duration of each stage is 60 seconds, until the subject can no longer maintain the prescribed rhythm or the subjective fatigue score reaches 18 points or more.

[0009] Step S2: Collect brain structural imaging data of the subjects, use a 3.0 Tesla magnetic resonance imaging system to obtain high-resolution T1 weighted images and diffusion tensor imaging data, reconstruct the whole brain gray matter volume distribution map and white matter fiber tract connectivity matrix, and quantitatively extract the anterior cingulate cortex thickness, insular gray matter density, mean voxel intensity of the brainstem respiratory center region and anisotropy fraction of the corpus callosum genu.

[0010] Step S3: Construct a multimodal fusion feature vector by normalizing the maximum number of completed stages, peak heart rate, and heart rate rise slope in the late stage of exercise data with the anterior cingulate cortex thickness, insular gray matter density, and brainstem region signal intensity in the brain structural parameters, and then jointly encoding them to form a spatial vector containing 12-dimensional physiological-neural joint features.

[0011] Step S4: Establish a nonlinear mapping model for oxygen uptake prediction, using a deep feedforward neural network architecture. The input layer receives the multimodal fusion feature vector, and the hidden layer contains three fully connected layers with 64, 32, and 16 neurons respectively. The activation function is a modified linear unit. The output layer generates the predicted value of the individual's maximum oxygen uptake. During model training, an adaptive moment estimation optimization algorithm is used. The loss function is defined as the square root of the mean square error between the predicted and measured values. The training set accounts for 80% of the total sample size, the validation set accounts for 10%, and the test set accounts for 10%.

[0012] Step S5: Implement individualized cardiorespiratory endurance level assessment. Divide the predicted maximum oxygen uptake value into 5 levels according to the percentile after age and gender correction. Level 1 corresponds to below the 20th percentile, Level 2 is from the 20th to the 40th percentile, Level 3 is from the 40th to the 60th percentile, Level 4 is from the 60th to the 80th percentile, and Level 5 is above the 80th percentile. Generate a visual assessment report, including a heatmap of the contribution of exercise performance and a radar map of brain region weight distribution.

[0013] In step S1, the standardized running track is 20 meters long, and pressure sensing pads are set at both ends to accurately capture the moment of foot contact. The cadence sequence is obtained by sampling through a wearable inertial measurement unit with a sampling frequency of 100 Hz. The dynamic heart rate response curve is continuously recorded by a chest strap heart rate monitor with a time resolution of 1 second.

[0014] In step S2, the T1-weighted image spatial resolution is 1 mm x 1 mm x 1 mm, the diffusion tensor imaging parameters are set to b = 1000 seconds per square millimeter, the number of gradient directions is 64, the average number of voxels is 2, the image reconstruction uses the Siemens three-dimensional fast gradient echo sequence, the anterior cingulate cortex thickness measurement range is limited to Brodmann 24 and 32 zones, and the mean intercortical distance is calculated using the surface substrate modeling method.

[0015] The normalization process in step S3 uses Z-score transformation, as shown in the following formula: , where x represents the original feature value, μ represents the overall mean of the feature in the training set, and σ represents the standard deviation. During the joint encoding process, cross terms are introduced to construct interactive features. For example, the product of the anterior cingulate cortex thickness and the slope of the heart rate rise in the later stage of exercise is included in the feature space.

[0016] In step S4, the input features of the deep feedforward neural network also include body mass index, resting heart rate, and systolic blood pressure. The upper limit of the number of model training iterations is set to 500 rounds. The early stop mechanism is triggered when the validation set loss does not decrease for 10 consecutive rounds. The initial learning rate is set to 0.001, and it decays to 0.9 times the original value every 100 rounds. Finally, the model's coefficient of determination R² on the independent test set is greater than or equal to 0.91, and the mean absolute error is less than or equal to 1.2 ml / kg / min.

[0017] In step S5, age correction uses a piecewise linear regression model. For men aged 20 to 50, the age decreases by 0.55% per year, and the decrease increases to 0.75% for those over 50. The corresponding decrease rates for women are 0.45% and 0.65%. The gender correction factor is set so that the baseline value for men is 8.3% higher than that for women. The visual assessment report is generated using HTML5 and JavaScript technology stacks and supports mobile browsing and data export functions.

[0018] The test also includes: repeating the test procedure after subjects undergo 12 weeks of aerobic intervention training, comparing the predicted change in oxygen uptake ΔVO2max before and after the intervention, and judging it as a significant improvement if ΔVO2max is greater than or equal to 1.5 ml per kilogram per minute. At the same time, the correlation between the degree of improvement in white matter integrity and ΔVO2max in brain structural parameters is analyzed, and a strong correlation is considered to exist when the correlation coefficient r is greater than 0.6.

[0019] The multimodal fusion feature vector introduces a dynamic coupling index, defined as the ratio of the heart rate rise slope in the later stages of exercise to the activation potential of the anterior cingulate cortex (ACC). The ACAC activation potential is determined by the gray matter density and local consistency score of this region, and is calculated using the following formula: GM represents the actual gray matter density, FC represents the functional connectivity strength, and the subscript 0 represents the average reference value for healthy young people.

[0020] The oxygen uptake prediction nonlinear mapping model is deployed on edge computing terminal devices with an inference latency of less than 200 milliseconds. It supports offline operation and uses the national cryptographic SM4 algorithm for data encryption to ensure that the security and privacy protection of the subjects' biomedical information meet the requirements of the national level 2 information security protection.

[0021] The method is applied to large-scale screening of elderly people in the community. It can complete no less than 150 tests per day. The entire test process takes less than 18 minutes, including 9 minutes for the exercise test and 9 minutes for the preparation and recovery phases. The system's automatic interpretation accuracy rate has reached 93.7% after double-blind verification.

[0022] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0023] This invention integrates load-bearing back-and-forth walking exercise data with the structural features of key regulatory regions of the human brain, and for the first time establishes a quantitative correlation model between peripheral motor performance and the anatomical basis of the central nervous system, breaking through the limitations of traditional cardiopulmonary endurance assessment that relies solely on surface parameters and macroscopic motor output.

[0024] This invention utilizes high-field magnetic resonance imaging to obtain morphological parameters of core brain regions such as the anterior cingulate cortex, insula, and brainstem respiratory center, and integrates these parameters with physiological signals such as dynamic heart rate response and gait stability during backwalking to construct a biologically sound oxygen uptake prediction framework.

[0025] The deep neural network model used in this invention can capture nonlinear interaction effects, significantly improving the accuracy of cardiopulmonary function assessment for patients in the early stages of neurodegenerative diseases, individuals with subclinical cognitive impairment, and the elderly. The test results have a determination coefficient of over 0.91 on the independent validation set, and the mean absolute error is less than 1.2 ml / kg / min, which is about 37% better than existing methods based on pure motion distance regression.

[0026] This invention introduces a dynamic coupling index and a personalized grading mechanism, achieving a leap from "single numerical prediction" to "mechanism analysis + risk stratification". It can not only be used for monitoring the efficacy of clinical rehabilitation, but also guide the formulation of precise exercise prescriptions. The system supports local deployment and rapid screening applications, meets the needs of large-sample epidemiological research, and has important promotional value in the construction of public health service system. Attached Figure Description

[0027] Figure 1 is a schematic diagram of the overall technical scheme of the oxygen uptake cardiopulmonary endurance test method based on load back-and-forth walking and human brain structure proposed in this invention. Detailed Implementation

[0028] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0029] Example 1

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0031] Currently, in the field of cardiorespiratory endurance assessment, non-invasive testing methods based on the dynamic correlation between load shuttle walking and oxygen uptake are widely used. However, their modeling logic has long been limited to the combination of peripheral physiological variables and macroscopic exercise performance, failing to fully incorporate the intrinsic mechanisms of the central nervous system's regulation of cardiorespiratory function. This results in systematic bias in models when faced with populations exhibiting significant neurophysiological heterogeneity. To address these technical problems, this invention proposes a biologically sound framework for predicting oxygen uptake by integrating load shuttle walking exercise behavior data with structural features of key regulatory regions in the human brain. This framework is then applied to a cardiorespiratory endurance testing method based on load shuttle walking and human brain structure.

[0032] Referring to Figure 1, the overall technical architecture of this invention comprises five core processing stages: motion data acquisition, brain structural image acquisition, multimodal feature fusion, nonlinear oxygen uptake mapping modeling, and individualized level assessment. The entire process begins with a standardized shuttle walk test, simultaneously or sequentially acquires high-resolution brain structural images, and ultimately outputs a mechanistically interpretable predicted maximum oxygen uptake value and a visual assessment report.

[0033] Step S1 involves acquiring the subject's load-bearing shuttle walk data. A progressive speed shuttle walk test is performed on a standardized track, recording the actual walking distance, total time, cadence sequence, and dynamic heart rate response curve for each stage of the prescribed number of round trips. Specifically, in Step S1, the initial test speed is set to 1.0 m / s, increasing by 0.2 m / s per stage, with each stage lasting 60 seconds, until the subject can no longer maintain the prescribed pace or their subjective fatigue score reaches 18 points or higher. This test protocol is designed based on a variant of the classic Beep Test, ensuring a linear increase in exercise intensity to elicit a physiological response close to maximal oxygen uptake. The standardized track is 20 meters long, with pressure-sensing pads at both ends to accurately capture the foot contact time, used to calculate the actual number of round trips and the effective walking distance. The cadence sequence was obtained through sampling using a wearable inertial measurement unit (IMU) at a sampling frequency of 100 Hz. This unit, fixed to the subject's lumbosacral region, recorded triaxial acceleration and angular velocity signals in real time. A zero-velocity correction algorithm was used to eliminate integral drift, thereby extracting the ground contact time and take-off time for each step, forming a high-temporal-resolution cadence sequence. The dynamic heart rate response curve was continuously recorded by a chest-strap heart rate monitor with a temporal resolution of 1 second. This device employed photoplethysmography (PPG) technology and transmitted data in real time to a central processing terminal via Bluetooth 5.0. The motion data acquisition process was controlled by dedicated testing software. The software interface displayed the current speed, remaining time, cumulative distance, and instantaneous heart rate in real time, while automatically recording the subject's subjective fatigue score (using the Borg 6-20 scale) at the end of each phase. When the subject voluntarily terminated the test or failed to complete all round trips within the specified time, the system automatically marked the test endpoint and saved a complete motion trajectory log. The log contains start and end timestamps for each stage, number of round trips completed, average speed within the stage, instantaneous cadence fluctuation coefficient, and heart rate rise slope, providing a raw data foundation for subsequent feature extraction.

[0034] Step S2 involves acquiring brain structural imaging data of the subject. A 3.0 Tesla magnetic resonance imaging system is used to obtain high-resolution T1-weighted images and diffusion tensor imaging data. Whole-brain gray matter volume distribution maps and white matter fiber tract connectivity matrices are reconstructed, and the thickness of the anterior cingulate cortex, gray matter density of the insula, mean voxel intensity of the brainstem respiratory center, and anisotropy fraction of the genu of the corpus callosum are quantitatively extracted. Specifically, in step S2, the T1-weighted image spatial resolution is 1 mm x 1 mm x 1 mm, acquired using a Siemens three-dimensional fast gradient echo sequence, with a repetition time (TR) of 2300 ms, an echo time (TE) of 2.98 ms, a flip angle of 9 degrees, and an acquisition time of 5 minutes and 30 seconds. The diffusion tensor imaging parameters are set to a b-value of 1000 seconds per square millimeter, a gradient direction number of 64, an intravoxel mean number of times of change of 2, a spatial resolution of 2 mm with isotropic properties, and a total acquisition time of approximately 12 minutes. Image reconstruction was performed on the scanner's local workstation, accelerated using parallel imaging technology, and an eddy current distortion correction algorithm was applied to eliminate geometric distortion caused by gradient switching. Subsequently, the original DICOM format images were imported into the FSL and FreeSurfer joint processing pipeline. First, T1-weighted images underwent N4 off-field correction, skull dissection, and tissue segmentation to generate probability maps of gray matter, white matter, and cerebrospinal fluid. Using FreeSurfer's surface-base modeling method, the thickness of the entire cerebral cortex was calculated, with the measurement range of the anterior cingulate cortex thickness specifically limited to Brodmann areas 24 and 32, regions proven to be closely related to autonomic regulation and motor intention generation. Its thickness was defined as the mean of the shortest distance from the gray matter-white matter boundary to the gray matter-cerebrospinal fluid boundary at all vertices within this anatomical region. Second, diffusion tensor imaging data, after Eddy current correction, was fitted with a tensor model to generate anisotropic fractional FA maps. Using a deterministic fiber tract tracking algorithm, the main white matter pathways were reconstructed, and the average FA value of the genu of the corpus callosum was extracted as an indicator of cross-hemispheric information integration efficiency. Simultaneously, the brainstem respiratory center region was manually delineated in the ventrolateral medulla oblongata. Referring to standard anatomical atlases, the mean signal intensity of all voxels within this region on T1-weighted images was calculated to characterize local tissue density. Insular gray matter density was obtained by multiplying the segmented insular mask by the gray matter probability map, summing the results, and then dividing by the total number of voxels. All brain structural parameters were extracted independently by two experienced neuroimaging analysts under double-blind conditions. Discrepancies exceeding 5% were arbitrated by a third party to ensure data reliability.

[0035] Step S3 involves constructing a multimodal fusion feature vector. This involves normalizing the maximum number of completed stages, peak heart rate, and the slope of heart rate increase in the later stages of exercise from the motion data, and then jointly encoding these parameters with the anterior cingulate cortex thickness, insular gray matter density, and brainstem region signal intensity from the brain structural parameters. This results in a spatial vector containing 12-dimensional physiological-neurological joint features. Specifically, in step S3, the normalization process uses Z-score transform, the mathematical expression of which is:

[0036]

[0037] Where x represents the original feature value, μ represents the overall mean of the feature in the training set, and σ represents the standard deviation. This transformation maps features of different dimensions to a standard normal distribution space with a mean of 0 and a standard deviation of 1, eliminating the adverse effects of scale differences on model training. In the joint encoding process, in addition to the six basic features (maximum number of completed stages, peak heart rate, late-exercise heart rate rise slope, anterior cingulate cortex thickness, insular gray matter density, and brainstem region signal intensity), interactive features are constructed by introducing cross-terms. For example, the product of anterior cingulate cortex thickness and late-exercise heart rate rise slope is included in the feature space to capture the synergistic effect between central regulatory capacity and peripheral cardiovascular response. Furthermore, a dynamic coupling index is introduced into the multimodal fusion feature vector, defined as the ratio of late-exercise heart rate rise slope to the anterior cingulate cortex activation potential. The anterior cingulate cortex activation potential is determined by the region's gray matter density and local consistency score, and its calculation formula is:

[0038]

[0039] Wherein, GM represents actual gray matter density, and FC represents functional connectivity strength, specifically the resting-state functional connectivity strength between the anterior cingulate cortex and the insula, obtained through independent component analysis; the subscript 0 represents the mean reference value in healthy young adults. This dynamic coupling index aims to quantify the cardiovascular response efficiency driven by a unit of central activation and is a key biomarker reflecting the neuro-cardiopulmonary coupling efficacy. The final 12-dimensional feature vector also includes body mass index, resting heart rate, and systolic blood pressure, these traditional risk factors have been shown to provide additional predictive gain. All features underwent missing value imputation (using the K-nearest neighbor algorithm) and outlier truncation (set to mean ± 3 standard deviation) before being input into the model to ensure data quality.

[0040] Step S4: Establish a nonlinear mapping model for predicting oxygen uptake using a deep feedforward neural network architecture. The input layer receives the multimodal fusion feature vector, and the hidden layer contains three fully connected layers with 64, 32, and 16 neurons respectively. The activation function is a modified linear unit (MRU). The output layer generates the predicted maximum oxygen uptake value for each individual. Specifically, in step S4, an adaptive moment estimation optimization algorithm is used during model training. The loss function is defined as the square root of the mean square error between the predicted and measured values. The training set accounts for 80% of the total sample size, the validation set accounts for 10%, and the test set accounts for 10%. The input feature dimension of the deep feedforward neural network is 12, corresponding to the aforementioned multimodal fusion feature vector. The first hidden layer contains 64 neurons, receiving all input features and linearly combining them using a weight matrix, then introducing nonlinearity through a modified linear unit activation function. Its output is a high-dimensional abstract feature representation. The second hidden layer contains 32 neurons, further compressing and refining the output of the first layer. The third hidden layer contains 16 neurons, forming a bottleneck layer to prevent overfitting and promote feature generalization. The output layer consists of a single neuron, directly outputting the predicted maximum oxygen uptake in milliliters per kilogram per minute (mL / kg / min). The maximum number of training iterations is set to 500 epochs, with an early stopping mechanism triggered when the validation set loss does not decrease for 10 consecutive epochs to avoid overfitting. The initial learning rate is set to 0.001, decaying to 0.9 times its original value every 100 epochs. L2 regularization (weight decay coefficient of 0.0001) is used to constrain model complexity. Real oxygen uptake labels used for training are simultaneously collected by a gold-standard laboratory gas metabolism analyzer when subjects complete the same shuttle walk test, ensuring the accuracy of the supervision signal. The final model achieves a determination coefficient R² greater than or equal to 0.91 and a mean absolute error less than or equal to 1.2 mL / kg / min on the independent test set, significantly outperforming traditional regression models that only use movement distance and body surface parameters. Deployed on edge computing terminals, the model has an inference latency of less than 200 milliseconds, supports offline operation, and uses the national standard SM4 encryption algorithm to ensure the security and privacy protection of subjects' biomedical information meets the national level-two information security protection requirements.

[0041] Step S5 involves implementing an individualized cardiorespiratory endurance level assessment. The predicted maximum oxygen uptake (VO2 max) value is divided into five levels based on age and gender-adjusted percentiles: Level 1 corresponds to below the 20th percentile, Level 2 to the 20th to 40th percentiles, Level 3 to the 40th to 60th percentiles, Level 4 to the 60th to 80th percentiles, and Level 5 above the 80th percentile. A visual assessment report is generated, including a heatmap of the contribution of exercise performance and a radar map of brain region weight distribution. Specifically, in Step S5, age adjustment uses a piecewise linear regression model. For men, the decrease is 0.55% per year between the ages of 20 and 50, and the decrease increases to 0.75% for those over 50. The corresponding decrease rates for women are 0.45% and 0.65%, respectively. The gender adjustment factor is set so that the baseline value for men is 8.3% higher than that for women. The corrected oxygen uptake values ​​were mapped to a norm database established through large-scale epidemiological studies. This database is divided into 5-year age groups, storing the percentile distribution of oxygen uptake for each age group and gender. The rating results not only reflect the absolute cardiopulmonary function level but also the relative position compared to peers of the same age and gender. The visualized assessment report is generated using HTML5 and JavaScript, supporting mobile browsing and data export. The first page of the report displays the rating labels and brief interpretations. The second page is a heatmap of the contribution of exercise performance, which calculates the contribution of each exercise feature to the final prediction using SHAP values ​​and presents it intuitively with color gradations. The third page is a radar map of brain region weight distribution, showing the SHAP values ​​of key brain region structural parameters such as the anterior cingulate cortex, insula, and brainstem, revealing the relative importance of central nervous system factors in individual cardiopulmonary endurance assessment. This report provides mechanistic decision support for clinicians or health managers.

[0042] To further verify the effectiveness and practicality of this invention, a specific application example was constructed. 120 elderly individuals aged 65 to 80 years from the community were selected as subjects, including 30 patients with mild cognitive impairment. All subjects first completed the complete testing procedure described in this invention: a progressive shuttle walk test on a 20-meter standardized track, simultaneously wearing a heart rate monitoring device and an inertial measurement unit; subsequently, a 3.0 Tesla MRI scan was completed within 48 hours. The acquired data, after processing according to the above steps, generated individualized oxygen uptake prediction values ​​and grade assessment reports. The results showed that the prediction error of the method of this invention for patients with mild cognitive impairment (mean absolute error of 1.15 ml / kg / min) was significantly lower than that of the traditional model based solely on shuttle walk distance (mean absolute error of 1.82 ml / kg / min), demonstrating the effectiveness of incorporating brain structure information. Subsequently, 60 of these subjects underwent a 12-week aerobic intervention training program (3 times a week, 40 minutes of moderate-intensity brisk walking each time). After the intervention, the testing procedure was repeated. The predicted change in oxygen uptake (ΔVO2max) before and after the intervention was compared. A significant improvement was defined as ΔVO2max being greater than or equal to 1.5 ml / kg / min. Analysis revealed that 38 subjects met the significant improvement criteria. Further analysis of changes in brain structural parameters showed a correlation coefficient (r) of 0.63 between the improvement in the anisotropy fraction of the corpus callosum genu and ΔVO2max, indicating a strong association between improved white matter integrity and improved cardiopulmonary function. This example not only verifies the high accuracy of this invention in elderly screening scenarios but also demonstrates its unique value in monitoring rehabilitation efficacy and exploring mechanisms. The entire testing process takes less than 18 minutes, with the exercise test portion taking 9 minutes and the preparation and recovery phases totaling 9 minutes. It can efficiently complete no fewer than 150 tests per day. The system's automatic interpretation accuracy reached 93.7% after double-blind validation, fully meeting the needs of large-scale community screening.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for testing cardiopulmonary endurance based on load shuttle walk and human brain structure, characterized in that: The specific steps include: Step S1: Acquire the subject's load-bearing shuttle walk exercise data. Perform a progressively increasing speed shuttle walk test on a standardized track, and record the actual walking distance, total time, cadence sequence, and dynamic heart rate response curve for each stage of the specified number of round trips; Step S2: Collect the subject's brain structure imaging data. Use a 3.0 Tesla magnetic resonance imaging system to acquire high-resolution T1-weighted images and diffusion tensor imaging data, reconstruct the whole-brain gray matter volume distribution map and white matter fiber tract connectivity matrix, and quantitatively extract the anterior cingulate cortex thickness, insular gray matter density, mean voxel intensity of the brainstem respiratory center region, and anisotropy fraction of the corpus callosum genu; Step S3: Construct a multimodal fusion feature vector, incorporating the maximum number of completed stages and peak heart rate from the exercise data. Step S4: The slope of heart rate rise during the later stage of exercise is normalized and jointly encoded with the anterior cingulate cortex thickness, insular gray matter density, and brainstem signal intensity parameters to form a spatial vector containing 12-dimensional physiological-neurological joint features; Step S5: Establish a nonlinear mapping model for oxygen uptake prediction, using a deep feedforward neural network architecture. The input layer receives the multimodal fusion feature vector, the hidden layer contains 3 fully connected layers, the activation function is a modified linear unit, and the output layer generates the individual's maximum oxygen uptake prediction value; Step S6: Implement individualized cardiopulmonary endurance level assessment, divide the predicted maximum oxygen uptake value into 5 levels according to the percentile after age and gender correction, and generate a visual assessment report, including a heatmap of the contribution of exercise performance and a radar map of brain region weight distribution.

2. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: In step S1, the initial speed of the progressive speed increase shuttle walk test is set to 1.0 m / s, increasing by 0.2 m / s in each stage, with a single stage lasting 60 seconds, until the subject can no longer maintain the prescribed rhythm or the subjective fatigue score reaches 18 points or more; the standardized track length is 20 meters, with pressure sensing pads at both ends to accurately capture the moment of foot contact; the cadence sequence is obtained by sampling through a wearable inertial measurement unit at a sampling frequency of 100 Hz; and the dynamic heart rate response curve is continuously recorded by a chest strap heart rate monitor with a time resolution of 1 second.

3. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: In step S2, the spatial resolution of the high-resolution T1-weighted image is 1 mm x 1 mm x 1 mm. The parameters of the diffusion tensor imaging data are set to a b value of 1000 seconds per square millimeter, a gradient direction number of 64, an intravoxel mean number of 2, and the image reconstruction uses the Siemens three-dimensional fast gradient echo sequence. The anterior cingulate cortex thickness measurement range is limited to Brodmann zones 24 and 32. The mean intercortical distance is calculated using the surface substrate modeling method.

4. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: In step S3, the normalization process uses Z-score transformation, and the joint encoding process introduces cross terms to construct interactive features. The cross terms to construct interactive features include the product of the anterior cingulate cortex thickness and the slope of the heart rate rise in the later stage of exercise.

5. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: In step S4, the oxygen uptake prediction nonlinear mapping model is trained using an adaptive moment estimation optimization algorithm. The loss function is defined as the square root of the mean square error between the predicted and measured values. The training set accounts for 80% of the total sample size, the validation set accounts for 10%, and the test set accounts for 10%. The input features of the deep feedforward neural network also include body mass index, resting heart rate, and systolic blood pressure. The upper limit of the number of model training iterations is set to 500 rounds. The early stop mechanism is triggered when the validation set loss does not decrease for 10 consecutive rounds. The initial learning rate is set to 0.001, and it decays to 0.9 times the original value every 100 rounds.

6. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: The five levels in step S5 are as follows: Level 1 corresponds to below the 20th percentile, Level 2 to the 20th to 40th percentile, Level 3 to the 40th to 60th percentile, Level 4 to the 60th to 80th percentile, and Level 5 to above the 80th percentile. The age correction uses a piecewise linear regression model, with the male population decreasing by 0.55% per year in the 20-50 age range and increasing to 0.75% for those over 50. The corresponding decrease rates for women are 0.45% and 0.65%, respectively. The gender correction factor is set so that the male baseline value is 8.3% higher than that for women.

7. The method for testing oxygen uptake and cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: The test also included repeating the procedure after subjects underwent 12 weeks of aerobic intervention training, comparing the predicted change in oxygen uptake ΔVO2max before and after the intervention. If ΔVO2max was greater than or equal to 1.5 ml / kg / min, it was considered a significant improvement. At the same time, the correlation between the degree of improvement in white matter integrity and ΔVO2max was analyzed. A strong correlation was considered to exist when the correlation coefficient r was greater than 0.

6.

8. The method for testing cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: The multimodal fusion feature vector introduces a dynamic coupling index, defined as the ratio of the heart rate rise slope in the later stage of exercise to the activation potential of the anterior cingulate cortex, wherein the activation potential of the anterior cingulate cortex is jointly determined by the gray matter density and local consistency score of the region.

9. The method for testing cardiopulmonary endurance based on load shuttle walk and human brain structure according to claim 1, characterized in that: The oxygen uptake prediction nonlinear mapping model is deployed on edge computing terminal devices with an inference latency of less than 200 milliseconds. It supports offline operation and uses the national cryptographic SM4 algorithm for data encryption to ensure that the security and privacy protection of the subjects' biomedical information meet the requirements of the national level 2 information security protection.

Images