2023, Vol. 27 ›› Issue (8): 1224-1231

Constructing the prediction model of maximal oxygen uptake by back-propagation neural network based on the cardiorespiratory optimal point

Wu Dongzhe, Gao Xiaolin, Li Chuangtao, Wang Hao   

Abstract: BACKGROUND: Maximal oxygen uptake is considered as the “gold standard” for evaluating aerobic exercise capacity and cardiopulmonary health. However, the measurement of maximal oxygen uptake requires a strong exercise load, and there are some limitations such as low reproducibility of the index and subjective effect of the test participants.
OBJECTIVE: To construct the prediction model of maximal oxygen uptake using the new sub-maximal exercise evaluation index - cardiorespiratory optimal point - through the back-propagation neural network.
METHODS: The trial protocol was approved by the Ethics Committee of the China Institute of Sports Science. Eighty healthy college students (40 males and 40 females) were randomly recruited. They were fully informed of the trial process and purpose and voluntarily signed informed consent to cooperate with the whole trial process. The participants underwent an incremental load cardiopulmonary exercise test to identify the maximal oxygen uptake, cardiorespiratory optimal point, and other related indicators for correlation analysis to obtain statistically significant indicators. Then, the prediction model of maximal oxygen uptake was built.
RESULTS AND CONCLUSION: There were significant correlations between maximal oxygen uptake and cardiorespiratory optimal point, body mass index, sex, oxygen uptake and power corresponding to the cardiorespiratory optimal point (P < 0.01). The prediction model of maximal oxygen uptake with classical three-layer topology was established using the back-propagation neural network, including 5 input layers, 10 hidden layers and 1 output layer. The mean absolute and relative errors between the predicted and measured values of the model were 0.227 L/min and 12%, respectively. This indicated that the back propagation neural network model built based on the cardiorespiratory optimal point could accurately and effectively predict the maximal oxygen uptake. There was no significant difference between the maximal oxygen uptake predicted value of the back propagation neural network model and that of the multiple linear regression model (P > 0.05). However, the prediction accuracy of the back propagation neural network model constructed according to the cardiorespiratory optimal point was better than that of the multiple linear regression model.
Key words: cardiorespiratory optimal point, maximal oxygen uptake, back-propagation neural network, cardiopulmonary exercise test