EEG-based Assessment of Human Endurance: Association between Endurance and Brain-wave Activity

Endurance is a fundamental human potential that researchers have focused on widely. Studies on athletes have shown that human endurance is linked to brain signals. To improve modern assessment and intervention systems for human endurance, it is necessary to identify certain cortical brain-signal activity reflecting endurance potential. This study proposes a deep learning-based model that automatically extracts features from electroencephalograms (EEGs) to predict human endurance performance and identify the frequency ranges and brain regions reflecting key endurance characteristics in running athletes during task and resting states. The EEGs of 50 individuals (27 running athletes and 23 non-athletes) were acquired during resting and bicycle-riding states. The signals were preprocessed, transformed into spectrograms, and fed into a convolutional neural network (CNN) for feature extraction and inter-individual classification. The results showed that when applying this model to the data of three different states, one riding and two resting, the average accuracies were 82.0%, 82.0%, and 74.7% respectively, outperforming the baseline methods in all cases. Furthermore, the features were visualized by class activation mapping (CAM) to identify the frequencies and brain regions associated with key endurance characteristics. Accuracies were improved by selecting the featured regions, but were significantly compromised if masked, thereby confirming the reliability of the extracted features. It is indicated that high beta (20-30 Hz) and gamma (31-40 Hz) activities in the parietal and occipital regions were most related to endurance characteristics during tasks, while gamma activities in the premotor and primary motor cortex differed between athletes and non-athletes at rest. The results were further confirmed using Wilcoxon signed-rank tests.