A cross-subject MDD detection approach based on multiscale nonlinear analysis in resting state EEG

Exploring multi-scale nonlinear patterns from different frequency bands in resting state electroencephalogram (EEG) signals is significant for major depressive disorder (MDD) detection. The study aims to investigate potential EEG biomarkers and realize cross-subject detection of MDD. This study used multiscale LZC (MLZC) to extract nonlinear features of resting state EEG. Brain topography analysis was used to investigate the difference between MDD and healthy controls (HC) among different scales. A multiscale feature fusion method was proposed to realize the cross-subject detection of MDD. Two public datasets (MPHC and MODMA) and three classifiers were used to validate the performance of the proposed method. Compared with other scales, the difference between the two groups was larger in the high frequency scale, as demonstrated by the higher complexity of brain activity in the HC group than in the MDD group. For the classification, the high frequency scale LZC had the best classification results, with accuracies of 68.75%, 82.61%, and 73.44% in MODMA, MPHC, and fused datasets. Through the multiscale feature fusion analysis, it is found that it retains a large amount of high-frequency channel information for the three datasets, highlighting the importance of high frequency features. By combining the multiscale nonlinear feature fusion, it achieves the best classification results on the three dataset experiments, with accuracies of 72.42%, 84.81%, and 76.13%, respectively. The high frequency scale LZC is more effective for MDD diagnosis in resting state EEG. The cross-subject MDD patients detection accuracy can be promoted by multiscale nonlinear feature fusion.