Cross-Sampling Rate Transfer Learning for Enhanced Raw EEG Deep Learning Classifier Performance in Major Depressive Disorder Diagnosis

Transfer learning offers a route for developing robust deep learning models on small raw electroencephalography (EEG) datasets. Nevertheless, the utility of transferring representations from large datasets with lower sampling rates to smaller datasets with higher sampling rates is relatively unexplored. In this study, we transfer representations learned by a convolutional neural network on a large, publicly available sleep dataset (100 Hertz sampling rate) to a major depressive disorder (MDD) diagnosis task (200 Hertz sampling rate) and examine their effects upon the MDD models via spectral and spatial explainability analyses. Importantly, we find that the early convolutional layers have representations that are generalizable across tasks. Moreover, our approach significantly increases mean model accuracy from 82.33% to 86.99%, increases use of lower frequencies, (θ-band), and increases robustness to channel loss. We expect this analysis to provide useful guidance and enable more widespread use of transfer learning in EEG deep learning studies.