Interpreting Deep Learning Models for Multimodal Neuroimaging

Accurately analyzing both structural and functional brain data from multimodal neuroimaging is a challenge for deep learning methods. Recent progress in explainable AI (XAI) has helped to gain insight into structural relationships across brain regions and complex dynamics of cognitive states, both in healthy and diseased. In this brief note, we will touch upon selected recent directions of our research where machine learning techniques help to analyze brain measurements from EEG, fNIRS, sMRI and fMRI. We owe these steps that we are summarizing mainly to activities of members of the BBCI team and their collaborators. Clearly, unavoidably and intentionally this abstract will have a high overlap to prior own contributions as it reports about and discusses these novel ideas and directions.