Interpretable artificial intelligence for detection of psychiatric disorders based on multi-modal neuroimaging data

Faculty-development competitive research grants program for 2020-2022 (batch 2)

The medical field, in particular, must provide a reliable interpretation for their decision making since groundless decisions can lead to irreparable results.
In this study, we propose highly accurate, subject-independent, and explainable AI systems for the diagnosis of psychiatric disorders. This project includes three purposes: 1) An optimization algorithm to extract subject-invariant features in spatial, frequency, and temporal domains from the large size and multi-modal EEG/NIRS dataset, 2) Advanced interpretable interface based on the layer-wise decomposition technique where one can visually explain the output of the prediction, 3) Finally, we will collect a large neuronal dataset based on multi-modal setting (EEG, NIRS, PPG, diagnostic reports, etc.), that will greatly contribute to the related research field.