EEG to fMRI Synthesis for Medical Decision Support: A Case Study on Schizophrenia Diagnosis

Electroencephalography (EEG) measures the neuronal activity at the scalp, while functional magnetic resonance imaging (fMRI) provides a sub-cortical view of blood supply in the human brain. Although fMRI is known for providing rich spatial information, it is expensive and of restricted use. EEG to fMRI synthesis is a cross modal research area that bridges the gap between the two and has recently received attention. Although these studies promise lower healthcare costs and ambulatory assessments, their utility in diagnostic settings is still largely untapped. Using simultaneous EEG and fMRI recordings, this study combines a state-of-the-art synthesis model with a modified contrastive loss, and subsequent prediction layering, to unprecedentedly assess its predictive power in schizophrenia diagnosis. In addition, we perform an exhaustive search for the (synthesized) hemodynamic brain patterns able to discriminate schizophrenia. Schizophrenia diagnosis using synthesized hemodynamics yield an area under the ROC curve of 0.77, confirming the validity of the undertaken neuroimaging synthesis. Experiments further revealed schizophrenia-related patterns in frontal, left temporal and cerebellum regions of the brain. Altogether, our results suggest that a synthesized fMRI view is able to discriminate this pathology, and it contains discriminative patterns of brain activity in accordance with related work on schizophrenia.