Predicting psychosis in at-risk patients using abnormal neural oscillations and synchrony in conjunctions with machine learning algorithms

In the last 20 years, there has been a marked increase in interest in the early detection and treatment of psychosis. Despite the various potential “prodromes” that have been identified and have helped to increase the accuracy in the detection of persons at-risk of developing psychoses, it is still not possible to predict the transition to psychosis with sufficient accuracy. Although some electroencephalography (EEG) studies, based on basic power-spectral and event-related potential analyses, have been conducted in the field of early detection, neural oscillations and their phase-synchronization across brain areas have been ignored.
The present dissertation covers three different studies which, together, demonstrate that neural oscillations are disturbed in emerging psychosis. The first paper shows that at-risk patients with later transition to psychosis are characterized by abnormal localized brain activity and that inter-cortical areas of the brain are poorly synchronized. The second study shows that machine learning algorithms can detect patterns of abnormal brain activity predictive of later transitions to psychosis with promising accuracy. The third study reveals, in a cross-sectional manner, that patients who already had a first episode of psychosis at inclusion, already demonstrated the same abnormal patterns of brain activity revealed in at-risk patients with later transition to psychosis.