Supervised machine learning in psychiatry: towards application in clinical practice

In recent years, the field of machine learning (often named with the more general term artificial intelligence) has literally exploded and its application has been proposed in basically all fields, including psychiatry and mental health. This has been motivated by the promise of using machine learning to develop new clinical tools that could help perform personalized predictions and recommendations, ultimately improving the results achievable in the psychiatric clinical practice that still faces only a limited success in the fight against mental diseases. However, despite this huge interest, there is still a substantial lack of tools in psychiatry that are based on machine learning algorithms. Massimiliano Grassi, in his Ph.D. thesis, investigates the challenges of translating machine learning algorithms into clinical practice and proposes innovative solutions to these challenges. The thesis presents the development and validation of new algorithms for the prediction of the onset of Alzheimer’s disease, the remission of obsessive-compulsive disorder, and the automatization of sleep staging in polysomnography, a method to diagnose sleep disorders. The results from these studies demonstrate that the use of machine learning in psychiatric clinical practice is not just a promise, and it is possible to develop machine learning algorithms that achieve clinically relevant performance even if based solely on information that can be easily accessible in the daily clinical routine.