“Can we catch it before we lose it?”:
Developing an Integrated Digital-Clinical Phenotype of Risk for Psychosis

Yonatan (Yoni) Stern#, and Danny Koren - Department of Psychology
Hagit Hel-Or - Department of Computer Science
Roy Salomon - Gonda Brain Research Center, Bar-Ilan University

Artificial Intelligence
computer vision
Image processing


PhD Grant 2021

Thirty years ago, a Natur editorial described schizophrenia as “arguably the worst disease affecting mankind.” Unfortunately, the situation has not improved significantly. Schizophrenia still constitutes a grievous unsolved medical, social, and economic problem. It is widely acknowledged that a pre-onset phase typically precedes full-blown schizophrenia3,4, during which general psychiatric and sub-threshold psychotic symptoms are manifest, often persisting unidentified for over a year. As with other medical diseases (e.g., cancer, cardiac disease), early intervention in schizophrenia has been hailed as a promising avenue to improve its harsh prognosis.  Nonetheless, only approximately a third of the individuals identified as at-risk proceed to develop schizophrenia. Hence, more reliable, and precise methods of identification and risk assessment are necessary. The development of such tools requires an integrated understanding of schizophrenia’s clinical phenotype.  

A growing body of clinical and empirical literature suggests that a disturbed sense of the “Self” is a core deficit in schizophrenia, arising from aberrant processing of the Sense of Agency (SoA), the feeling of control over one’s actions, and the ability to accurately attribute them to oneself. Utilizing virtual reality (VR) technology we have developed state-of-the-art experimental paradigms that enable us to manipulate in an ecologically valid and well-controlled manner one’s experience of the body and SoA. We have developed VR environments in which we introduce sensorimotor conflicts between the participant’s actions and perceived consequences. Work from these paradigms allow for: (i) bridging the gap between the real-world experience of psychotic symptoms and how they are assessed in the lab, and (ii) providing well-defined quantitative measures that may be related to participants’ clinical states. Work published from these paradigms has found that: (1) VR-based task performance is significantly correlated to self-reported psychotic symptoms in the general population. (2) Sensitivity to such changes is drastically different between controls and chronic schizophrenia patients, and a classifier can distinguish between chronic schizophrenia patients and controls at ~90% accuracy. (3) Furthermore, these experimental paradigms have provided novel insight to the temporal dynamics and computational mechanisms leading to psychotic experiences. 

The current research uses the rich multi-faceted data collected in a VR environment mimicking psychotic symptoms to develop a digital-clinical phenotype of psychosis that integrates physiological, kinematic, and cognitive data. Such a clinical phenotype will enable improved risk assessment and early detection tools that rely on subtle changes that the person may not be aware of. We hope that improving early detection of schizophrenia will allow ius to catch it before we lose it.