Psychosis is a frequent and disabling illness, occurring in adolescence and young adulthood, whose socio-functional impact could be prevented by early intervention. Although it's possible to clinically screen subjects at risk, it's very difficult to identify those who will progress to psychosis (25% at three years). The emergence of psychosis results from progressive interactions between genetic vulnerability and environmental stress factors, which can disrupt the neurodevelopmental trajectory. The search for prognostic biomarkers is therefore crucial to improving prevention and early care. In this context, structural neuroimaging is a promising tool, as variations in brain development and structural brain abnormalities can reveal underlying pathophysiology.
In this work, the authors used voxel-based morphometry (VBM), a measure of local gray matter volume for each voxel in the brain, from anatomical MRI data of 2024 healthy subjects and 82 subjects at Ultra-High Risk (UHR) of psychosis, 27 of whom developed psychosis after one year. In UHR subjects, they were able to automatically identify regions predictive of conversion to psychosis using a supervised learning model (Enet-TV-LR) that provided them with an interpretable map of distinct predictive regions. Using both the whole brain and each predictive region separately, a brain age predictor was then constructed and validated in 1605 controls, externally tested in 419 controls from an independent cohort, and applied to UHR subjects. Brain age gaps were calculated as the difference between chronological age and predicted age, providing an approximation of global and regional brain maturation. With an area under the curve of 80%, the prediction of the onset of psychosis was very good, thanks in particular to volumetric increases in the ventromedial prefrontal cortex and decreases in the left precentral gyrus and right orbitofrontal cortex, regions predicted to show delayed and accelerated maturation patterns, respectively.
Study design (top) and main brain areas affected in subjects who developed psychosis (bottom). © E.Duchesnay/CEA
By providing the first interpretable models of psychosis predictors based on epigenetics and neuroanatomy in subjects at very high risk of psychotic transition, this work supports the hypothesis of an underlying neurodevelopmental pathophysiology, marked by complex patterns of asynchronous maturation.
Contacts : Édouard Duchesnay (edouard.duchesnay@cea.fr) et Anton Iftimovici (anton.iftimovici@inserm.fr)
Area under the curve (AUC) is a measure of the performance of a classification model. It is commonly used in machine learning and data analysis to assess the accuracy of models in predicting binary events.