Asynchronous neural maturation predicts transition to psychosis.

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Neuroimaging and machine learning are used to predict psychosis onset based on brain structure.
• Interpreting these predictors in relation to brain structure remains a challenge.
• Understanding brain maturation is crucial for identifying psychosis risk.

Purpose of the Study:

• To develop an interpretable machine learning model to predict psychosis onset.
• To identify specific brain regions and their maturation patterns predictive of psychosis.
• To investigate the role of neuroanatomical age in psychosis prediction.

Main Methods:

• Used voxel-based morphometry on healthy and at-risk mental state (UHR) cohorts.
• Applied Elastic-Net-Total-Variation (Enet-TV) with cross-validation for psychosis prediction.
• Developed and validated a brain age predictor to assess regional maturation and brain age gaps.

Main Results:

• Achieved high performance in psychosis prediction (80% AUC, 69% accuracy).
• Identified predictive regions including ventromedial prefrontal cortex (volumetric increase) and left precentral gyrus/right orbitofrontal cortex (volumetric decrease).
• Observed delayed maturation in predictive regions for psychosis conversion and accelerated maturation in others.

Conclusions:

• Interpretable machine learning combined with brain age modeling reveals asynchronous maturation patterns.
• Asynchronous inter-regional brain maturation is a key component of the psychosis predictive signature.
• This approach enhances the understanding of neurobiological underpinnings of psychosis development.