Dissecting Psychiatric Heterogeneity and Comorbidity with Core Region-Based Machine Learning.

Area of Science:

• Neuroscience
• Psychiatry
• Artificial Intelligence

Background:

• Machine learning (ML) is increasingly used with neuroimaging data for psychiatric disorder diagnosis and prognosis.
• Current ML applications in psychiatric neuroimaging require refinement for better clinical utility.

Purpose of the Study:

• To review current practices for evaluating ML in obsessive-compulsive and related disorders.
• To propose a novel ML strategy using a core set of brain regions for enhanced diagnostic and prognostic accuracy.
• To provide a roadmap for developing neuroimaging-based biomarkers for psychiatric disorders.

Main Methods:

• A novel strategy is presented for constructing ML models using a defined set of 'core regions' identified from neuroimaging data.
• Both hypothesis-driven and data-driven approaches are discussed for identifying these core brain regions.
• The strategy focuses on brain areas central to the psychopathology of psychiatric disorders.

Main Results:

• A core set of co-altered brain regions can efficiently build predictive ML models.
• These models can identify distinct symptom dimensions or clusters in individual patients.
• The proposed strategy aims to improve performance, interpretability, and generalizability of ML models.

Conclusions:

• A core set-based ML strategy offers a promising approach for developing neuroimaging biomarkers.
• This method can accelerate the discovery of reliable markers for psychiatric disorder diagnosis and prognosis.
• The strategy is broadly applicable across various psychiatric disorders.