Machine Learning of Schizophrenia Detection with Structural and Functional Neuroimaging

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Machine Learning

Background:

• Schizophrenia (SZ) affects 1% of the population, posing diagnostic challenges.
• Multimodal brain imaging (fMRI, sMRI) and machine learning (ML) show promise for improving SZ diagnosis.
• Conventional ML methods struggle to effectively integrate multimodal brain data.

Purpose of the Study:

• To introduce a novel multimodal imaging and multilevel characterization with multiclassifier (M3) method for SZ classification.
• To investigate the influence of global signal regression (GSR) on ML-based SZ diagnosis.
• To assess the effectiveness of M3 in distinguishing schizophrenia patients from healthy controls.

Main Methods:

• Utilized functional MRI (fMRI) and structural MRI (sMRI) data.
• Implemented the multimodal imaging and multilevel characterization with multiclassifier (M3) method.
• Compared classification performance with and without global signal regression (GSR) and using different brain parcellation schemes (e.g., Brainnetome 246 atlas).

Main Results:

• The M3 method achieved 83.49% classification accuracy, 68.69% sensitivity, 93.75% specificity, and an AUC of 0.8491 without GSR using the Brainnetome 246 atlas.
• Classification performance was robust across different processing methods.
• Models without GSR demonstrated higher accuracy and specificity compared to models with GSR.

Conclusions:

• The M3 method is an effective tool for differentiating schizophrenia patients from healthy controls.
• The study suggests that global signal regression may remove important neuronal information, potentially hindering SZ detection.
• The M3 method can identify discriminative brain regions, aiding in the exploration of neural mechanisms underlying schizophrenia.