Can a single brain region predict a disorder?

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Predicting neurological and psychiatric disorders from brain fMRI data is challenging.
• Voxel-based feature methods show instability for group classification tasks.

Purpose of the Study:

• To develop and validate a robust method for predicting diverse brain disorders in unseen subjects using fMRI.
• To improve upon state-of-the-art prediction techniques for conditions such as Cocaine Use Disorder, Schizophrenia, and Alzheimer's disease.

Main Methods:

• Utilized a region-based feature selection approach with a majority vote classifier.
• Employed a leave-one-subject-out cross-validation strategy for robust generalization assessment.
• Integrated multiple experimental conditions and diverse fMRI acquisition parameters.

Main Results:

• Achieved high prediction accuracies: 89.3-90.9% for Cocaine Use, 96.4% for Schizophrenia, and 81.5% for Alzheimer's disease.
• Demonstrated method's effectiveness across various experimental designs, magnetic field strengths, and acquisition speeds.
• Showcased the method's ability to generate a meaningful, low-dimensional representation that preserves discriminative information.

Conclusions:

• The proposed region-based method offers a stable and effective approach for multi-subject disorder prediction from fMRI.
• This technique generalizes well across diverse disorders, experimental paradigms, and data acquisition settings.
• The findings suggest a promising direction for developing reliable neuroimaging-based diagnostic tools.