Disease state prediction from resting state functional connectivity.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Multivoxel pattern analysis (MVPA) is increasingly used for brain state prediction and real-time functional MRI (fMRI).
• Support vector classification (SVC) is a popular MVPA technique known for its accuracy and noise resilience.

Purpose of the Study:

• To apply SVC to functional connectivity patterns for distinguishing between patients with depression and healthy volunteers.
• To develop and evaluate novel feature selection algorithms incorporating reliability information for improved diagnostic accuracy.

Main Methods:

• Support vector classification (SVC) was employed to analyze functional connectivity data derived from fMRI.
• Two new feature selection algorithms, one filter and one wrapper method, were developed, integrating reliability metrics.
• These novel methods were compared against two established feature selection techniques.

Main Results:

• A support vector classifier trained with the proposed methods reliably differentiated clinically depressed patients from healthy controls.
• The reliability-based feature selection methods demonstrated superior performance compared to previously used approaches.
• The developed framework successfully identified distinct functional connectivity patterns associated with major depression.

Conclusions:

• SVC is a powerful tool for analyzing functional connectivity data in psychiatric research.
• Incorporating reliability into feature selection significantly enhances the accuracy of diagnostic classifiers for depression.
• The proposed framework is adaptable for classifying other neurological and psychiatric disease states using fMRI data.