Random Forest Classification of Depression Status Based On Subcortical Brain Morphometry Following Electroconvulsive...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Machine Learning

Background:

• Central nervous system disorders often present with brain abnormalities detectable via MRI.
• Advanced pattern detection and classification algorithms aid in diagnosing brain disorders and predicting treatment efficacy.

Purpose of the Study:

• To classify depressed subjects versus controls using subcortical structure shape information.
• To classify patients before and after electroconvulsive therapy (ECT).
• To compare the performance of shape-based features against traditional volumetric measures.

Main Methods:

• Utilized a novel feature-selection method based on regularized random forests.
• Applied shape descriptors derived from subcortical structures in magnetic resonance imaging (MRI) data.
• Compared classification accuracy of high-dimensional shape features with volumetric data.

Main Results:

• Shape-based models demonstrated superior classification performance compared to volumetric predictors in certain cases.
• The method successfully differentiated between depressed subjects and controls.
• The approach effectively classified patients' response to electroconvulsive therapy.

Conclusions:

• High-dimensional shape features derived from subcortical structures are valuable for classifying brain disorders like depression.
• Shape-based analysis offers a promising automated alternative for diagnosis and predicting treatment response, outperforming traditional volumetric methods.
• This approach has potential applications in clinical settings for improved patient management.