Degree of contribution (DoC) feature selection algorithm for structural brain MRI volumetric features in depression...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Accurate individual-level depression detection using structural magnetic resonance imaging (sMRI) is challenging.
• Brain volumetric changes are significant in depression biomarker research.
• Automated feature selection is crucial for improving diagnostic accuracy.

Purpose of the Study:

• Develop an automated algorithm for selecting brain sMRI volumetric features for depression detection.
• Introduce the degree of contribution (DoC) feature selection algorithm.
• Evaluate the performance of DoC in identifying major depressive disorder.

Main Methods:

• Developed the degree of contribution (DoC) feature selection algorithm using an ensemble approach.
• Implemented a four-stage process: feature ranking, subset generation, subset evaluation, and DoC analysis.
• Evaluated DoC on a dataset of 115 sMRI scans (88 healthy controls, 27 depressed subjects) using 44 volumetric features.

Main Results:

• The DoC algorithm determined feature importance scores (DoC scores) across varying accuracy thresholds (Acc_Thresh).
• DoC consistently outperformed four existing feature selection algorithms in both average and maximum classification scores.
• The developed DoC method demonstrated superior performance in feature selection for depression detection.

Conclusions:

• The DoC algorithm effectively generates smaller feature subsets with high classification accuracy for depression.
• Volumetric features from the left-brain region were identified as the most discriminant based on DoC scores.
• DoC shows promise for advancing automated depression detection using neuroimaging data.