Classifying Treated vs. Untreated MDD Adolescents from Anatomical Connectivity using Nonlinear SVM

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Psychiatry

Background:

• Identifying treatment responders in adolescent Major Depressive Disorder (MDD) is crucial for personalized medicine.
• Current diagnostic methods lack precision in predicting treatment outcomes for adolescent MDD.

Purpose of the Study:

• To utilize machine learning to identify neuroanatomical features differentiating treated from untreated adolescent MDD patients.
• To enhance the accuracy of classifying MDD patients based on brain connectivity and network topology.

Main Methods:

• Employed machine learning classifiers on diffusion tensor imaging (DTI) derived anatomical connectivity and network topological measurements.
• Utilized p-value and minimum redundancy maximum relevance (mRMR) for feature selection.
• Performed classification using a leave-one-out cross-validation on 52 subjects (37 treated, 15 untreated).

Main Results:

• Achieved 73% accuracy, 100% specificity, and 100% precision in distinguishing treated from untreated adolescent MDD patients.
• Identified key distinguishing features including mean diffusivity (MD) and track-count (TR) network properties of the right hippocampus, and radial diffusivity (RD) network participation coefficient of the left middle temporal gyrus.
• Highlighted the importance of axial diffusivity (AD) and apparent diffusion coefficient (ADC) connectivity between specific brain regions.

Conclusions:

• Machine learning effectively identifies neuroanatomical biomarkers for treatment response in adolescent MDD.
• These findings provide a foundation for developing objective tools to guide treatment selection in adolescent depression.
• Further research can refine these features for clinical application in predicting treatment outcomes.