Application of machine learning to structural connectome to predict symptom reduction in depressed adolescents with cognitive behavioral therapy (CBT)

PURPOSE: Adolescent major depressive disorder (MDD) is a highly prevalent, incapacitating and costly illness. Many depressed teens do not improve with cognitive behavioral therapy (CBT), a first-line treatment for adolescent MDD, and face devastating consequences of increased risk of suicide and many negative health outcomes. “Who will improve with CBT?” is a crucial question that remains unanswered, and treatment planning for adolescent depression remains biologically unguided. The purpose of this study was to utilize machine learning applied to patients' brain imaging data in order to help predict depressive symptom reduction with CBT. METHODS: We applied supervised machine learning to diffusion MRI-based structural connectome data in order to predict symptom reduction in 30 depressed adolescents after three months of CBT. A set of 21 attributes was chosen, including the baseline depression score, age, gender, two global network properties, and node strengths of brain regions previously implicated in depression. The practical and robust J48 pruned tree classifier was utilized with a 10-fold cross-validation. RESULTS: The classification resulted in an 83% accuracy of predicting depressive symptom reduction. The resulting tree of size seven with only three attributes highlights the role of the right thalamus in predicting depressive symptom reduction with CBT. Additional analysis showed a significant negative correlation between the change in the depressive symptoms and the node strength of the right thalamus. CONCLUSIONS: Our results demonstrate that a machine learning algorithm that exclusively uses structural connectome data and the baseline depression score can predict with a high accuracy depressive symptom reduction in adolescent MDD with CBT. This knowledge can help improve treatment planning for adolescent depression.