Machine Learning Models Only Moderately Accurate in Diagnosing Depression from Brain Scans Alone
-
Researchers used machine learning on brain MRI data to try to classify patients with major depressive disorder (MDD) vs healthy controls.
-
The study included over 5,000 people across 30 sites/datasets. Linear and nonlinear machine learning models were tested.
-
The models achieved only about 50-60% balanced accuracy, just slightly better than chance, in classifying MDD.
-
Model performance was mainly driven by cortical thickness measures, rather than surface area.
-
More sophisticated machine learning approaches may be needed to better distinguish MDD based on structural brain measures alone.