Background and purpose: Posttraumatic stress disorder (PTSD) is a heterogeneous condition associated with a range of brain imaging abnormalities. Early life stress (ELS) contributes to this heterogeneity, but we do not know how a history of ELS influences traditionally defined brain signatures of PTSD. Here, we used a novel machine learning method - evolving partitions to improve classification (EPIC) - to identify shared and unique structural neuroimaging markers of ELS and PTSD in 97 combat-exposed military veterans.
Methods: We used EPIC with repeated cross-validation (CV) to determine how combinations of cortical thickness, surface area, and subcortical brain volumes could contribute to classification of PTSD (n = 40) versus controls (n = 57), and classification of ELS within the PTSD (ELS+ n = 16; ELS- n = 24) and control groups (ELS+ n = 16; ELS- n = 41). Additional inputs included intracranial volume, age, sex, adult trauma, and depression.
Results: On average, EPIC classified PTSD with 69% accuracy (SD = 5%), and ELS with 64% accuracy in the PTSD group (SD = 10%), and 62% accuracy in controls (SD = 6%). EPIC selected unique sets of individual features that classified each group with 75-85% accuracy in post hoc analyses; combinations of regions marginally improved classification from the individual atlas-defined brain regions. Across analyses, surface area in the right posterior cingulate was the only variable that was repeatedly selected as an important feature for classification of PTSD and ELS.
Conclusions: EPIC revealed unique patterns of features that distinguished PTSD and ELS in this sample of combat-exposed military veterans, which may represent distinct biotypes of stress-related neuropathology.
Keywords: PTSD; early life stress; machine learning; neuroimaging.
© 2019 by the American Society of Neuroimaging.