Abstract
Post-traumatic Stress Disorder (PTSD) affects many US military veterans, but current diagnoses rely on clinical interviews and self-reports that often show inconsistencies. This study developed an explainable machine learning approach to predict PTSD severity using brain imaging data from 174 US Army soldiers with varying PTSD symptoms. Resting-state functional MRI data were analyzed to identify neural signatures most relevant to PTSD. Six machine learning algorithms were tested to predict four severity categories. The Support Vector Machine model achieved exceptional performance with 97.1% accuracy, 97.5% precision, 97.1% recall, and 97.2% F1 score. Feature analysis revealed distinct brain connectivity signatures for each category, with specific neural connections identified for different levels. Explainable AI analysis revealed critical threshold effects and identified left insula-left middle temporal connectivity as the strongest predictor. This framework demonstrates that objective brain-based assessment can predict PTSD severity with high accuracy while providing clinically meaningful insights for improved treatment planning.
Recommended Citation
Naeimaeimousavi, Seyed Amirhossein; Gupta, Ashish; Joghataee, Mohammad; and Deshpande, Gopikrishna, "Predicting PTSD Severity in Veterans: An Explainable FMRI-Based Machine Learning Approach" (2025). Proceedings of the 2025 Pre-ICIS SIGDSA Symposium. 12.
https://aisel.aisnet.org/sigdsa2025/12