Framework for Accurate Classification of Self-Reported Stress From Multisession Functional MRI Data of Veterans With
Rahul Goel1, Teresa Tse1, Lia J Smith2,3
1Department of Neuroscience, Baylor College of Medicine, Houston, TX, USA.
Chronic Stress (Thousand Oaks, Calif.)
|October 2, 2023
Summary
Machine learning accurately predicts combat Veterans
Area of Science:
- Neuroscience
- Psychiatry
- Computer Science
Background:
- Posttraumatic stress disorder (PTSD) significantly impacts combat Veterans.
- Existing treatments for PTSD need improvement in effectiveness.
- Functional magnetic resonance imaging (fMRI) neurofeedback and virtual reality (VR) show promise for PTSD treatment.
Purpose of the Study:
- To develop a method for classifying internal brain stress levels in Veterans with PTSD during VR exposure.
- To facilitate real-time fMRI neurofeedback for PTSD treatment.
Main Methods:
- Collected fMRI data from 8 male combat Veterans with PTSD symptoms during 2 sessions.
- Veterans self-reported stress levels every 15 seconds.
- Applied machine learning (ML) algorithms to preprocessed fMRI data to predict stress responses.
Main Results:
- Accurately classified 8 classes of self-reported stress responses.
- Achieved a mean root mean square error of 0.6 (± 0.1) across Veterans using the best ML approach.
Conclusions:
- ML algorithms can predict internal brain states from whole-brain cortical fMRI data.
- The developed framework enables individualized, real-time fMRI neurofeedback for PTSD exposure therapy.
Keywords:
algorithm design and analysisbehavioral sciencesfunctional magnetic resonance imagingmachine learning algorithmsmental disordersposttraumatic stress disorder (PTSD)psychiatryveterans
