Whole-Brain Patterns Related to Emotionally Valenced Images Predict PTSD Symptoms

1972 

Abstract Submission 

Vansh Bansal¹, Katherine McCurry², Jonathan Lisinski³, John Wang⁴, Dong-Youl Kim⁵, Kelsey Winkeler⁵, Stephen LaConte³, Pearl Chiu⁴, Brooks Casas⁴

¹Fralin Biomedical Research Institute, Virginia Tech, Roanoke, VA, ²Michigan Medicine, University of Michigan, Ann Arbor, MI, ³Fralin Biomedical Research Institute at VTC, Virginia Tech, Roanoke, VA, ⁴Fralin Biomedical Research Institute at Virginia Tech Carilion, Roanoke, VA, ⁵Fralin BIomedical Research Institute, Virginia Tech, Roanoke, VA

Vansh Bansal  
Fralin Biomedical Research Institute, Virginia Tech
Roanoke, VA

Katherine McCurry, PhD  
Michigan Medicine, University of Michigan
Ann Arbor, MI

Jonathan Lisinski  
Fralin Biomedical Research Institute at VTC, Virginia Tech
Roanoke, VA

John Wang  
Fralin Biomedical Research Institute at Virginia Tech Carilion
Roanoke, VA

Dong-Youl Kim  
Fralin BIomedical Research Institute, Virginia Tech
Roanoke, VA

Kelsey Winkeler  
Fralin BIomedical Research Institute, Virginia Tech
Roanoke, VA

Stephen LaConte  
Fralin Biomedical Research Institute at VTC, Virginia Tech
Roanoke, VA

Pearl Chiu  
Fralin Biomedical Research Institute at Virginia Tech Carilion
Roanoke, VA

Brooks Casas  
Fralin Biomedical Research Institute at Virginia Tech Carilion
Roanoke, VA

Prior research on biomarkers of posttraumatic stress disorder (PTSD) has leveraged functional magnetic resonance imaging (fMRI) and emotion processing paradigms to examine brain activity. This work has found overall affective processing to be a deficit in PTSD [1]. Additionally, neural habituation to negatively valenced images compared to neutral images has a strong link to hyperarousal and re-experiencing symptom severity, with opponent effects found in which decreased habituation was associated with hyperarousal symptoms and increased habituation was associated with re-experiencing symptoms [2]. Given this prior work, we implemented machine learning methods with both habituation-related neural data and average neural signal of emotion processing to identify multivariate activation patterns that predict hyperarousal and re-experiencing symptoms.

Participants: This study included 132 military veterans (23 female; mean age: 36.1 years (s.d. 10.8) who were deployed during post-9/11 conflicts (Operation Enduring Freedom, Operation Iraqi Freedom, and Operation New Dawn) and were recruited for this study from rural and urban regions of southwest Virginia and Houston, Texas.

Clinical Assessment: Diagnosis and symptoms of PTSD were assessed using the Clinician Administered PTSD Scale (CAPS) prior to scanning. Given our prior work [2], the present analyses focused on hyperarousal and re-experiencing symptoms.

Experimental Paradigm: Participants were shown eight blocks each of positive, negative, and neutral-valenced images (8-10 unique images per block) in pseudo-randomized order from the International Affective Picture System [3] while undergoing fMRI.

fMRI Analysis: All participants underwent fMRI in a 3T Siemens Tim Trio with standard preprocessing methods [2]. At the individual level, valence-specific average responses were modeled as boxcar functions corresponding to each block's duration, and valence-specific patterns of habituation and/or sensitization of hemodynamic responses were modeled as a linear parametric modulation corresponding to block order over time [2]. General linear models (GLMs) were used to derive individual-level beta maps.

Machine Learning: For every participant, average and habituation beta maps for each valence (positive, negative, neutral) and each valence contrast were each used for a series of 12 support vector regressions (SVR). Beta maps from 100 veterans were used for each SVR, with the remaining 32 held for future out-of-sample testing. Nested cross-validation and grid search hyperparameter tuning was applied, and correlations of predicted hyperarousal score, predicted re-experiencing score, and predicted difference between hyperarousal and re-experiencing to actual scores were recorded to capture model performance. Performance was compared between SVRs predicting hyperarousal versus re-experiencing symptoms, as well as SVRs trained on average signal versus habituation.

The predicted difference between hyperarousal and re-experiencing scores from an SVR trained on average neural activation in negative image blocks correlated with true difference of hyperarousal and re-experiencing across folds (r = 0.45, s.d. = 0.40). For all beta map inputs, SVR models predicting re-experiencing symptoms were significantly more accurate (p=0.002) than those predicting hyperarousal symptoms. Furthermore, beta maps from average neural responses to valenced images were more predictive of all symptom scores than were beta images from habituation beta maps (p=0.001).

These findings support previous results showing that neural signals related to negatively valenced images predict hyperarousal symptoms, re-experiencing symptoms, and differentials in these symptom clusters in veterans. Future work will apply whole-brain multivariate weight patterns from this analysis to develop a real-time neurofeedback protocol targeting modulation of brain patterns associated with PTSD symptoms.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ²

Emotional Perception

Activation (eg. BOLD task-fMRI)

Multivariate Approaches ¹

ADULTS

Affective Disorders

FUNCTIONAL MRI

Machine Learning

Multivariate

Trauma