Leveraging features from EEG and free-hand sketches for effective diagnosis of post traumatic stress disorder

Abstract

Due to war, violence, and other traumatic events, marginalized communities, e.g., refugees and internally displaced people are more vulnerable to develop Post-traumatic Stress Disorder (PTSD). The diagnosis of PTSD suffers from various human-centered design issues particularly in case of underserved communities. Language and cultural barriers, lack of skilled clinicians, lack of literacy and technological skills, sensitivity around disclosing traumatic experiences, etc., greatly undermine the effectiveness of the existing interview or self-report based diagnostic tools of PTSD. To address these issues, here, we present an automated and quantitative diagnostic model of PTSD based on the features from EEG and free-hand sketches.

In this regard, we performed an in-depth study among four diverse communities, e.g., Rohingya refugees (n = 71), slum-dwellers (n = 35), engineering students (n = 85), and healthy Bangladeshi individuals (n = 45). We used a consumer-grade, low-cost, and portable EEG headset to collect brainwave signals from Rohingya refugees and Bangladeshi citizens during three different activities. Besides, we collected free-hand sketches from the refugees, slum-dwellers, and engineering students. Based on the reported post-traumatic stress symptoms of the refugees, we developed PTSD regulatory network to determine causal associations among PTSD and its different symptoms. This model combines concepts from both reflective and formative models of PTSD. To the best of our knowledge, this is the first of its kind. Besides, we identified several neurobiological abnormalities related to PTSD using the EEG signals collected via our portable EEG headset while talking. Moreover, we used corner and edge detection algorithms to extract three features (number of corners, number of strokes, and average length of strokes) from the images of free-hand sketches. We used these features along with sketch themes, participants’ gender and group to train multiple logistic regression models for potentially screening PTSD (accuracy: 82.9-87.9%). We improved the accuracy (99.29%) by integrating EEG data with sketch features in a Random Forest model for the refugee population. Since