PTSD Diagnosis using Deep Transfer Learning: an EEG Study

Post Traumatic Stress Disorder (PTSD) is a chronic mental and behavioral disorder that can develop following being exposed to a traumatic event. PTSD is diagnosed according to self-reports, which is prone to error in children and adults due to the fact that avoidance is one of the major symptoms of PTSD. In this paper, an automatic approach for diagnosing PTSD is proposed. We propose an EEG-based method since it is a low cost easily available imaging modality. Eyes closed resting-state EEG signals are recorded from 15 war-related PTSD and 15 matched control participants. After preprocessing, signals are divided into 1s segments. Time-frequency maps corresponding to each segment are achieved by applying the continuous wavelet transform. RGB images are generated using these time-frequency maps. They are fed to a convolutional neural network. In this paper, we use pre-trained VGG16 with proper modifications in its fully connected and classifier layers. To our best knowledge, this is the first study that uses deep transfer learning for diagnosing PTSD based on EEG signals. Our results suggest that the proposed approach can be an appropriate method for this purpose.