Identifying Patients With PTSD Utilizing Resting-State fMRI Data and Neural Network Approach

Abstract

Purpose: The primary aim of the study is to identify the existence of post-traumatic stress disorder (PTSD) in an individual and to detect the dominance level of each affected brain region in PTSD using rs-fMRI data. This will assist the psychiatrists and neurologists to distinguish impartially between PTSD individuals and healthy controls for the brain-based treatment of PTSD. Methods: Twenty-eight individuals (14 with PTSD, 14 healthy controls) were assessed to obtain rs-fMRI data of their six brain regions-of-interest. The rs-fMRI data analyzed by the Artificial Neural Network (ANN), adopting the training-validation-testing approach to classify PTSD and to identify the most affected brain region due to PTSD. The classification accuracy is justified by a variety of different methods and metrics. Results: Three ANN models were established to attain the study’s purpose using the susceptible regions in the right, left, and both hemispheres and the classification accuracy of ANN models achieved 79%, 93.5%, and 94.5%, respectively. The prediction accuracy even increased in the independent holdout sample using trained models. The developed models are reliable, intellectually attractive, and generalize. Additionally, the most dominant region in the PTSD individuals was the left hippocampus and the least was the right hippocampus. Conclusion: The present investigation achieved high classification accuracy and identified the brain regions that highly contributed to differentiating PTSD individuals from healthy controls. The results indicated that the left hippocampus is the most affected brain region in PTSD individuals. Therefore, our findings are helpful for practitioners for diagnostic, medication, and therapy of the affected brain regions by knowing the strength of infected regions.

Keywords

• Artificial neural network
• amygdala
• calibration plot
• health-care
• hippocampus
• medial prefrontal cortex