Multi-atlas Based Classification For Early Prediction Of Post-traumatic Stress Disorder

Since individuals may develop into post-traumatic stress disorder(PTSD)within days to six months after trauma exposure and PTSD can be prevented if early intervention is taken,identifying the patients of high risks to develop PTSD becomes more and more important.Although many studies have explored a variety of related predictors to predict the onset of PTSD through group comparisons,the findings are not consistent.72 participants underwent T1 and diffusion tensor imaging(DTI)scans after suffering a traffic accident.Thirty-two subjects were eventually diagnosed with PTSD.We parcellated regions-of-interest(ROIs)based on multiple atlases and extracted T1 or DTI features within those ROIs.Features were further fed into feature selection and classification to build an optimal model through which we can predict whether a new trauma-exposed subject would develop into PTSD.Meanwhile,features mostly involved in classification were selected as PTSD-related biomarkers.The accuracy of identifying PTSD can achieve 0.9167.A careful examination of the atlases was critical to ensure high-performance PTSD prediction.T1 features have higher contributions than DTI features and Lasso-based feature selection outperforms other feature selection methods.Four T1 features,measuring the cortical thickness of left lingual,right middle temporal,left cingulate gyrus and left parahippocampal gyrus and two DTI features,measuring white matter connectivity of left fusiform gyrus and left orbital gyrus,are eventually selected as biomarkers.This study turns out to be hopeful to help doctors not only evaluate the risk of having but also better understand the underlying cause of PTSD.