Combination of grey matter and white matter features for early prediction of posttraumatic stress disorder

Abstract

Posttraumatic stress disorder (PTSD) is a prevalent psychiatric disorder. In previous researches, there are few studies about structural and functional alterations of the whole brain simultaneously about PTSD prediction. Early alterations could provide evidence of early diagnosis and treatment. Early diagnosis of PTSD plays an important role during the treatment. In this work, we extract discriminant features from multi-modal images and implement classification-based prediction for PTSD onset. Specifically, discriminant features are a collection of measures derived from grey matter (GM) and white matter (WM). We choose cortical thickness of GM and three descriptions of WM connection which are fiber count, fractional anisotropy (FA), and mean diffusivity (MD). After applying automated anatomical labeling (AAL) to parcellate the whole brain into 90 regions-of-interest (ROIs), the descriptions can be quantified. Then, a weighted clustering coefficient of every ROI connected with the remaining ROIs is extracted as feature. GM features and WM features are combined and selected automatically, which are later utilized by support vector machine (SVM) for early identification of the patients. The classification accuracy is around 79.86 % as the area of receiver operating characteristic (ROC) curve is 0.816 evaluated via dual leave-one-out cross-validation.