Detection of Schizophrenia Disorder from Ventricle Region in MR Brain Images via Hu Moment Invariants Using Random Forest

摘要

In this work, the classification of SZ from normal is identified using random forest classifier with Hu moments derived from ventricle region. At the beginning, the ventricle segmentation from MR brain images was implemented using multiplicative intrinsic component optimization method. The analysis shows that a multiplicative intrinsic component optimization method is capable to improve the bias correction and achieve better segmentation of ventricle from the skull stripped images. The calculated area from segmented ventricle region with ground truth shows high correlation (>0.98) for both normal and abnormal images. The high similarity measure shows the ability of the proposed method in segmenting the ventricle region. The Hu moment invariant along with random forest classifier gives better results compared to conventional methods. The classification accuracy is 82.73% for the entire feature set. The sensitivity and specificity are 82% and 83.33%, respectively. The AUC is 0.827. Hu moment invariant features extracted from segmented ventricle gives significant variation between the normal and SZ subjects. Hence, the Hu moment invariants extracted from ventricle along with random forest classifier could aid the physician in better diagnosis of schizophrenia subjects.