Automatic segmentation of intracerebral hemorrhage from brain CT images

Abstract

Intracerebral hemorrhage (ICH) diagnosis is a neurological deficit that can occur in the patients suffering from high blood pressure and head trauma. Manual segmentation of ICH is tedious and operator dependent, therefore the purpose of this study is to present a robust fully automated system for hemorrhage detection from Computed Tomography (CT) scan images. The proposed method is based on White Matter Fuzzy c-Means (WMFCM) clustering and wavelet-based thresholding. The suggested method starts with the removal of components which do not look like brain tissues including skull by using a new WMFCM technique. After brain extraction, a new segmentation technique based on wavelet thresholding is used for detection and localization of hemorrhagic stroke. The proposed segmentation method is fast and accurate where standard evaluation metrics like dice similarity coefficients, Jaccard distance, Hausdorff distance, precision, recall, and F1 score are used to measure the accuracy of the proposed algorithm. Our method is demonstrated on a dataset of 20 brain computed tomography (CT) images suffered ICH and results obtained are compared with the ground truth of images. We found that our method can detect ICH with an average dice similarity of 0.82 and perform better as compared to standard fuzzy c-means (FCM) and spatial FCM (SFCM) clustering methods.