Entropy based automatic unsupervised brain intracranial hemorrhage segmentation using CT images

The present work proposes entropy based automatic unsupervised brain intracranial hemorrhage segmentation using CT images. The proposed work is consisting of fuzzy c-mean (FCM), automatic selection of cluster, skull removal, thresholding and edge-based active contour methods. The FCM is used to divide the image into different cluster and among these clusters one is automatically selected for the segmentation process which have skull and hemorrhage region. Due to major benefits of level set method i.e. use large time step which reduce the number of iterations and does not need reinitialization process thus it has fast convergence speed is used for smoothness of hemorrhage region. The seed point for level set method is initialized by entropy based thresholding techniques. The exhaustive experimentations have been carried out on 35 different patients CT images collected from SGRR Institute of Medical & Health Sciences and SMI Hospital, Dehradun, Uttarakhand, India. The performance of developed automatic segmentation techniques for hemorrhage detection is analyzed by using sensitivity, specificity, Dice coefficient, Jaccard index, Precision and Accuracy. After the evaluation of proposed method it can be observed that hemorrhagic regions segmented by the proposed method, has high accuracy comparison to FCM method and manual fuzzy based active contour method.