Intensity population based unsupervised hemorrhage segmentation from brain CT images

Abstract

This article has proposed an intelligent knowledge driven method to segment hemorrhage from brain CT images using the information of pixel intensity population and distribution. A mathematical model is designed to identify the unexpected variation in pixel intensity population in a brain CT image having hemorrhage. Complete batch of multi-slice CT scan images is taken as input. Fusion of knowledge of brain anatomy with intensity distribution information of CT brain image results in a unique solution for hemorrhage segmentation. To test the robustness, segmentation of different types of hemorrhage of different patients is done using the proposed method. The results are accepted and validated by radiology experts. A fully automatic and fast Computer Aided Diagnosis (CAD) is designed, using the proposed method, to segment hemorrhage automatically, in the absence of an expert, for further inspections like checking severity, volume, size, shape and type of hemorrhage. Competence of the CAD is tested against mostly used established clustering methods to demonstrate its potential.