Segmentation of Liver Tumor for Computer Aided Diagnosis

Tumor segmentation provides a milestone for determination of the exact tumor size for computer aided diagnosis (CAD). Liver lesion segmentation is a significant step in liver cancer diagnosis, treatment planning and treatment evaluation. Liver Tumor Segmentation Challenge (LiTS) offers a common testbed for comparing different automated liver lesion segmentation techniques. The main use of digital image processing is to increase the quality of images for interpretation of human and machine understanding. Tumor segmentation in liver computed tomography (CT) volumes is considered as a complex task because of the varying tumor shape and texture. The location of the tumor also presents a challenge. Manual segmentation of tumors is a time-consuming task that can be inaccurate in some cases. The aim of this research is to propose an automated method, which can detect the tumor in each slice in volumetric CT liver images. The proposed algorithm is fully automated to segment out tumors by using the information provided by the sequence of CT volumes. The method is evaluated on LiTS 2017 dataset using dice score, volumetric overlap error and relative volume difference parameter. An average dice score of 0.60 is achieved, when evaluated on 80 CT volumes from LiTS, taking 34 seconds per slice. The results are significant when compared with state-of-the-art segmentation techniques used in CAD systems.