Fuzzy C-Means Based CAD Sytem for Liver Tumors Segmentation from CT Scans

Abstract

Mortality rate worldwide due to liver tumors are relatively high. Liver tumors are the leading cause of liver cancer. Hence, liver cancer mortality rates can be reduced with prediction and accurate staging of liver tumors in early stages. To assist radiologist diagnosis decision making, computer-aided diagnosis (CAD) is used along with different imaging modalities. Precise and reliable segmentation of the liver and tumors are essential in CAD systems pipeline. Computer-Aided Detection (CADe) systems and Computer-Aided Diagnosis (CADx) systems, are the two primary CAD systems categories. This paper mainly aims to propose a CADe system framework to automatically segment the liver along with liver tumors using Fast-Generalized Fuzzy C-Means (FG-FCM) and Kernel-Based Fuzzy C-Means (K-FCM), respectively. Computed tomography (CT) is used in this paper as the imaging modality. The applicability of our framework was verified with 250 CT volumes from various benchmark datasets namely; LiTS17, 3Dircadb and MICCAI-Sliver07. Reliability, reasonability and applicability of our system was proved through experimental results with an overall average accuracy of 93.96%, recall/sensitivity of 88.79%, selectivity/specificity of 97.36%, Dice similarity coefficient of 90.93%. The proposed framework has the advantages of being unsupervised, fully-automatic with low noise sensitivity and high in-homogeneity correction in comparison with other frameworks.