AutoLiv: Automated Liver Tumor Segmentation in CT Images

Abstract

Gastrointestinal (GI) cancer consists of a group of ten cancers that affect the various accessory organs of the digestive system and liver cancer is one of them. In India, it is ranked twelfth in terms of new cases, eight in terms of deaths and increasing as per the Global cancer Observatory data of last year. Like other cancers, it can be cured if detected early. But the diagnostic performance of Computerized Tomography (CT) images for Liver cancer is interpreter-dependent and prone to human errors. Medical image segmentation and analysis of tumor can help in Computer-aided diagnosis (CAD). Automatic Segmenting of liver and tumor is a complex task as it depends on the shape, location, texture and intensity. Therefore, to develop a general-purpose algorithm that fits all is not possible. Both these tasks can be performed either manually or in a semi-automated manner. In this paper we present AutoLiv, automated liver-tumor detection in CT images. In the first stage, threshold-based slope difference differentiation (SDD) technique is used for segmentation of liver and using this in the second stage we carry out tumor detection by alternative fuzzy c-means (AFCM) clustering algorithm. MATLAB based results and manual segmentation results are compared. A close correlation is observed between both the manual and automated approach with very high degree of spatial overlap seen in the regions-of-interest (ROIs) isolated by both methods.