Multi-Class Liver Cancer Diseases Classification Using CT Images

Abstract

Liver cancer is the fourth common cancer in the world and the third leading reason of cancer mortality. The conventional methods for detecting liver cancer are blood tests, biopsy and image tests. In this paper, we propose an automated computer-aided diagnosis technique for the classification of multi-class liver cancer i.e. primary, hepatocellular carcinoma, and secondary, metastases using computed tomography (CT) images. The proposed algorithm is a two-step process: enhancement of CT images using contrast limited adaptive histogram equalization algorithm and extraction of features for the detection and the classification of the different classes of the tumor. The overall achieved accuracy, sensitivity and specificity with the proposed method for the classification of multi-class tumors are 97%, 94.3% and 100% with experiment 1 and 84% all of them with experiment 2, respectively. By automatic feature selection scheme accuracy is deviated maximum by 10.5% from the overall and the ratio features accuracy decreases linearly by 5.5% with 20 to 5 selected features. The proposed methodology can help to assist radiologists in liver cancer diagnosis.