INTRODUCTION (Liver cancer, , image-guided surgery, medical image segmentation, deep learning)
Primary liver cancer is the sixth most frequent cancer that accounts 6% globally and the second leading cause for mortality from cancer about 9%. The most frequent liver cancer, about 75% of all primary liver cancers is Hepatocellular Carcinoma (HCC) which is also called as hepatoma. Once after diagnosis, the cancer can be categorized into different levels such as potentially resectable, potentially transplantable and unresectable. Liver resection or hepatectomy is considered as the most effective and potential curative method. The main goal of the liver resection is to completely remove the tumor and the appropriate surrounding liver tissue without leaving any tumor cells behind. For technical reasons this procedure is challenging if the distribution of the tumors is within complex vasculature. Also the need for maintaining adequate functional liver volume with intact vascular inflow and outflow, biliary drainage accounts for increasing the complexity of whole procedure.
Current development in the field of image guided
Primary liver cancer is the fifth most cause for mortality in the world. Medical image segmentation has gained greater attention over the past decade, especially in the field of image-guided surgery. Accurate and fast liver segmentation tools are important for liver resection planning and navigation [1] . In this work we explore Convolutional Neural Network (CNN) based approach to segment liver from CT examinations. CNN has already proven efficiency in obtaining better results for different tasks such as object recognition, image classification, hand-written character recognition and many more. In this study we propose a comparison analysis of different initialization schemes for fast and accurate liver segmentation using CNN.