Automatic Liver Segmentation in CT Images Using Improvised Techniques

Prerna Kakkar, Sushama Nagpal, Nalin Nanda

Abstract: Computer aided automatic segmentation of liver can serve as an elementary step for radiologists to trace anomalies in the liver. In this paper, we have explored two techniques for liver segmentation - Region growing technique of Morphological Snake and a graph-based technique called Felzenszwalb. The aforementioned techniques have been modified by incorporating Artificial Neural Network (ANN) for automatic seed generation eliminating any user intervention. It has been tested on an open-source dataset of Liver CT Scans. Compared to the algorithms that have been used in past, the algorithms discussed in this paper are computationally much efficient in terms of time. Both algorithms were able to segment liver with high accuracy but Morphological Snake outperformed Felzenszwalb in terms of segmentation by achieving a dice index of 0.88 and a very high accuracy of 98.11%. However, Felzenszwalb computed results at a faster rate.