An Efficient Semiautomatic Active Contour Model of Liver Tumor Segmentation from CT Images

Abstract

Computed tomography (CT) image is one of the most extensively used imaging modalities for revealing and analyzing tumors as it has a higher spatial resolution, faster imaging speed relatively lower cost compared to MRI. The liver tumor volumetry requires the tumor segmentation in three dimensions which is performed manually tracing the tumor regions on slices which are tiresome and prolonged also, and the volume of manual demarcation is subjective. The major challenging task in liver tumor segmentation is due to the significant variation in location, shape, intensity and texture. This makes it difficult to develop a universal computer algorithm that is applicable for all cases. In this paper, an efficient semiautomatic technique for segmentation of liver tumor from CT images is proposed. The technique is based on semiautomated segmentation approach based on active contour segmentation using the level set method. The proposed approach consists of mainly four stages. In the first stage, the region of interest (ROI) image is initialized, which contains the liver tumor region in the CT image extraction using seed points; in the second stage, resampling and segmentation of ROI is done for reducing the noise and enhancing the boundaries; in the third stage, threshold values adjustment and seed point calculations are carried out; and in the fourth stage, the post-processing is done to extract and refine the liver tumor boundaries. The proposed technique is compared with the region growing algorithms that require filling holes and removing small connected components that can be achieved by using binary morphological operations: opening to remove small connected components and closing to fill holes. The liver tumors detected by the scheme were compared with those manually traced by experts, used as the ground truth results. The study was evaluated on two datasets of tumors. The proposed scheme obtained the Dice similarity coefficient 0.938 and the Jaccard similarity coefficient 0.883. The mean surface distance, the median surface distance and the maximal surface distance were 0.025, 0.712 and 2.828 mm, respectively. With the proposed scheme, the time requisite is reduced significantly and proves to be helpful to the observer in better visualizing the probable tumors inside the liver. The proposed method represents the 3D reconstruction of segmented tumors within the liver.