Performance evaluation of computed tomography liver image segmentation approaches

This paper presents and evaluates the performance of two well-known segmentation approaches that were applied on liver computed tomography (CT) images. The two approaches are K-means and normalized cuts. An experiment was applied on ten liver CT scan images, with reference segmentations, in order to test the performance of the two approaches. Experimental results were compared using an evaluation measure that highlights segmentation accuracy. Based on the obtained results in this study, it has been observed that K-means clustering algorithm outperformed normalized cuts segmentation algorithm for cases where region of interest depicts a closed shape, while, normalized cuts algorithm obtained better results with non-circular clusters. Moreover, for K-means clustering, different initial partitions can result in different final clusters.