A Discussion on the Evaluation of A New Automatic Liver Volume Segmentation Method for Specified CT Image Datasets

This paper presents discussions on experimental result evaluation outcomes of a new liver volume segmentation method developed for 10 specified CT image datasets. Precise liver surface segmentation is the first step and one of the major tasks in individual surgical resection virtual reality simulations. There are five major difficulties: Firstly, the automatic initialization of liver detection is often unreliable. Secondly, the liver surface shows great anatomical variations amongst patients and even in the same patient over time; Thirdly, the intensities of liver and kidney are similar in certain scanning planes, which makes their contacting boundaries weaker and fuzzier than most hepatic vessel or portal vein walls; Fourthly, in some cases the liver tail and the apex of the heart can easily be confused. Finally, the boundary between subcostal fat tissue and the liver in many cases is ambiguous. This paper presents a new automatic strategic active contour method for accurate and reproducible liver volume segmentation, which contains different ways to tackle different difficulties. Our method integrates a rotational template matching, and k-means clustering followed by rib cage area local edge enhancement, with a GVF (Gradient Vector Flow) geometric Snake. This proposed method has been trained by 20 specified CT image datasets, and implemented and evaluated on another 10 testing datasets.