Fully Automatic Liver Segmentation from Multi-slice Spiral Computed TomographicImages

Liver segmentation from Multi-slice spiral computed tomography (MSCT) is the key techniquein liver volume estimation, which is important for livercancer treatment. This paper proposes a fully automaticmethod to segment liver regions from MSCT. To beginwith, we took advantage of the correlations between neighbor slice images to get the initial fronts. Then a modifiedfast marching algorithm is applied to propagate the frontsuntil the stop criterion was satisfied. The areas includedby the propagated fronts were the liver regions. Completeliver regions of one case could be fetched from each sliceusing the similar approach one by one. Our method wastested on 562 slice images from 8 cases collected from a hospital in China. With the comparison of the doctors' manualsegmentation results, the difference percentage was 4.7%averagely with the maximum value 9.1% and the minimumone 1.4% respectively.