This paper proposes a method for automated liver region extraction by analyzing distributions of CT values from multi-phase 3D abdominal X-ray CT images. A simple thresholding method does not work well for extracting liver regions from single phase CT images, since CT values of a liver region are very similar to those of other organs such as the spleen and muscle. The proposed method estimates CT value distributions of liver, spleen, and muscle regions using early and late arterial phase CT images. We assume these distributions as Gaussian and their mixture, and estimate the distributions by the EM algorithm. The estimated distributions are used for determining the optimum threshold values for extracting initial liver region, Hepatocellular carcinoma (HCC), and other organs. Then, the liver region is obtained by combining initial liver region and HCC region, and other regions are used for eliminating over extraction of the liver region. Finally, the surface of the extracted liver region is smoothed along the border of the liver region. We applied the proposed method to 26 cases of multi-phase abdominal CT images. The experimental results showed that the proposed method succeeded in the satisfactory extraction of liver regions.

View full abstract