抄録
This paper presents several procedures for extraction of liver vessel regions from EOB-primovist MR (EOB-MR) images with evaluation of their performances. In development of liver CAD systems using EOB-MR images, it is very important to provide a function for extracting accurately liver vessel regions from the images, because vessel information is essentially needed for discrimination between small tumors and others. We have already developed a procedure for liver vessel region extraction using a region growing technique, and then have obtained 57% of the similarity value (JI) between the extracted result by the procedure and the true region in average. In this paper, new types of extraction procedures using Top-Hat operation and LOG (Laplacian of Gaussian) respectively are developed, and performances of three procedures including the previous one are compared. In the experiments using actual EOB-MR 8 cases (including 5 cancer cases), it was known that the relationship among similarity values obtained by three procedures was region growing > Top-Hat > LOG. In the paper, the difference of shape features between vessel regions extracted by three procedures will be also discussed.
