Simple Atlas Selection Strategies for Liver Segmentation in CT Images

The paper present two atlas selection strategies: segmentation with a fixed, single individual atlas and segmentation with the best atlas for liver CT images. These two strategies was implemented and results were compared using DICE similarity coefficient, mean surface and Hausdorff distance. The average mean surface distance for single individual atlas equals 3.49 and 3.08 mm for Sum of Squared Differences and Mutual Information respectively. Average Hausdorff distance for the similarity measures mentioned above, which measure outliers, equal to 6.02 and 4.65 mm. The average DICE similarity coefficient are 0.49 and 0.59. The better results were obtained for Mutual Information similarity measure.