Fig. 3.

Overview of our proposed method. Training: TR1) computation of ground-truth Dice ratio between each pair of atlas label maps after non-rigid registration, TR2) computation of pairwise features from the key regions between each pair of atlas images after affine alignment, and TR3) learning of the relationship between pairwise features and ground-truth DR. Testing: TS1) affine alignment of the target image to the common space, TS2) computation of pairwise features between the target and all the atlas images, TS3) prediction of the segmentation performance by using the learned model, and TS4) selection of atlases with the highest scores for multiple-atlas segmentation.