Figure 4:

Example of segmentation error metrics. (A) An axial section of the heart and the corresponding ground truth and predicted segmentation from one of the patients; model combinations are shown. The Dice coefficient was used to assess overall segmentation accuracy and is illustrated by the fused image (right), with agreement shown in white and disagreement in black (in this section, the left ventricular [LV] Dice coefficient = 0.910). (B) Segmentation errors due to the compressed image representations are expected to occur primarily at the boundary. Therefore, we isolated the boundary of each segmentation and calculated a border Dice score (in this section, the LV border Dice coefficient = 0.402).