Fig. 7
From: Quality assurance for automatically generated contours with additional deep learning
Performance of the quality prediction model (MAE and rank correlation of predicted Dice scores) on different cases of failed segmentations: completely empty contours, pure noise, matrices full of ones, and shifted ground truths (GTs). The performance on real GT segmentation maps is also shown. All results are aggregated over 5 different validation splits for a total of 80 samples each