Table 1.

Results obtained by the U-Net model using the active semi-supervised learning procedure at each round. We report the number of images used for the training, the number of images used for the validation, the number of correctly segmented validation images (according to the expert evaluation), and the metric scores achieved (on the test set) after 150 epochs for each round of training, respectively. The percentages of training and validation images are referred to the whole set of available samples, i.e., 1564 images. The percentage of correct segmentation is referred to the total number of validated images per each round.

	Round 0	Round 1	Round 2	Round 3
N° training images	145 (9%)	368 (24%)	916 (59%)	1365 (87%)
N° validation images	1419 (91%)	1196 (76%)	648 (41%)	199 (13%)
N° correct segmentation	223 (16%)	548 (46%)	449 (69%)	112 (56%)
DSC metric	0.95	0.98	0.97	0.96
Precision metric	0.93	0.98	0.97	0.96
Recall metric	0.96	0.98	0.97	0.96