. Author manuscript; available in PMC: 2023 Mar 27.

Published in final edited form as: Med Image Anal. 2022 Dec 14;84:102726. doi: 10.1016/j.media.2022.102726

Table 1.

The performance of different methods on five datasets. Output $ℬ$ : the output of the U-Net. Segmentation S_f₁₈ the SAM on the last layer (the 18th layer in U-Net). It shows that training the network with the SA score in the loss does not hurt the performance.

Dataset	Offline ProtoSeg		Online ProtoSeg
Dataset	Output $ℬ$	Segmentation S_f₁₈	Output $ℬ$	Segmentation S_f₁₈
BraTS	0.8471 ± 0.23	0.7944 ± 0.25	0.8428 ± 0.24	0.8514 ± 0.23
ISIC	0.8330 ± 0.18	0.8260 ± 0.17	0.8353 ± 0.17	0.8445 ± 0.16
COVID	0.4862 ± 0.22	0.5123 ± 0.21	0.4840 ± 0.22	0.5597 ± 0.17
Prostate	0.7232 ± 0.28	0.6365 ± 0.34	0.7557 ± 0.20	0.7585 ± 0.21
Pancreas	0.6475 ± 0.25	0.3439 ± 0.22	0.6441 ± 0.23	0.6635 ± 0.22