TABLE 1.

Evaluation results on dice similarity coefficient (%), Jaccard (%), average symmetric surface distance (mm), and Root‐Mean‐Squared Error (mm) of different segmentation models on the National Institutes of Health (NIH) pancreas dataset

Models	DSC	Jaccard	ASD	RMSE
DAU‐Net	82.38 ± 5.46[70.01, 89.39]	70.39 ± 7.58[54.05, 80.82]	1.22 ± 0.32[0.58, 2.71]	2.26 ± 0.35[1.45, 3.65]
ADAU‐Net‐A1	82.68 ± 5.55[69.54, 89.94]	70.84 ± 7.65[53.71, 81.74]	1.09 ± 0.18[0.55, 2.20]	2.21 ± 0.34[1.44, 3.58]
ADAU‐Net‐A2	83.76 ± 4.94[71.60, 90.25]	72.38 ± 6.95[56.08, 82.24]	1.07 ± 0.23[0.49, 2.31]	2.17 ± 0.32[1.38, 3.52]
ADAU‐Net‐A3	82.96 ± 5.51[68.49, 89.85]	71.27 ± 7.58[52.28, 81.61]	1.19 ± 0.29[0.53, 2.65]	2.23 ± 0.38[1.43, 3.76]

Abbreviations: ADAU‐Net, attention‐guided duplex adversarial U‐Net; ASD, average symmetric surface distance; AU‐Net, adversarial U‐Net; DAU‐Net, duplex adversarial U‐Net; DSC, dice similarity coefficient.