Table 2.

Comparison of different models for the GVA (844 samples) and IMIM (381) datasets.

GVA Dataset
Model	R1 vs. Model	R2 vs. Model	AND vs. Model	Closest Radiologist
ECNN	0.71 ± 0.23	0.72 ± 0.23	0.72 ± 0.23	0.77 ± 0.21
UNet	0.76 ± 0.15	0.76 ± 0.15	0.76 ± 0.16	0.81 ± 0.12
YNet (param.)	0.70 ± 0.21	0.70 ± 0.21	0.70 ± 0.22	0.75 ± 0.20
YNet (mask)	0.80 ± 0.14	0.78 ± 0.15	0.78 ± 0.16	0.84 ± 0.12
AND-ECNN	0.68 ± 0.21	0.64 ± 0.21	0.62 ± 0.22	0.72 ± 0.20
AND-UNet	0.76 ± 0.15	0.79 ± 0.15	0.81 ± 0.14	0.83 ± 0.12
AND-YNet (param.)	0.65 ± 0.22	0.68 ± 0.22	0.69 ± 0.23	0.72 ± 0.21
AND-YNet (mask)	0.77 ± 0.15	0.79 ± 0.15	0.82 ± 0.14	0.84 ± 0.12
CM-ECNN	0.71 ± 0.20	0.72 ± 0.20	0.73 ± 0.20	0.78 ± 0.18
CM-UNet	0.80 ± 0.13	0.80 ± 0.13	0.79 ± 0.15	0.85 ± 0.10
CM-YNet (param.)	0.73 ± 0.18	0.75 ± 0.17	0.77 ± 0.17	0.80 ± 0.15
CM-YNet (mask)	0.79 ± 0.13	0.80 ± 0.13	0.79 ± 0.15	0.84 ± 0.10
IMIM Dataset
Model	R1 vs. Model	R2 vs. Model	AND vs. Model	Closest Radiologist
ECNN	0.58 ± 0.24	0.59 ± 0.23	0.55 ± 0.25	0.65 ± 0.23
UNet	0.67 ± 0.22	0.69 ± 0.18	0.64 ± 0.23	0.74 ± 0.18
YNet (param.)	0.60 ± 0.27	0.60 ± 0.24	0.56 ± 0.27	0.66 ± 0.25
YNet (mask)	0.69 ± 0.24	0.69 ± 0.20	0.64 ± 0.24	0.76 ± 0.20
AND-ECNN	0.58 ± 0.26	0.57 ± 0.23	0.51 ± 0.26	0.64 ± 0.24
AND-UNet	0.68 ± 0.24	0.72 ± 0.21	0.68 ± 0.25	0.76 ± 0.20
AND-YNet (param.)	0.58 ± 0.26	0.60 ± 0.24	0.57 ± 0.26	0.65 ± 0.24
AND-YNet (mask)	0.68 ± 0.24	0.71 ± 0.21	0.68 ± 0.24	0.76 ± 0.21
CM-ECNN	0.63 ± 0.25	0.65 ± 0.21	0.60 ± 0.25	0.71 ± 0.21
CM-UNet	0.69 ± 0.22	0.72 ± 0.18	0.66 ± 0.24	0.77 ± 0.17
CM-YNet (param.)	0.67 ± 0.23	0.69 ± 0.20	0.65 ± 0.23	0.74 ± 0.19
CM-YNet (mask)	0.68 ± 0.23	0.70 ± 0.19	0.64 ± 0.24	0.76 ± 0.18

DICE scores are shown as mean ± standard deviation. The last column shows the mean score between the model segmentation and the label of the closest radiologist for each sample. The highest value for each column is highlighted in bold.