An Edge-Assisted Computing and Mask Attention Based Network for Lung Region Segmentation

Table 3

Results of EAM-Net and Res-UNet using different encoder backbones on the JSRT, Shenzhen, and Montgomery datasets.

Methods	JSRT			Shenzhen			Montgomery
	PA (%)	Dice (%)	JA (%)	PA (%)	Dice (%)	JA (%)	PA (%)	Dice (%)	JA (%)
Res-UNet18	98.40	97.37	94.88	97.69	95.37	91.17	98.56	97.01	95.03
EAM-Net18	98.71	97.88	95.88	98.00	95.92	92.07	99.09	98.11	96.30
Res-UNet34	98.43	97.42	94.98	97.77	95.42	91.31	98.58	97.34	95.39
EAM-Net34	98.83	97.99	96.00	98.04	96.00	92.18	99.15	98.22	96.51
Res-UNet50	98.59	97.69	95.49	97.85	95.53	91.40	98.65	97.41	95.60
EAM-Net50	98.96	98.11	96.15	98.20	96.13	92.27	99.16	98.23	96.52

Bold values represent the the highest performance for each performance metric.