Table 2.

Quantitative evaluation of different networks for lung opacification segmentation. The BCE-Dice loss was used for training.

Methods	DSC (%)	SEN (%)	PPV (%)	VA (%)	RLP (%)	RLR (%)	95% HD(mm)
PSPNet [23]	80.86	75.67	88.87	84.42	89.12	76.24	59.93
ESPNetv2 [46]	83.19	79.77	88.61	89.03	67.84	78.31	63.96
DenseASPP [47]	86.87	85.76	88.98	94.83	88.62	78.71	51.96
DeepLabV3+ [24]	85.26	83.97	88.33	93.75	89.16	78.57	53.61
U-Net [10]	83.61	82.96	85.57	92.57	86.18	76.48	73.50
COPLE-Net [17]	83.70	84.27	83.42	93.45	77.46	74.60	59.21
CE-Net [48]	85.78	84.46	87.88	94.70	82.79	79.45	55.85
Attention U-Net [25]	82.66	79.95	86.58	90.43	88.20	75.22	60.97
UNet+ [26]	81.83	80.29	84.03	91.87	80.30	76.72	74.32
Proposed	88.99	87.85	90.28	96.25	90.87	84.83	29.16