. 2022 Dec 15;84:102722. doi: 10.1016/j.media.2022.102722

Table 2.

Quantitative segmentation results of the lung regions on CT slices. The performance is reported as Dice (%), B-Acc (%) and MAE (%). 95% confidence intervals are presented in brackets. We performed experiments with classic segmentation methods such as U-Net (Ronneberger et al., 2015), U-Net++ (Zhou et al., 2019), and cutting-edge methods such as PraNet (Fan et al., 2020a), RBA-Net (Meng et al., 2020a), CABNet (Meng et al., 2020b), GRB-GCN (Meng et al., 2021c) and BI-Gconv (Meng et al., 2021a). Notably, we sampled 120 vertices for CABNet (Meng et al., 2020b) and RBA-Net (Meng et al., 2020a) to construct a smooth boundary.