Skip to main content

Table 2 Result comparisons of different segmentation models

From: Automated segmentation and diagnosis of pneumothorax on chest X-rays with fully convolutional multi-scale ScSE-DenseNet: a retrospective study

Method

MPA

\({\mathrm {PA}}_1\)

DSC

\({\mathrm {DSC}}_1\)

Parameters

GFLOPS

U-Net [18]

0.90(0.16)*

0.81(0.31)*

0.89(0.17)*

0.79(0.33)*

7,764,098

11.59

SegNet [32]

0.91(0.15)*

0.81(0.31)*

0.90(0.16)*

0.80(0.31)*

29,444,162

40.14

DeepLab v3+ [33]

0.90(0.15)*

0.81(0.30)*

0.89(0.16)*

0.78(0.32)*

59,351,458

61.07

DenseASPP [34]

0.90(0.15)*

0.81(0.30)*

0.89(0.16)*

0.78(0.32)*

35,365,762

39.11

FC-DenseNet [25]

0.91(0.15)*

0.83(0.29)*

0.91(0.15)

0.82(0.29)

5,415,278

15.39

MS_scSE_U-Net

0.92(0.14)

0.85(0.28)

0.90(0.15)*

0.81(0.31)*

8,204,011

11.6

MS_scSE_FC-DenseNet (Ours)

0.93(0.13)

0.86(0.27)

0.92(0.14)

0.84(0.27)

5,989,096

15.59

  1. The number with * represents a significant difference comparing other methods to our method, according to student’s T-test for two independent samples (\(p<0.05\))