Table 2 Comparison of segmentation performance across different models without data augmentation.

From: Attention-based multi-residual network for lung segmentation in diseased lungs with custom data augmentation

Method

Year

Params (M)

FLOPs (G)

MJ

GMH

COVID

Shenzhen set

DSC

JI

DSC

JI

DSC

JI

DSC

JI

U-Net11

2015

7.76

24.1

0.9546 ± 1.93

0.9214 ± 2.88

0.8475 ± 2.38

0.7479 ± 3.39

0.9140 ± 0.98

0.8510 ± 1.35

0.9168 ± 1.05

0.8519 ± 1.60

U-Net++13

2018

9.03

58.4

0.9543 ± 2.76

0.9248 ± 3.62

0.8610 ± 1.59

0.7650 ± 2.21

0.9291 ± 0.56

0.8718 ± 0.86

0.9250 ± 0.34

0.8647 ± 0.47

AttentionUNet16

2018

0.84

16.5

0.9570 ± 2.08

0.9257 ± 2.82

0.8524 ± 1.83

0.7512 ± 2.65

0.9284 ± 0.27

0.8699 ± 0.48

0.9197 ± 0.61

0.8550 ± 1.05

ResUNet++12

2019

4.07

32.0

0.9655 ± 1.04

0.9367 ± 1.60

0.8598 ± 2.45

0.7679 ± 3.58

0.9317 ± 0.40

0.8765 ± 0.55

0.9217 ± 0.81

0.8619 ± 1.23

DC-UNet18

2021

10.06

39.2

0.9570 ± 2.69

0.9288 ± 3.40

0.8303 ± 0.74

0.7221 ± 1.24

0.9256 ± 0.99

0.8666 ± 1.56

0.9107 ± 1.04

0.8417 ± 1.68

DCSAU-Net19

2023

2.59

6.88

0.9632 ± 1.59

0.9342 ± 2.35

0.8336 ± 1.64

0.7430 ± 2.51

0.9220 ± 1.31

0.8647 ± 1.75

0.9255 ± 0.57

0.8683 ± 0.88

I2U-Net20

2024

6.75

3.57

0.9590 ± 2.10

0.9297 ± 2.79

0.8856 ± 1.17

0.8080 ± 1.58

0.9352 ± 0.57

0.8823 ± 0.87

0.9270 ± 0.75

0.8710 ± 1.02

AMRU++ (Proposed)

 

10.65

51.0

0.9628 ± 1.67

0.9346 ± 2.44

0.9097 ± 1.54

0.8403 ± 2.36

0.9422 ± 0.45

0.8930 ± 0.74

0.9338 ± 0.38

0.8797 ± 0.66

  1. Fivefold cross validation was used and the standard deviation were computed as percentages (%). The best results are indicated in bold.