Table 1 Comparison with state-of-the-art methods across five datasets (Kvasir-SEG, CVC-ClinicDB, ColonDB, EndoScene, ETIS)

From: PrysmNet a polyp refining system using salience and multimodal guidance for reproducible cross domain segmentation

Method

Complexity

Kvasir

ClinicDB

ColonDB

EndoScene

ETIS

 

Params

FLOPs

mDice

mIoU

BF

mDice

mIoU

BF

mDice

mIoU

BF

mDice

mIoU

BF

mDice

mIoU

BF

U-Net16

24.56

38.26

81.86 ± 0.42

74.61 ± 0.45

68.42 ± 0.58

82.33 ± 0.40

75.56 ± 0.43

69.16 ± 0.55

51.21 ± 0.85

44.46 ± 0.88

38.58 ± 1.10

71.08 ± 0.52

62.61 ± 0.55

58.28 ± 0.68

39.80 ± 0.95

33.58 ± 0.98

28.48 ± 1.20

UNet++17

25.09

84.30

82.11 ± 0.40

74.31 ± 0.43

69.10 ± 0.55

79.49 ± 0.41

72.94 ± 0.44

67.82 ± 0.58

48.39 ± 0.82

41.08 ± 0.85

36.24 ± 1.05

70.76 ± 0.50

62.49 ± 0.53

58.08 ± 0.65

46.18 ± 0.88

34.49 ± 0.92

32.12 ± 1.15

SFA36

–

–

72.33 ± 0.55

61.13 ± 0.58

58.38 ± 0.70

70.00 ± 0.52

60.73 ± 0.56

56.48 ± 0.68

46.97 ± 0.90

34.77 ± 0.95

33.18 ± 1.25

46.79 ± 0.75

32.96 ± 0.78

30.58 ± 0.95

29.73 ± 1.10

21.74 ± 1.15

18.91 ± 1.30

ACSNet37

46.02

29.45

89.87 ± 0.20

83.03 ± 0.22

79.55 ± 0.30

88.09 ± 0.21

82.64 ± 0.23

78.28 ± 0.32

71.62 ± 0.45

64.99 ± 0.48

59.47 ± 0.60

86.30 ± 0.25

78.79 ± 0.28

75.11 ± 0.35

57.98 ± 0.65

50.05 ± 0.68

45.35 ± 0.80

PraNet7

32.50

221.90

89.84 ± 0.18

84.08 ± 0.20

80.22 ± 0.28

89.90 ± 0.19

84.95 ± 0.21

81.51 ± 0.29

70.95 ± 0.42

64.08 ± 0.45

58.83 ± 0.58

87.16 ± 0.22

79.70 ± 0.25

76.44 ± 0.33

62.84 ± 0.60

56.75 ± 0.63

51.24 ± 0.75

SANet38

–

–

90.44 ± 0.16

86.47 ± 0.18

82.12 ± 0.25

93.74 ± 0.15

88.96 ± 0.17

85.36 ± 0.24

75.35 ± 0.38

67.08 ± 0.40

62.12 ± 0.52

88.85 ± 0.20

81.57 ± 0.22

78.25 ± 0.30

75.08 ± 0.55

65.42 ± 0.58

60.59 ± 0.70

TransFuse24

115.59

38.73

91.89 ± 0.12

86.85 ± 0.14

83.43 ± 0.20

93.48 ± 0.11

88.69 ± 0.13

85.13 ± 0.19

74.43 ± 0.35

67.63 ± 0.38

61.84 ± 0.48

90.42 ± 0.15

83.86 ± 0.18

80.54 ± 0.25

73.72 ± 0.50

66.15 ± 0.53

61.21 ± 0.65

TransUNet39

105.28

24.66

91.34 ± 0.14

85.70 ± 0.16

82.51 ± 0.22

93.56 ± 0.12

88.79 ± 0.14

84.98 ± 0.20

78.14 ± 0.32

69.99 ± 0.35

64.29 ± 0.45

89.30 ± 0.18

82.47 ± 0.20

79.11 ± 0.28

73.11 ± 0.48

66.08 ± 0.50

59.82 ± 0.62

HarDNet–MSEG40

33.80

192.74

92.12 ± 0.11

85.97 ± 0.13

83.20 ± 0.19

93.27 ± 0.10

88.25 ± 0.12

84.80 ± 0.18

73.16 ± 0.33

66.09 ± 0.36

60.55 ± 0.45

90.88 ± 0.14

83.29 ± 0.16

80.14 ± 0.24

77.78 ± 0.45

69.02 ± 0.48

63.46 ± 0.58

DS–TransUNet41

177.44

30.97

93.44 ± 0.09

88.82 ± 0.11

85.61 ± 0.17

93.83 ± 0.09

89.13 ± 0.10

86.22 ± 0.16

79.83 ± 0.28

71.72 ± 0.31

66.31 ± 0.40

90.06 ± 0.12

82.84 ± 0.14

80.45 ± 0.22

77.09 ± 0.42

69.01 ± 0.45

64.13 ± 0.55

SwinE-Net42

–

–

92.08 ± 0.10

87.04 ± 0.12

84.54 ± 0.18

93.87 ± 0.09

89.20 ± 0.11

86.59 ± 0.17

80.44 ± 0.26

72.51 ± 0.29

67.17 ± 0.38

90.63 ± 0.13

84.09 ± 0.15

81.27 ± 0.23

75.87 ± 0.40

68.75 ± 0.43

62.88 ± 0.53

Polyp-PVT21

–

–

91.77 ± 0.10

86.44 ± 0.12

83.82 ± 0.18

93.73 ± 0.09

88.90 ± 0.11

85.71 ± 0.17

80.69 ± 0.25

72.78 ± 0.28

67.48 ± 0.36

90.08 ± 0.12

83.31 ± 0.15

80.39 ± 0.22

78.77 ± 0.38

70.67 ± 0.41

65.22 ± 0.50

CaraNet43

46.64

21.69

91.80 ± 0.11

86.61 ± 0.13

83.55 ± 0.20

93.62 ± 0.10

88.83 ± 0.12

85.50 ± 0.18

77.35 ± 0.29

68.91 ± 0.32

63.26 ± 0.42

90.31 ± 0.14

83.31 ± 0.16

80.64 ± 0.24

74.71 ± 0.42

67.20 ± 0.45

61.56 ± 0.56

ColonFormer44

52.94

22.94

92.42 ± 0.08

87.66 ± 0.10

85.11 ± 0.16

94.01 ± 0.08

89.70 ± 0.10

86.84 ± 0.15

81.14 ± 0.24

73.33 ± 0.27

68.51 ± 0.35

90.67 ± 0.11

84.21 ± 0.13

81.05 ± 0.20

80.13 ± 0.35

72.21 ± 0.38

66.81 ± 0.48

SegT45

–

–

92.77 ± 0.07

88.01 ± 0.09

85.58 ± 0.15

94.02 ± 0.07

89.74 ± 0.09

87.18 ± 0.14

81.43 ± 0.23

73.07 ± 0.26

68.84 ± 0.34

89.54 ± 0.11

82.88 ± 0.13

79.59 ± 0.21

81.03 ± 0.32

73.26 ± 0.35

67.54 ± 0.45

Polyper (final)25

29.11

43.54

94.87 ± 0.05

90.30 ± 0.07

88.24 ± 0.12

94.48 ± 0.05

89.89 ± 0.08

87.51 ± 0.11

83.73 ± 0.18

74.56 ± 0.21

70.12 ± 0.28

92.46 ± 0.09

86.70 ± 0.11

83.50 ± 0.16

86.52 ± 0.22

78.52 ± 0.25

78.23 ± 0.32

Prysm-Net (Ours)

34.72

52.87

95.14 ± 0.04

90.93 ± 0.06

89.47 ± 0.11

94.98 ± 0.05

90.31 ± 0.07

88.61 ± 0.10

84.61 ± 0.15

75.89 ± 0.18

71.29 ± 0.24

92.96 ± 0.08

87.39 ± 0.10

84.28 ± 0.13

87.26 ± 0.19

79.69 ± 0.21

79.17 ± 0.29

  1. All models were re-implemented and evaluated under a unified protocol. We report mDice, mIoU, and BF (%) as well as model complexity (Params, M; FLOPs, G). Bold indicates the best performance. “–" indicates the metric was not reported. Prysm-Net (Ours) follows the same protocol. In inference, only the ViT backbone, decoder, and BSM are active; the training-time FMDM/MGM provide guidance only and are disabled at test.