Skip to main content

Table 3 Performance of HFS-Net in the test set

From: A hierarchical fusion strategy of deep learning networks for detection and segmentation of hepatocellular carcinoma from computed tomography images

  

Segmentation

 

Per tumor volume

Per tumor cut

Per slice

Per patient

All

(n = 179)

Dice per case

58.7%

Sensitivity

88.9%

82.1%

84.3%

92.2%

Dice global

82.8%

Precision

51.5%

67.6%

75.5%

93.2%

MTD MAE

0.73 cm

F1-Score

65.3%

74.2%

79.6%

92.7%

1 ~ 2 cm

(n = 33)

Dice per case

32.0%

Sensitivity

69.4%

52.4%

57.0%

72.7%

Dice global

32.7%

Precision

45.5%

34.1%

39.5%

77.4%

MTD MAE

0.56 cm

F1-Score

54.9%

41.3%

46.7%

75.0%

2 ~ 3 cm

(n = 28)

Dice per case

47.3%

Sensitivity

90.3%

75.0%

77.8%

92.9%

Dice global

55.5%

Precision

57.1%

58.7%

66.7%

92.9%

MTD MAE

0.57 cm

F1-Score

70.0%

65.9%

71.7%

92.9%

3 ~ 5 cm

(n = 52)

Dice per case

60.4%

Sensitivity

89.1%

75.7%

79.8%

94.2%

Dice global

69.1%

Precision

49.0%

60.8%

69.0%

94.2%

MTD MAE

0.48 cm

F1-Score

63.2%

67.4%

74.0%

94.2%

> 5 cm

(n = 66)

Dice per case

76.1%

Sensitivity

98.5%

88.1%

89.5%

100%

Dice global

84.5%

Precision

54.0%

75.6%

84.2%

100%

MTD MAE

1.04 cm

F1-Score

65.7%

81.4%

86.8%

100%

1 tumor

(n = 151)

Dice per case

58.2%

Sensitivity

89.2%

81.7%

83.7%

92.7%

Dice global

84.2%

Precision

52.0%

66.7%

74.7%

93.3%

MTD MAE

0.70 cm

F1-Score

65.7%

73.4%

78.9%

93.0%

> 1 tumors

(n = 28)

Dice per case

62.9%

Sensitivity

87.5%

83.3%

86.0%

89.3%

Dice global

78.1%

Precision

49.1%

70.6%

78.2%

92.6%

MTD MAE

0.90 cm

F1-Score

62.9%

76.4%

82.0%

90.9%

  1. MTD, maximum tumor diameter; MAE, mean absolute error