Table 2 Prospective comparison of PAN-VIQ and radiologists in vessel-specific vascular invasion prediction across five target vessels

From: A clinically validated 3D deep learning approach for quantifying vascular invasion in pancreatic cancer

Vessel

 

Classification

Accuracy

Precision

Recall

F1 score

Specificity

P

CA

Model

0

87.78%

88.33%

100.00%

93.75%

75.61%

NA

1

91.15%

91.15%

91.15%

96.54%

2

88.89%

88.89%

88.89%

87.35%

Senior radiologist

0

87.49%

90.63%

98.96%

94.66%

83.54%

0.673

1

87.50%

96.00%

91.67%

84.49%

2

92.31%

90.00%

91.15%

92.86%

Junior radiologist

0

87.50%

97.39%

98.42%

97.90%

44.44%

0.000

1

40.00%

28.57%

31.58%

98.44%

2

42.86%

42.86%

42.86%

97.93%

CHA

Model

0

90.22%

93.50%

96.74%

95.08%

91.43%

NA

1

93.33%

100.00%

96.55%

91.43%

2

96.77%

90.91%

93.75%

93.55%

Senior radiologist

0

89.71%

91.41%

99.43%

95.31%

86.74%

1.000

1

87.50%

96.84%

91.94%

89.71%

2

93.33%

93.75%

93.54%

92.86%

Junior radiologist

0

56.86%

50.00%

51.43%

50.70%

73.13%

0.000

1

59.26%

57.14%

58.18%

85.14%

2

61.54%

61.54%

61.54%

76.19%

SMA

Model

0

87.43%

88.54%

99.48%

93.75%

76.86%

NA

1

86.48%

100.00%

92.78%

87.43%

2

100.00%

90.53%

95.12%

98.43%

Senior radiologist

0

87.96%

92.67%

99.48%

95.96%

85.34%

0.000

1

85.71%

96.84%

90.91%

87.96%

2

100.00%

84.21%

91.30%

95.79%

Junior radiologist

0

63.06%

76.67%

50.00%

60.53%

89.23%

1.000

1

56.10%

65.71%

60.53%

76.32%

2

60.00%

80.00%

68.57%

80.25%

SMV

Model

0

88.55%

92.47%

98.18%

95.24%

84.85%

NA

1

86.27%

98.33%

92.11%

88.55%

2

95.00%

90.00%

92.31%

93.55%

Senior radiologist

0

87.88%

91.41%

99.40%

95.32%

86.74%

0.554

1

87.50%

96.43%

91.89%

87.88%

2

93.33%

90.00%

91.67%

92.86%

Junior radiologist

0

63.44%

54.29%

61.29%

57.58%

74.19%

0.000

1

70.37%

67.86%

69.09%

87.69%

2

67.74%

61.76%

64.62%

83.05%

PV

Model

0

87.36%

90.66%

98.90%

94.65%

83.51%

NA

1

87.50%

96.36%

91.76%

87.36%

2

92.31%

91.11%

91.72%

92.86%

Senior radiologist

0

87.92%

92.11%

99.45%

95.67%

58.97%

0.000

1

72.73%

54.55%

62.22%

87.92%

2

85.71%

66.67%

74.74%

93.55%

Junior radiologist

0

66.00%

66.67%

47.37%

55.38%

85.48%

0.000

1

62.86%

78.57%

69.84%

81.94%

2

68.42%

76.47%

72.22%

81.82%

  1. Performance metrics include accuracy, precision, recall, F1-score, and specificity. Classification labels (0, 1, 2) denote different diagnostic categories (e.g., absence, suspected, or confirmed vascular invasion). The p values (P) indicate statistical significance in comparisons between model and radiologist performance. Missing values (−) denote cases where metrics were not applicable or unavailable.
  2. CA celiac artery, CHA common hepatic artery, SMA superior mesenteric artery, SMV superior mesenteric vein, PV portal vein, NA not applicable.