Table 2 Performance of the model in all cohorts.

From: A clinically applicable deep-learning model for detecting intracranial aneurysm in computed tomography angiography images

Cohort

Accuracy

Patient-level sensitivity

Specificity

PPV

NPV

Lesion-level sensitivity

FPs/case

Dice ratio

Internal cohort 1

86.0% (79.5%–90.7%)

97.3% (90.8%–99.3%)

74.7% (63.8%–83.1%)

79.4% (70.0%–86.4%)

96.6% (88.3%–99.1%)

95.6% (89.1%–98.3%)

0.29

0.75

Internal cohort 2

88.6% (83.7%–92.1%)

94.4% (87.8%–97.7%)

83.9% (76.5%–89.5%)

82.3% (74.1%–88.3%)

95.0% (89.1–98%)

84.1% (76.9%–89.5%)

0.26

0.65

Internal cohort 3

83.6% (78%–88.1%)

78.7% (66%–87.7%)

85.5% (78.9%–90.3%)

66.7% (54.5%–77.1%)

91.6% (85.7%–95.2%)

68.8% (57.3%–78.4%)

0.19

0.52

Internal cohort 4

85.8% (81.8%–89.1%)

73.6% (59.4%–84.3%)

87.9% (3.6%–91.1%)

50.0% (39.2%–60.8%)

95.3% (81.8%–89.1%)

60.6% (48.2%–71.7%)

0.20

0.45

Internal cohort 5

89.2% (85.2%–92.2%)

78.6% (48.8%–94.3%)

89.7% (85.7%–92.7%)

25.0% (13.7%–40.6%)

99.0% (96.7%–99.7%)

68.8% (41.5%–87.9%)

0.13

0.47

NBH cohort

81.0% (75–86%)

84.6% (68.8%–93.6%)

80.2% (73.3%–85.7%)

49.3% (37%–61.6%)

95.8% (90.8%–98.3%)

76.1% (62.1%–86.1%)

0.27

0.53

TJ cohort

8% (66.9%–81.5%)

76.1% (66.9%–83.6%)

71.1% (53.9%–84.0%)

88.3% (79.6%–93.7%)

50.9% (37%–64.7%)

73.0% (64.8–80%)

0.44

0.45

LYG cohort

76.6% (71.4%–81.1%)

85.0% (72.9%–92.5%)

74.6% (68.7%–79.7%)

44.0% (34.9%–53.5%)

95.5% (91.4%–97.8%)

78.9% (67.8%–87.1%)

0.32

0.56

  1. The data in parentheses are 95% confidence interval.
  2. FPs false positives, NPV negative predictive value, PPV positive predictive value.