Skip to main content

Table 4 Performance of standalone AI the historical first reader – by site and mammography equipment vendor

From: Multi-vendor evaluation of artificial intelligence as an independent reader for double reading in breast cancer screening on 275,900 mammograms

A) MK / IMS Giottoa

  

Performance Metric

Historical first reader (%)

Standalone AI (%)

On ten-year cohort

  

 Sensitivity

79.4 (76.3, 82.2)

85.7 (83.0, 88.1)

 Specificity

95.4 (95.0, 95.7)

96.1 (95.7, 96.4)

On 2015-year cohort: with more complete IC data available

 Sensitivity

82.0 (72.8, 88.6)

85.4 (76.6, 91.3)

 Specificity

96.5 (95.4, 97.3)

96.2 (95.0, 97.0)

B) NUH / GEb

Performance Metric

Historical first reader (%)

Standalone AI (%)

On ten-year cohort

 Sensitivity

77.8 (74.6, 80.7)

76.9 (73.7, 79.9)

 Specificity

97.3 (97.0, 97.5)

89.6 (89.2, 90.0)

On 2015-year cohort: with more complete IC data available

 Sensitivity

67.2 (58.4, 75.0)

72.3 (63.6, 79.5)

 Specificity

97.2 (96.8, 97.5)

90.6 (89.9, 91.3)

C) LTHT / Hologicb

Performance Metric

Historical first reader (%)

Standalone AI (%)

On ten-year cohort

 Sensitivity

81.0 (77.8, 84.0)

79.9 (76.5, 82.9)

 Specificity

95.0 (94.7, 95.3)

89.2 (88.8, 89.6)

On 2015-year cohort: with more complete IC data available

 Sensitivity

82.8 (73.9, 89.1)

84.9 (76.3, 90.8)

 Specificity

95.7 (95.1, 96.2)

89.3 (88.5, 90.1)

D) ULH / Siemensb

Performance Metric

Historical first reader (%)

Standalone AI (%)

On ten-year cohort

 Sensitivity

76.7 (73.3, 79.8)

77.3 (73.9, 80.4)

 Specificity

96.4 (96.1, 96.8)

89.9 (89.3, 90.5)

On 2015-year cohort: with more complete IC data available

 Sensitivity

70.5 (62.9, 77.1)

73.1 (65.6, 79.4)

 Specificity

97.1 (96.6, 97.6)

89.7 (88.7, 90.6)

  1. 95% confidence intervals are presented in parentheses
  2. aThe positive pool for sensitivity includes screen-detected positives and two-year ICs, which are relevant for HU
  3. bThe positive pool for sensitivity includes screen-detected positives and three-year ICs, which are relevant for the UK