Table 1 Comparison of the current reader system and the AI-integrated scenarios by the screening outcomes and workload

From: Comparison of AI-integrated pathways with human-AI interaction in population mammographic screening for breast cancer

Variable

Current reader system

AI reader-replacement

AI band-pass

AI triage

Sensitivity (%)

79.8

82.3 (81.5, 83.1)

81.7

77.2

Specificity (%)

96.0

96.3 (96.2, 96.3)

96.6

95.7

No. of episodes

149,105

149,105

149,105

149,105

True positive

1061

1094 (+3.1%)

1086 (+2.4%)

1026 (−3.3%)

True negative

141,915

142,269 (+0.2%)

142,694 (+0.5%)

141,490 (−0.3%)

False positive

5861

5507 (−6%)

5082 (−13.3%)

6286 (+7.3%)

False-negative

268

235 (−12.4%)

243 (−9.3%)

303 (+13.1%)

Workload

    

Assessments

6922

6602 (−4.6%)

6168 (−10.9%)

7312 (+5.6%)

Human reads

30,8091

160,132 (-48%)

59,453 (−80.7%)

166,292 (−46%)

*Third reads

9881

11,027 (+11.6%)

3173 (−67.9%)

2276 (−77%)

  1. *The third reads are part of the human reads. A separate entry is presented to show the workload impact on the third reader.
  2. For the AI reader-replacement scenario, mean values based on 1000 simulations are presented (rounded to the nearest integer, and the 95% confidence intervals (CI) are presented for the sensitivity and specificity). The AI band-pass and AI triage scenarios do not require simulating the third reader, and the results are entirely based on the real data (i.e. no CI). Percentages in the brackets with the plus/minus sign indicate the percentage change compared with the current reader system.
Back to article page