Table 2.
Scenarios | Selection rate of AI and radiologists/setup based on AI score | Consensus, n (% of all examinations) | Recall, n (% of all examinations) | Screen-detected cancer, n (n per 1000) | Potential interval cancers detected as SDC among consensus and/or recall cases, n (n per 1000)∞ | Potential rate of SDC (per 1000) with 95% confidence interval§ | Potential rate of IC (per 1000) with 95% confidence interval§ |
---|---|---|---|---|---|---|---|
Independent double reading | 10,790 (8.8) | 3896 (3.2) | 752 (6.1) | 0 | – | 1.7 (n = 205)€ | |
1 | AI and R1 selects 5.8% each* | 12,724 (10.4) | 3357 (2.7) | 721 (5.9) | 77 (0.6) | 6.5 (6.1–7.0) | 1.0 (0.9–1.2) |
2 | AI selects 10.1% (AI score = 10) and R1 selects 5.8%* | 17,394 (14.2) | 3412 (2.8) | 729 (5.9) | 100 (0.8) | 6.7 (6.3–7.2) | 0.9 (0.7–1.0) |
3 |
AI score 1–5: negative AI score 6–10: R1+R2 |
7955 (6.5) | 3201 (2.6) | 735 (6.0) | 45 (0.4) | 6.3 (5.9–6.8) | 1.3 (1.1–1.5) |
4 |
AI score 1–7: negative AI score 8–10: R1+R2 |
5894 (5.0) | 2564 (2.1) | 719 (5.8) | 43 (0.3) | 6.2 (5.8–6.7) | 1.3 (1.1–1.5) |
5 |
AI score 1–9: negative AI score 10: R1+R2 |
2805 (2.3) | 1437 (1.2) | 653 (5.3) | 33 (0.3) | 5.6 (5.2–6.0) | 1.4 (1.2–1.6) |
6 |
AI score 1–5: negative AI score 6–10: R1 |
5301 (4.3) | 2669 (2.2) | 651 (5.3) | 28 (0.2) | 5.5 (5.1–6.0) | 1.4 (1.2–1.7) |
7 |
AI score 1–5: R1 AI score 6–10: R1+R2 |
9683 (7.9) | 3749 (3.1) | 747 (6.1) | 46 (0.4) | 6.4 (6.0–6.9) | 1.3 (1.1–1.5) |
8 |
AI score 1–7: R1 AI score 8–10: R1+R2 |
8880 (7.2) | 3614 (2.9) | 746 (6.1) | 45 (0.4) | 6.4 (6.0–6.9) | 1.3 (1.1–1.5) |
9 |
AI score 1–5: negative AI score 6–7.5: R1 AI score 7.6–10: R1+R2 |
6934 (5.6) | 3028 (2.5) | 731 (5.9) | 44 (0.4) | 6.3 (5.9–6.8) | 1.3 (1.1–1.5) |
10 |
AI score 1–5: negative AI score 6–7.5: R1+R2 AI score 7.6–10: R1 |
6322 (5.1) | 2842 (2.3) | 655 (5.3) | 29 (0.2) | 5.6 (5.2–6.0) | 1.4 (1.2–1.7) |
11 | 3.2%* | - | 3896 (3.2) | 555 (4.5) | 44 (0.4) | 4.9 (4.5–5.3) | 1.3 (1.1–1.5) |
*5.8% mimics the average selection rate of the individual radiologists in the study sample, and 3.2% mimics the recall rate in the study sample after independent double reading
∞Number of interval cancers among consensus cases in Scenario 1–10. Number of interval cancers among cases that AI as a standalone system selected to be recalled in scenario 11
§If all IC with prior exams among consensus and/or recall cases were present at screening and detected as screen-detected and not interval cancers
€The actual rate and number of interval cancers in the study sample