Table 3. Comparison of the AI-Unaided and AI-Aided Reader Studies in Terms of the AUC, Recall Rate, Sensitivity, and Specificity according to Each Radiologist and Average Radiologists.
| Radiologist | AI-Unaided Reading | AI-Aided Reading | P | |
|---|---|---|---|---|
| AUC* | ||||
| 1 | 0.76 [0.69–0.83] | 0.90 [0.85–0.95] | < 0.001 | |
| 2 | 0.78 [0.71–0.85] | 0.86 [0.80–0.92] | < 0.001 | |
| 3 | 0.85 [0.80–0.90] | 0.92 [0.88–0.97] | < 0.001 | |
| Average radiologists | 0.79 [0.74–0.85] | 0.89 [0.85–0.94] | < 0.001 | |
| Recall rate† | ||||
| 1 | 44.1 (350/793) [40.6–47.7] | 36.9 (293/793) [33.6–40.4] | < 0.001 | |
| 2 | 73.6 (584/793) [70.4–76.7] | 57.1 (453/793) [53.6–60.6] | < 0.001 | |
| 3 | 63.3 (502/793) [59.8–66.7] | 54.4 (431/793) [50.8–57.9] | < 0.001 | |
| Average radiologists | 60.4 [57.8–62.9] | 49.5 [46.5–52.4] | < 0.001 | |
| Sensitivity† | ||||
| 1 | 79.6 (43/54) [66.5–89.4] | 90.7 (49/54) [79.7–96.9] | 0.031 | |
| 2 | 92.6 (50/54) [82.1–97.9] | 92.6 (50/54) [82.1–97.9] | 1.000 | |
| 3 | 96.3 (52/54) [87.3–99.5] | 94.4 (51/54) [84.6–98.8] | 1.000 | |
| Average radiologists | 89.5 [83.1–95.9] | 92.6 [86.2–99.0] | 0.053 | |
| Specificity† | ||||
| 1 | 58.5 (432/739) [54.8–62.0] | 67.0 (495/739) [63.5–70.4] | < 0.001 | |
| 2 | 27.9 (206/739) [24.7–31.3] | 45.7 (338/739) [42.1–49.4] | < 0.001 | |
| 3 | 39.4 (291/739) [35.8–43.0] | 49.0 (362/739) [45.3–52.7] | < 0.001 | |
| Average radiologists | 41.9 [39.3–44.5] | 53.9 [50.9–56.9] | < 0.001 | |
*Numbers in brackets are the 95% confidence intervals of the AUC values, †Numbers are percentages, raw data are in parentheses, and 95% confidence intervals are in brackets. AI = artificial intelligence, AUC = area under the receiver operating characteristic curve