Table 2. Agreement Rates of the DLS and General Pathologists With the Subspecialists’ Majority Opinion at 3 Clinically Important Decision Cutoffsa.
| Clinical task, evaluation metric | % (95% CI) | |
|---|---|---|
| DLS | General pathologist | |
| Nontumor vs tumor determination (n = 752) | ||
| Agreement with subspecialist majority opinion | 94.3 (92.4-95.9) | 94.7 (92.8-96.3)b |
| Sensitivity | 95.5 (93.7-96.8)b | 92.8 (90.0-95.1) |
| Specificity | 91.7 (88.2-94.6) | 97.0 (95.1-98.6)b |
| Grading of tumor-containing biopsy specimensc | ||
| Agreement with subspecialist majority opinion for GG1 vs GG2-5 (n = 498) | 86.1 (83.1-89.2)b | 80.6 (77.9-83.5) |
| Agreement with subspecialist majority opinion for GG1-2 vs GG3-5 (n = 498) | 92.8 (90.8-94.9)b | 86.0 (83.2-88.5) |
Abbreviations: DLS, deep learning system; GG, grade group.
a
Similar to Figure 1, the agreement rate of the general pathologists represents the mean rate across all general pathologists.
b
The higher value in the row.
c
Agreement on 2 Gleason grading thresholds.