Table 5.
Per-Patient Performance Comparison with 95% Confidence Intervals for Readers, CAD Systems and in Combination.
| Study | Endpoint | Zone | Reader(s) Alone | CAD System Alone | Combination | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cut-Off | SN % | SP % | AUC | Chosen Threshold | SN % | SP % | AUC | Interaction | SN % | SP % | AUC | |||
| ROI Classification (ROI-C) | ||||||||||||||
| Bonekamp [21] |
GS ≥ 3 + 4 | WP | PI-RADSv2, ≥4 | 89 (76–96) |
43 (33–54) |
NR | Matched to reader SN in training set | 96 (85–99) |
51 (40–62) |
NR | NA | NA | NA | NA |
| Dinh [23] |
GS ≥ 3 + 4 | WP | Likert (1–5), ≥3 | 100 (100–100) |
9 (2–15) |
0.88 (0.68–0.96) |
CAD SN of 95% in training set | 100 (100–100) |
40
(28–51) |
0.95
(0.90–0.98) |
NA | NA | NA | NA |
| Woźnicki [31] | GS ≥ 3 + 4 | WP | PI-RADSv2, ≥4 | 91 (82–98) |
28 (13–46) |
NR | NR | NA | NA | NA | Radiomics model ensembled with PI-RADS, PSAd and DRE models | 91 (81–98) |
57 (38–74) |
NR |
| Lesion Localization and Classification (LL&C) | ||||||||||||||
| Gaur [34] |
GS ≥ 3 + 3 | WP | PI-RADSv2, ≥3 | 94 (91–96) |
45 (38–52) |
0.82 | NR | NA | NA | NA | CAD identified lesions reviewed by radiologist |
82
(75–88) |
72
(63–80) |
0.83 |
| Giannini [35] | GS ≥ 3 + 4 | WP | PI-RADSv2, ≥3 and max diameter ≥7 mm | 81 (70–90) |
75 (68–92) |
NR | NR |
96
(78–100) |
NR | NR | CAD identified lesions reviewed by radiologist |
91
(82–97) |
78 (71–85) |
NR |
| Greer [36] |
GS ≥ 3 + 3 | WP | PI-RADSv2, ≥3 | 91 (87–95) |
70 (62–79) |
0.88 (0.83–0.92) |
NR | NA | NA | NA | CAD identified lesions reviewed by radiologist | 90 (85–95) |
57
(47–66) |
0.85 (0.79–0.90) |
| Litjens [37] |
GS ≥ 3 + 4 * | WP | PI-RADSv1, ≥3 | ≈100 † | ≈52 † | NR | NA | NR | NR | 0.83 | NA | NA | NA | NA |
| Mehralivand [38] |
GS ≥ 3 + 4 | WP | PI-RADSv2, ≥3 | 82 | NR | 0.82 | NR | NA | NA | NA | CAD identified lesions reviewed by radiologist | 84 | NR | 0.78 |
| Schelb [39] |
GS ≥ 3 + 4 | WP | Mix of PI-RADSv1/v2, ≥3 | 96 (80–100) |
22 (10–39) |
NR | Point that most closely matched PI-RADS ≥3 performance in training set |
96 (80–100) |
31 (16–48) |
NR | NA | NA | NA | NA |
| Schelb [40] |
GS ≥ 3 + 4 | WP | PI-RADSv2, ≥3 | 98 (94–100) |
17 (11–24) |
NR | Iterative dynamic threshold that most closely matches PI-RADS ≥ 3 performance in most recent cases |
99 (95–100) |
24 (17–31) |
NR | NA | NA | NA | NA |
| Zhu [42] |
GS ≥ 3 + 4 | WP | PI-RADSv2, ≥3 | 84 (75–91) |
56 (43–69) |
0.83 (0.76–0.88) |
NR | NA | NA | NA | Radiologist reported with but not limited by CAD probability map |
93
(86–98) |
66 (53–77) |
0.89
(0.83–0.94) |
| Patient Classification (PAT-C) | ||||||||||||||
| Deniffel [43] | GS ≥ 3 + 4 | WP | PI-RADSv2, ≥3 and PSAd ≥0.15 ng/mL2 | 95 (84–100) |
35 (19–52) |
NR | CSPCa likelihood ≥ 0.2 | 100 (100–100) |
52 (32–68) |
0.85 (0.76–0.97) |
NA | NA | NA | NA |
(AUC—area under the receiver operating characteristic curve; CAD—computer-aided diagnosis; GS—Gleason score; NR—not reported; PI-RADS—Prostate Imaging-Reporting and Data System; ROI—region of interest; SN—sensitivity; SP—specificity; WP—whole prostate). Bold results indicate statistically significant differences to that of the reader(s) alone, p-value >0.05. * 3 + 4 vs. benign, Gleason 3 + 3 excluded. † Approximate values derived from study figures.