Table 3.
Subgroup analysis showing the performance of the artificial intelligence for the detection of AMD.
| Variables | No. of study | AUC (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | LR+ (95% CI) | LR- (95% CI) | DOR (95% CI) | Heterogeneity for DOR |
|
|---|---|---|---|---|---|---|---|---|---|
| I2,% | P value | ||||||||
| Methods | |||||||||
| CNN | 9 | 0.980 (0.975–0.985) | 0.88 (0.88–0.88) | 0.90 (0.90–0.91) | 16.0 (11.1–23.1) | 0.07 (0.06–0.10) | 225 (123–409) | 99.7 | <0.001 |
| SVM | 3 | 0.994 (0.988–1.000) | 0.94 (0.92–0.96) | 0.97 (0.95–0.99) | 30.9 (11.9–79.8) | 0.04 (0.01–0.16) | 917 (97–8861) | 71.4 | 0.030 |
| Images for validation | |||||||||
| <500 | 3 | 0.997 (0.995–1.000) | 0.99 (0.96–1.00) | 0.99 (0.97–1.00) | 54.6 (13.9–215.0) | 0.02 (0.01–0.07) | 2656 (286–24,635) | 42.0 | 0.178 |
| 500–5000 | 4 | 0.982 (0.970–0.994) | 0.93 (0.92–0.94) | 0.93 (0.93–0.94) | 16.3 (5.7–46.8) | 0.07 (0.03–0.13) | 252 (50–1274) | 98.1 | <0.001 |
| >5000 | 6 | 0.980 (0.974–0.985) | 0.88 (0.88–0.88) | 0.90 (0.90–0.91) | 17.0 (10.9–26.5) | 0.08 (0.06–0.11) | 216 (105–445) | 99.8 | <0.001 |
| Outcomes | |||||||||
| AMDa | 4 | 0.993 (0.984–1.000) | 0.92 (0.90–0.93) | 0.85 (0.83–0.87) | 29.2 (3.4–248.4) | 0.04 (0.01–0.14) | 853 (39–18,403) | 88.2 | <0.001 |
| Referable AMDb | 6 | 0.983 (0.978–0.988) | 0.88 (0.88–0.88) | 0.90 (0.90–0.90) | 15.8 (10.2–24.3) | 0.06 (0.05–0.08) | 276 (132–579) | 99.8 | <0.001 |
| Publication year | |||||||||
| Before 2015 | 4 | 0.989 (0.985–0.993) | 0.95 (0.94–0.96) | 0.96 (0.95–0.97) | 22.7 (16.9–30.3) | 0.05 (0.03–0.10) | 474 (197–1142) | 58.1 | 0.067 |
| After 2015 | 9 | 0.980 (0.975–0.985) | 0.88 (0.88–0.88) | 0.90 (0.90–0.91) | 15.9 (10.8–23.3) | 0.08 (0.06–0.10) | 224 (120–418) | 99.7 | <0.001 |
| Regions | |||||||||
| Asian countries | 3 | 0.979 (0.970–0.988) | 0.93 (0.91–0.94) | 0.89 (0.88–0.89) | 20.6 (2.6–163.5) | 0.08 (0.06–0.11) | 212 (73–613) | 77.9 | 0.011 |
| The western countries | 10 | 0.984 (0.980–0.988) | 0.88 (0.88–0.88) | 0.91 (0.91–0.91) | 19.0 (12.0–30.2) | 0.07 (0.05–0.09) | 288 (155–536) | 99.7 | <0.001 |
AUC: area under curve, CI: confidence interval, LR+: positive likelihood ratio, LR-: negative likelihood ratio, DOR: diagnostic odds ratio, CNN: convolutional neural networks, SVM: support vector machine.
a
Studies detecting dry-AMD only or wet-AMD only were excluded in this analysis.
b
Referable AMD was defined as intermediate and advanced AMD.