Table 3.
Prospective validation studies of AI in DR
| Algorithm | References | Population | Patients (n) | Dilation | DR grading system | Detection | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|---|---|---|---|---|---|
| Conventional retinal photography cameras | ||||||||||
| iGrading | Philip et al. [68] | Primary care. Scotland | 6722 | As needed | Scottish DR Grading Scheme | Any DR | 90.5% (89.3–91.6) | 67.4% (66.0–68.8) | N/A | N/A |
| Soto-Pedre et al. [69] | Primary care, Spain | 5253 | Yes | MA detection | Any DR | 94.52% (92.6–96.5) | 68.77% (67.2–70.4) | 34.1% (31.7–36.5) | 98.66% (89.2–99.2) | |
| Bosch | Bawankar et al. [59] | India | 560 | No | AAO PPP 2019 | Any DR | 91.2% (86.4–94.7) | 96.9% (94.5–98.5) | 94.4% (90.4–96.8) | 95.0% (92.5–96.8) |
| IDx-DR X2.1 | Abramoff et al. [38] | Ten primary care sites, USA | 819 | As needed | ETDRS | mtmDR | 87.2% (81.8–91.2) | 90.7% (88.3–92.7) | N/A | N/A |
| rDR | 99.3% (96.1–99.9) | 68.8% (61.5–76.2) | 74.6% (68.4–80.8) | 99.1% (97.2–100) | ||||||
| VTDR | 99.1% (95.1–99.9) | 80.4% (73.9–85.9) | 75.3% (68.4–82.3) | 99.3% (96.3–100) | ||||||
| IDx-DR 2.0 | van der Heijden et al. [39] | Hoorn DCS center, The Netherlands | 898 | As needed | ICDR | rDR | 68% (56–79) | 86% (84–88) | 30% (24–38) | 97% (95–98) |
| ICDR | VTDR | 62% (32–85) | 95% (93–96) | 14% (7–27) | 99% (99–100) | |||||
| EURODIAB | rDR | 91% (69–98) | 84% (81–96) | 12% (8–18) | 100% (99–100) | |||||
| EURODIAB | VTDR | 64% (36–86) | 95% (93–96) | 16% (8–29) | 99% (99–100) | |||||
| SELENA + | Bellemo et al. [42] | Zambia, Africa | 1574 | N/A | ICDR | rDR | 92.3% (90.1–94.1) | 89.0% (87.9–90.3) | N/A | N/A |
| VTDR | 99.4% (99.2–99.7) | N/A | N/A | N/A | ||||||
| DME | 97.2% (96.6–97.8) | N/A | N/A | N/A | ||||||
| VoxelCloud Retina | Zhang et al. [48] | Nationwide DR screening, China | 15,805 | No | ICDR | rDR | 83.3% (81.9–84.6) | 92.5% (92.1–92.9) | 61.8% (60.3–63.3) | 97.4% (97.2–97.7) |
| DLA | Scheetz et al. [47] | Australia | 203 | No | NHS | rDR | 96.9% (83.8–99.9) | 87.7% (81.8–92.2) | 59.6% (45.1–73.0) | 99.3% (96.4–100) |
| ARDA | Ruamviboonsuk et al. [43] | Nationwide DR screening, Thailand | 7651 | As needed | ICDR | VTDR | 91.3% (85.1–97.4) | 96.3% (95.1–97.4) | 79.2% (73.8–84.3) | 95.5% (92.8–97.9) |
| EyeArt v2.1 | Heydon et al. [40] | NHS DESP, UK | 30,405 | N/A | ETDRS | rDR | 95.7% (94.8–96.5) | 54.0% (53.4–54.5) | N/A | N/A |
| EyeArt v2.1 | Ipp et al. [41] | Multicenter, USA | 893 | No | ETDRS | mtmDR | 95.5% (92.4–98.5) | 85.0% (82.6–87.4) | 59.5% (53.9–63.9) | 98.8% (98.2–99.4) |
| Yes | mtmDR | 95.5% (92.6–98.4) | 85.3% (83.0–87.5) | 59.1% (53.8–64.4) | 98.8% (98.2–99.5) | |||||
| No | VTDR | 95.1% (90.1–100) | 89.0% (87.0–91.1) | 26.7% (19.5–33.0) | 99.8% (99.5–100) | |||||
| Yes | VTDR | 95.2% (90.4–100) | 89.5% (87.6–91.4) | 26.1% (19.6–32.6) | 99.8% (99.5–100) | |||||
| Smartphone-based retinal photography | ||||||||||
| EyeArt v2.1 (FOP) | Rajalakshmi et al. [49] | Tertiary care hospital, India | 301 | Yes | ICDR | Any DR | 95.8% (92.9–98.7) | 80.2% (72.6–87.8) | 89.7% (85.5–93.8) | 91.4% (85.7–97.1) |
| EyeArt v2.0 (FOP) | Kim et al. [51] | Two university hospitals, USA | 72 | Yes | ICDR | rDR | 77.8% (67.3–85.7) | 71.5% (48.7–86.9) | N/A | N/A |
| Remidio | Natarajan et al. [50] | Dispensaries in Mumbai, India | 213 | Yes | ICDR | Any DR | 85.2% (66.3–95.8) | 92.0% (97.1–95.4) | N/A | N/A |
| rDR | 100% (78.2–100) | 88.4% (83.2–92.5) | N/A | N/A | ||||||
AAO PPP American Academy of Ophthalmology Preferred Practice Pattern, DCS diabetic care system, DESP Diabetic Eye Screening Programme, ETDRS Early Treatment Diabetic Retinopathy Study Severity Scale, MA microaneurysm, mtmDR more than mild diabetic retinopathy, NHS National Health Screening guidelines