Table 1.
Summary of representative works of AI in ophthalmology.
| Year | Reference | Topic | Dataset | AUC | SE | SP |
|---|---|---|---|---|---|---|
| 2016 | Abramoff et al.14 | DR | CFPs, 1748 images | 0.980 | 96.8% | 87.0% |
| 2017 | Gargeya et al.15 | DR | CFPs, 75,137 images | 0.97 | 94% | 98% |
| 2017 | Ting et al.16 | DR, glaucoma, and AMD | CFPs, 494,661 images | 0.931–0.958 | 90.5%–100% | 87.2%–91.6% |
| 2020 | Pan et al.17 | DR lesions classification | FFAs, 4067 images | 0.870–0.965 | 79.7%–98.0% | 82.7%–99.5% |
| 2022 | Gao et al.18 | DR grading | FFAs, 11,214 images | 0.922–0.994 | / | / |
| 2020 | Varadarajan et al.20 | DME OCT-grading | CFPs, 7072 images | 0.89 | 85% | 80% |
| 2019 | Liu et al.23 | Glaucoma | CFPs, 274,413 images | 0.996 | 96.2% | 97.7% |
| 2021 | Sun et al.28 | Glaucoma | OCTs, 777 sets | 0.957 | 89.6% | 95.2% |
| 2020 | Thompson et al.29 | Glaucoma | OCTs, 20,806 images | 0.96 | 81% | 95% |
| 2019 | Phene et al.33 | Glaucoma | CFPs, 86,618 images | 0.945 | 80.0% | 90.2% |
| 2019 | Asaoka et al.34 | Glaucoma | OCTs, 4316 images | 0.937 | 82.5% | 93.9% |
| 2019 | Wen et al.37 | Forecast future Visual Fields | VFs, 32,443 VFs | / | / | / |
| 2022 | Huang et al.38 | Glaucoma visual field grading | VFs, 16,356 VFs | 0.93 | / | / |
| 2017 | Burlina et al.39 | AMD | CFPs, 213,997 iamges | 0.96 | 88.4% | 94.1% |
| 2019 | Peng et al.40 | AMD grading | CFPs, 59,302 images | 0.94 | 59.0% | 93.0% |
| 2020 | Yan et al.42 | AMD progression prediction | CFPs, 31,262 images | 0.85 | / | / |
| 2018 | Kermany et al.43 | AMD | OCTs, 207,130 images | 0.999 | 97.8% | 97.4% |
| 2020 | Erfurth et al.45 | Quantification of Fluid Volumes to AMD | OCTs, 24,362 images | / | / | / |
| 2021 | Yan et al.47 | AMD | OCTs, 56,091 images | 0.940–0.992 | 80.0%–96.5% | / |
| 2021 | Zhang et al.48 | Detect and quantify geographic atrophy | OCTs, 5049 images | / | / | / |
| 2021 | Xu et al.50 | AMD,PCV | CFPs and OCTs, 1099 CFPs, 821 OCTs | 0.939 | 88.8% | 95.6% |
| 2022 | Jin et al.51 | Identification of choroidal neovascularization activity in AMD | OCTs and OCTAs, 462 image pairs | 0.980 | 89.6% | 95.6% |
| 2011 | Chueng et al.52 | Cataract | Slit-lamp photographs, 5547 images | / | / | / |
| 2021 | Li et al.54 | Keratitis | Slit-lamp photographs, 6567 images | 0.998 | 97.7% | 98.2% |
| 2021 | Ye et al.55 | Myopic maculopathy | OCTs, 2342 images | 0.927–0.974 | 73.9%–92.8% | 84.8%–94.0% |
| 2019 | Yoo et al.56 | Identify candidate patients for corneal refractive surgery | Multi-instrument data, 13,201 subjects | 0.983 | / | / |
| 2020 | WANG et al.57 | Eyelid malignant melanoma | Pathological patches, 225,230 images | 0.989 | 94.7% | 95.3% |