From: Potential applications of artificial intelligence in image analysis in cornea diseases: a review
Year | Authors | Imaging modality | Sample size (eyes) | Study population | Outcome measures | AI algorithms | Diagnostic performance | Validation method |
---|---|---|---|---|---|---|---|---|
Pterygium | ||||||||
 2023 | Liu et al. [36] | ASP | 20,987 | Dataset of smartphone and slit-lamp eye images | Detection and segmentation of pterygium | SA-CNN, SRU-Net | Detection Acc: 95.24% Segment Acc: 89.81%, Sens: 87.09% Spec: 96.68%, AUC: 0.9295 | Hold-out validation |
 2022 | Wan et al. [123] | ASP | 489 | Healthy and pterygium eyes | Segmentation and measuring of pterygium | U-Net +  + CNN | Dice: 0.902–0.962 κ: 0.918 | Hold-out validation |
 2022 | Fang et al. [42] | ASP | 6311 | SEED eye images; healthy and pterygium eyes | Detection of pterygium | CNN | AUC: 0.995, Sens: 98.6% Spec: 99.0% | Cross validation |
 2022 | Hung et al. [45] | ASP | 237 | Healthy and pterygium eyes | Detection of pterygium, and prediction of recurrence post-excision | Deep learning network | Pterygium detection – Spec: 91.7% – 100%, Sens: 80% – 91.7% Prediction of recurrence – Spec: 81.8%, Sens: 66.7% | Hold-out validation |
 2022 | Zhu et al. [124] | ASP | 300 | Healthy and pterygium eyes | Detection and segmentation of pterygium | Multiple CNNs | Acc: 99.0%, κ: 0.98 Sens: 98.7%, Spec: 99.3% | Hold-out validation |
 2021 | Jais et al. [125] | BCVA | 93 | Pterygium patients undergoing surgery | Prediction of BCVA improvement post-pterygium surgery | SVM, NBC, decision tree, logistic regression | Acc: 94.44%, Spec: 100% Sens: 92.14% | Cross validation |
 2019 | Zulkifley et al. [126] | ASP | 120 | Healthy and pterygium eyes | Detection of pterygium | CNN | Acc: 81.1%, Sens: 95.0% Spec: 98.3% | Cross validation |
 2019 | Lopez et al. [127] | ASP | 3017 | Healthy and pterygium eyes | Detection of pterygium | CNN | AUC: 0.99, Acc: 93.5% Sens: 88.2% | N.A |
Infectious keratitis | ||||||||
 2023 | Essalat et al. [37] | IVCM | 1001 | IVCM-Keratitis dataset images | IK detection | Densenet161 CNN | Acc: 93.55%, Precis: 92.52% Sens: 94.77% | Cross validation |
 2023 | Liang et al. [36] | IVCM | 7278 | Dataset of FK eye images | FK detection | GoogLeNet, VGGNet CNNs | Acc: 97.73%, Sens: 97.02% Spec: 98.54% | Hold-out validation |
 2023 | Wei et al. [128] | ASP | 420 | FK, BK, AK and VK eyes | FK detection and discriminating | Binary logistic regression, decision tree classification, RF | AUC: 0.859–0.916 | Cross validation |
 2022 | Natarajan et al. [31] | ASP | 285 | HSV VK and BK eyes | VK detection | DenseNet CNN | Acc: 72%, AUC: 0.73 Sens: 69.6%, Spec: 76.5% | Hold-out validation |
 2022 | Redd et al. [129] | ASP | 980 | FK and BK eyes | FK discriminating from BK | MobileNet CNN | AUC: 0.86 | Cross validation |
 2022 | Ghosh et al. [35] | ASP | 194 | FK and BK eyes | FK discriminating from BK | VGG19, ResNet50, DenseNet121 CNNs | Acc: 68.0%–78.0% AUC: 0.60–0.86 | Hold-out validation |
 2021 | Kuo et al. [130] | ASP | 1512 | Clinically suspected IK eye images | BK detection | Multiple CNNs | Sens: 74% Spec: 64% | Cross validation |
 2021 | Wang et al. [28] | ASP | 6073 | Healthy, FK, BK and HSV VK eye images | FK, BK, HSV VK detection | InceptionV3 | κ: 0.538 – 0.913 AUC: 0.860 – 0.959 | Hold-out validation |
 2021 | Koyama et al. [131] | ASP | 4306 | BK, AK, HSV VK eye images | BK, AK, HSV VK detection | Gradient boosting decision tree | Acc: 92.3%–97.9% AUC: 0.946–0.995 | Cross validation |
 2021 | Li et al. [132] | ASP | 6925 | Healthy and keratitis eye images | Keratitis detection | DenseNet121 CNN | AUC: 0.96 | Hold-out validation |
 2020 | Kuo et al. [133] | ASP | 288 | IK confirmed eyes | FK detection among IK | DenseNet CNN | Sens: 71%, Spec: 68% Acc: 70% | Cross validation |
 2020 | Liu et al. [134] | HRT-3 confocal microscopy | 1213 | Healthy and FK eyes | FK detection | CNN | Acc: 100%, Sens: 99.9% Spec: 100% | N.A |
 2018 | Wu et al. [135] | HRT-3 confocal microscopy | 378 | Healthy and FK eyes | Hyphae detection | KNN, SVM, linear regression, decision tree | AUC: 0.86–0.98 Acc: 81.7%–99.1% Sens: 78.5%–98.5% | Cross validation |