TABLE 4.
Research summary of artificial intelligence in retinopathy of prematurity.
| Year | Country or region | Authors | Task | Dataset (disease images) | AI algorithm | Output |
|---|---|---|---|---|---|---|
| 2018 | America | Brown et al. (2018) | Diagnosis | 5,511 images (977 images) | U-Net, Inception version 1 | Sensitivity = 0.93, Specificity = 0.94, Accuracy = 0.91 |
| 2020 | America | Chen Q. et al. (2021) | Diagnosis | 10,894 images (1945 images) | Convolution neural network | AUROC = 0.99, AUPRC = 0.98, Sensitivity = 0.94 |
| 2020 | China | Mao et al. (2020) | Diagnosis | 3,311 images (1,393 images) | U-Net, Dense Net | Specificity = 0.978, Sensitivity = 0.951 |
| 2022 | China | Peng et al. (2022) | Diagnosis | 8,733 images (3,684 images) | DenseNet121 | Accuracy = 0.9776, Recall = 0.9714, Precision = 0.9835, F1-score = 0.9774, Kappa = 0.9552 |
| 2020 | China | Tong et al. (2020) | Classification | 36,231 images (36,231 images) | ResNet, Faster region-based convolutional neural network | Accuracy = 0.903, Sensitivity = 0.778, Specificity = 0.932, F1 score = 0.761 |
| 2021 | China | Peng et al. (2021) | Classification | 635 images (332 images) | ResNet18, DenseNet121, EfficientNetB2 | Recall = 0.9055, Precision = 0.9092, F1 score = 0.9043, Accuracy = 0.9827, Kappa = 0.9786 |
| 2020 | Taiwan | Huang et al. (2021) | Classification | 11,372 images (1,279 images) | Convolution neural network | Accuracy = 0.9223, Sensitivity = 0.9614, Specificity = 0.9595, Sensitivity and Specificity of stage 1 ROP = 0.9182, 0.9450, Sensitivity and Specificity of stage 2 ROP = 0.8981,0.9899 |
| 2022 | China | Li and Liu (2022) | Classification | 18,827 images (3,869 images) | U-Net, Dense Net | Sensitivity of diagnosing = 0.9593, Specificity of diagnosing = 0.9929, Sensitivity and Specificity of stage 1 ROP = 0.9021, 0.9767, Sensitivity and Specificity of stage 2 ROP = 0.9275,0.9874, Sensitivity and Specificity of stage 3 ROP = 0.9184,0.9929 |
| 2021 | India | Agrawal et al. (2021) | Evaluation | 4,250 images (2,350 images) | U-Net, Circle Hough Transform | Accuracy = 0.98 |
| 2022 | China | Wu et al. (2022) | Evaluation | 7,796 images (1984 images) | OC-Net, SE-Net | AUC, Accuracy, Sensitivity and Specificity of OC-Net = 0.94,0.333,1.00, and 0.075, respectively |
| AUC, Accuracy, Sensitivity and Specificity of SE-Net = 0.88, 0.560, 1.00, and 0.353, respectively |