TABLE 2.
AI-related studies utilizing external ocular images.
| Authors | Study goals | Imaging data type | Dataset | Network model | Accuracy | AUC |
|---|---|---|---|---|---|---|
| Moriyama et al. (2006) | Eye motion tracking | External ocular images | — | Generative eye region model | — | — |
| Van Brummen et al. (2021) | Segmentation of regions such as iris and eyebrow | Photographs of periorbital areas | 418 images | ResNet-50 | — | — |
| Bahceci Simsek and Sirolu, (2021) | Evaluation of postoperative changes | Full-face photographs | 55 patients | DLIBML toolkit | — | — |
| Chen et al. (2021) | Measurement of eyelid paraments | External ocular images | 411 participants | MAIA software | — | — |
| Tabuchi et al. (2022) | Classification of images taken with a tablet device of patients with blepharoptosis diagnosis | Eyelid images | 1,276 images | Pre-trained MobileNetV2 CNNi | 0.828 | 0.900 |
| Hung et al. (2022) | Identification of monocular appearance photos of ptosis patients | External ocular images | 782 images | VGG-16 | 0.90 | 0.987 |
| Song et al. (2021b) | Determination of the choices of ptosis surgery strategies | External ocular images | 152 eyes | Gradient-boosted decision tree (GBDT) | 0.826 | 0.795 |
| Lou et al. (2021) | Evaluation of ptosis surgery outcome | External ocular images | 103 patients (135 ptotic eyes) | U-Net (Attention R2U-Net) | — | — |
| Yixin et al. (2022) | Exploration of the effect of eyelid on oculoplastic surgery and aesthetic outcomes | External ocular images | 64 patients | Multichannel CNN | 0.988 | — |
| Huang et al. (2022) | Diagnosis of TAO | Facial images | 3,120 eyes | ResNet-50 U-Net | Eye location: 0.980. Cornea: 0.930. Sclera segmentation: 0.870 | Over 0.850 |
| Li et al. (2022b) | Automatic detection of malignant eyelid tumors | External ocular images | Development set (n = 1,258). External test set (n = 309) | Faster-RCNN | — | AUCs ranged from 0.899 to 0.955 |
| Karlin J et al. (2022) | Detection of thyroid eye disease | External ocular images | Training set (n = 1994). Test set (n = 344) | ResNet-18 | 0.892 | |
| Yoo et al. (2020) | Synthesis of realistic postoperative appearance for orbital decompression surgery | External ocular images | 500 preoperative images and 500 postoperative images | Generative adversarial network (GAN) | — | 0.957 |