Table 5.
Summary of the machine and deep learning segmentation approaches.
| Study | Method | Dataset | Segmentation Type | Performance | Modality |
|---|---|---|---|---|---|
| Yan et al. [74] | Encoder-decoder Network | STARE, DRIVE | Drusen segmentation | Accuracy: 97.13%, Sensitivity: 92.02%, Specificity: 97.30% | CFP |
| Pham et al. [67] | DeeplabV3,U-Net | Kangbuk Samsung hospital, STARE | Drusen segmentation | Accuracy: 0.99, 0.981%, Sensitivity: 0.662, 0.588%, Specificity: 0.997, 0.991% Dice Score: 0.625, 0.542 | CFP |
| Chen et al. [75] | Faster R-CNN | RETOUCH | Cysts/fluid | Accuracy: 0.665%, Dice Score: 0.997 | OCT |
| Schlegl et al. [66] | FCN | 1200 volumes OCT | Cysts/fluid | Sensitivity: IRC 0.84, SRF 0.81%, Precision: IRC 0.91, SRF 0.61, Area under the curve (AUC): IRC 0.94, SRF 0.92 | OCT |
| Sappa et al. [76] | RetFluidNet | 124 volumes OCT | Cysts/fluid | Accuracy: IRF 80.05, PED 92.74, SRF 95.53%, Dice Score: 0.885 | OCT |
| Kang et al. [77] | U-Net | RETOUCH | Cysts/fluid | Accuracy: 0.968%, Dice Score: 0.9 | OCT |
| Liu et al. [78] | FCN | RETOUCH | Cysts/fluid | Dice Score: 0.744 | OCT |
| Tennakoon et al. [79] | U-Net | RETOUCH | Cysts/fluid | Dice Score: 0.737 | OCT |
| Diao et al. [80] | CM-CNN, (CAM-UNet) | Heidelberg Engineering, Germany | Drusen and CNV lesions | Dice Score: 77.51% | OCT |