Table 8.
Comparison of the performance of our proposed model with previous studies of glaucoma detection combining structural and functional features.
| Studies | Method | Trained features | AUC | Accuracy (%) | Sensitivity (%) | Specificity (%) | No. of patients |
|---|---|---|---|---|---|---|---|
| Kim et al.22 | RF model | Age, IOP, CCT, Average RNFL thickness, GHT, MD and PSD | 0.98 | 98 | 98.3 | 97.5 | 202 healthy and 297 glaucoma |
| Brigatti et al.23 | NN (Back propagation) | MD, corrected loss, variance, short term, fluctuation, CDR, rim area, cup volume, and RNFL height | – | 88 | 90 | 84 | 54 healthy and 185 glaucoma |
| Bowd et al.24 | RVM | OCT RNFL thickness measurements, MD, and PSD | 0.85 | – | 81 | 72 | 69 healthy and 156 glaucoma |
| Grewal et al.25 | ANN | RNFL parameters on OCT, cup area, vertical CDR, cup volume, MD, loss variance, and GDx- Variable Corneal Compensation (VCC) parameters | – | – | 93.3 | 80 | 35 healthy and 35 glaucoma |
| Eliash et al.51 | SVM | Horizontal integrated rim width (HIRW), rim area, HCDR, vertical CDR, Mean NFL, NFL inferior, NFL superior, NFL 6, NFL 7, NFL 11, and MD | 0.98 | 96.6 | 97.9 | 92.5 | 47 healthy and 42 glaucoma |
| Proposed study | DL | MD, PSD, Average RNFL thickness and CDR | 0.98 | 97% (validation data), 96% (test data) | 100% (validation and test data) | 96% (validation data), 93% (test data) | 130 healthy and 125 glaucoma |