Research Article
Multiview Volume and Temporal Difference Network for Angle-Closure Glaucoma Screening from AS-OCT Videos
Table 2
Performance of different subnetworks on the private Casia video dataset.
| 3D CNN (B-Acc/F1 score) | Splits | 18-Layer | 34-Layer | 50-Layer | 101-Layer | 152-Layer |
| 5% | 0.638/0.632 | 0.464/0.282 | 0.625/0.627 | 0.562/0.430 | 0.558/0.463 | 10% | 0.692/0.695 | 0.518/0.463 | 0.612/0.589 | 0.594/0.487 | 0.562/0.430 | 20% | 0.589/0.589 | 0.482/0.437 | 0.509/0.492 | 0.589/0.514 | 0.562/0.430 |
| CNN-LSTM (B-Acc/F1 score) | Splits | 18-Layer | 34-Layer | 50-Layer | 101-Layer | 152-Layer |
| 5% | 0.531/0.367 | 0.643/0.614 | 0.777/0.763 | 0.607/0.562 | 0.719/0.678 | 10% | 0.500/0.371 | 0.679/0.662 | 0.781/0.789 | 0.714/0.707 | 0.656/0.589 | 20% | 0.500/0.297 | 0.754/0.757 | 0.710/0.695 | 0.714/0.707 | 0.571/0.505 |
|
|
The bold values indicate the optimal results.
|