Research Article
Diagnosing Diabetic Retinopathy in OCTA Images Based on Multilevel Information Fusion Using a Deep Learning Framework
Table 3
The ablation experiment results of our classification model.
| | Input | Model | Accuracy |
| | Segmentation results | Concatenated convolution (ResNet50) | 75.2% () | | Isolated convolution | 78.9% () | | Isolated and concatenated convolution | 80.6% () |
| | OCTA images | Concatenated convolution (ResNet50) | 77.6% () | | Isolated convolution | 84.4% () | | Isolated and concatenated convolution | 87.8% () |
| | Merged images | Concatenated convolution (ResNet50) | 79.6% () | | Isolated convolution | 84.7% () | | Isolated and concatenated convolution | 88.1% () |
|
|
