Research Article
Exudate Detection for Diabetic Retinopathy Using Pretrained Convolutional Neural Networks
Table 2
Comparative classification accuracy results of the proposed model with individual CNN models for exudate detection using DIARETDB1 dataset.
| CNN models | Data splitting | Training (%) | Testing (%) | F1 score | Recall | Precision | Classification accuracy (%) |
| Inception-v3 | 70 | 30 | 0.95 | 0.98 | 0.93 | 93.10 | 80 | 20 | 0.93 | 0.94 | 0.92 | 93.30 | 90 | 10 | 0.95 | 0.94 | 0.97 | 93.57 |
| ResNet-50 | 70 | 30 | 0.95 | 0.99 | 0.92 | 90.57 | 80 | 20 | 0.94 | 0.98 | 0.90 | 96.10 | 90 | 10 | 0.98 | 0.98 | 0.98 | 97.90 |
| VGG-19 | 70 | 30 | 0.94 | 0.98 | 0.90 | 93.12 | 80 | 20 | 0.93 | 0.92 | 0.95 | 95.50 | 90 | 10 | 0.91 | 0.93 | 0.90 | 93.76 |
| Proposed model | 70 | 30 | 0.96 | 0.97 | 0.96 | 98.72 | 80 | 20 | 0.95 | 0.96 | 0.95 | 98.91 | 90 | 10 | 0.96 | 0.96 | 0.96 | 97.92 |
|
|