DR Staging on in-house OCTA
95.8Precision (H)Qwen3-VL-8b-SVR-FT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen3-VL-8b-SVR-FTparameters=8b, logic=Synthetic Vasculature Reasoning, fine-tuning=FT2025.12 | 95.8 | 41.1 | 41.8 | 84 | 48.8 | 73.7 | 68.8 | |
| GNNbackbone=GNN2025.12 | 95 | 32.6 | 45.6 | 72 | 59.4 | 77.5 | 69.7 | |
| ResNet18backbone=ResNet182025.12 | 94.3 | 33.5 | 42.6 | 79.3 | 54.4 | 56.3 | 63.3 | |
| Qwen3-VL-8b-SVRparameters=8b, logic=Synthetic Vasculature Reasoning2025.12 | 92.8 | 29.5 | 35.7 | 78.6 | 45.3 | 51 | 58.3 | |
| Qwen3-VL-8b-FTparameters=8b, fine-tuning=FT2025.12 | 90.8 | 36.4 | 24.5 | 82.8 | 63.6 | 7.3 | 51.2 | |
| Qwen3-VL-8b-GFTparameters=8b, fine-tuning=GFT2025.12 | 88.9 | 53.5 | 45.4 | 88.1 | 50.2 | 55 | 64.5 | |
| Qwen3-VL-30bparameters=30b2025.12 | 78.7 | 22.8 | 44.4 | 84.9 | 24.1 | 4.1 | 37.7 | |
| Qwen3-VL-8bparameters=8b2025.12 | 77.5 | 19.8 | 6.5 | 57.6 | 16.3 | 25.5 | 33.1 |