3D Vessel Segmentation on Cerebral 3D OCTA (test)
79.46Dice (all)3D U-Net (Upper Bound)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| 3D U-Net (Upper Bound)Training Data=real (ours), Annotation=Manual (2 months effort)2024.03 | 79.46 | 85.29 | 74.31 | 87.86 | 82.54 | 70.93 | |
| 3D U-Net (SimLTA)Training Data=syn. (simLTA), Artifacts=Local noise, Tail, Angle-dependent loss2024.03 | 74.83 | 80.92 | 66.9 | 81.27 | 80.66 | 69.19 | |
| 3D U-Net (SimLTAC)Training Data=syn. (simLTAC), Artifacts=Local noise, Tail, Angle-dependent loss, Curvature2024.03 | 74.46 | 80.84 | 66.5 | 80.75 | 80.33 | 69.55 | |
| 3D U-Net (Ablation: simLT)Training Data=syn. (simLT), Artifacts=Local noise, Tail2024.03 | 70.38 | 72.8 | 57.5 | 73.6 | 79.81 | 59.48 | |
| 3D U-Net (Vessel Graph Ablation)Training Data=syn. (sim_graph), Context=Ablation on vessel graphs2024.03 | 60 | 74.48 | 66.77 | 80.88 | 51.84 | 52.92 | |
| 3D U-Net ([21])Training Data=real ([21])2024.03 | 54.13 | 65.41 | 52.78 | 67.57 | 55.07 | 48.5 | |
| 3D U-Net (Ablation: simL)Training Data=syn. (simL), Artifacts=Local noise2024.03 | 52.68 | 46.15 | 56.67 | 61.34 | 47.61 | 28.77 | |
| 3D U-Net (Ablation: sim)Training Data=syn. (sim), Artifacts=None2024.03 | 50.85 | 29.98 | 54.26 | 62.63 | 47.33 | 9.87 | |
| OtsuType=Traditional technique, Supervision=None2024.03 | 50.62 | 33.47 | 42.99 | 49.34 | 51.63 | 11.2 | |
| FrangiType=Traditional technique, Supervision=None2024.03 | 40.84 | 53.13 | 58.52 | 65.96 | 19.57 | 30.95 |