Cardiac MRI Segmentation on M&Ms Target Domain B Philips 192 volumes
90.92Dice LVUpper Bound
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Upper BoundBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=Joint Training, Input Resolution=256x256, Training/Pre-trained Protocol=Joint training on source and target domains2026.04 | 90.92 | 85.1 | 86.47 | 1.72 | 1.55 | 2.08 | |
| CRISPBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=Domain Generalization (no target access), Input Resolution=256x2562026.04 | 89.57 | 82.74 | 86.49 | 1.8 | 1.57 | 1.57 | |
| DDG-MedBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=Domain Generalization (no target access), Input Resolution=256x2562026.04 | 89.01 | 80.15 | 84.81 | 2.43 | 2.23 | 3.03 | |
| TEGDABackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=TTA (update params with target domain), Input Resolution=256x2562026.04 | 88.98 | 83.21 | 85.52 | 1.88 | 1.53 | 1.86 | |
| IPLCBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=SFDA (update params with target domain), Input Resolution=256x2562026.04 | 88.5 | 83.27 | 83.4 | 1.74 | 1.86 | 2.05 | |
| FedDGBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=Domain Generalization (no target access), Input Resolution=256x2562026.04 | 86.97 | 79.25 | 85.05 | 2.47 | 2.17 | 3.65 | |
| BaselineBackbone=DeepLabv3+ (MobileNetV2), Domain Adaptation/Generalization Protocol=Directly tested source model, Input Resolution=256x256, Training/Pre-trained Protocol=Pre-trained on source Domain A2026.04 | 85.22 | 75.91 | 78.25 | 2.64 | 2.15 | 3.8 |