MRI Reconstruction on COR-PD (test)
36.57PSNRCycMit-MRI + AT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CycMit-MRI + ATepsilon=0.012025.01 | 36.57 | 0.94 | |
| CycMit-MRI + MoDLInput Defense=Proposed (CycMit-MRI), Base Model=MoDL2025.01 | 35.14 | 92.1 | |
| CycMit-MRI + MoDLepsilon=0.012025.01 | 35.14 | 0.92 | |
| CycMit-MRI + SMUGepsilon=0.012025.01 | 34.85 | 0.92 | |
| Proposed Method + MoDLAttack Domain=k-space, Attack Type=l2, Backbone=MoDL2025.01 | 34.21 | 0.89 | |
| ATepsilon=0.012025.01 | 33.99 | 0.92 | |
| CycMit-MRI + ATepsilon=0.022025.01 | 33.42 | 0.93 | |
| Adversarial TrainingAttack Domain=k-space, Attack Type=l22025.01 | 33.37 | 0.88 | |
| CycMit-MRI + MoDLepsilon=0.022025.01 | 33.25 | 0.91 | |
| CycMit-MRI + SMUGepsilon=0.022025.01 | 32.97 | 0.91 | |
| ATepsilon=0.022025.01 | 30.91 | 0.88 | |
| Input RS + MoDLInput Defense=Randomized Smoothing (RS), Base Model=MoDL2025.01 | 30.71 | 85.8 | |
| SMUGepsilon=0.012025.01 | 28.22 | 0.79 | |
| SMUGepsilon=0.022025.01 | 21.86 | 0.61 | |
| Input JPEG + MoDLInput Defense=JPEG Compression, Base Model=MoDL2025.01 | 17.29 | 50.9 | |
| Input TV + MoDLInput Defense=Total-Variation (TV) Minimization, Base Model=MoDL, lambda=0.01, iterations=52025.01 | 16.36 | 48.8 | |
| MoDLInput Defense=None, Base Model=MoDL2025.01 | 16.3 | 48.6 |