Table 6 The results and their standard errors for denoising. The three experiments differ in the selected SNR ratio of the corruptions. The performance of the selected models is compared using regards to SSIM and VIF. The results in bold indicate the best performance without significant differences between them following a Nemenyi post-hoc test
From: Improving MR image quality with a multi-task model, using convolutional losses
SSIM | VIF | ||
|---|---|---|---|
5 dB | Original | \(0.921\pm 0.070\) | \(1.023\pm 0.056\) |
C.A.D. | \(\varvec{0.961\pm 0.030}\) | \(\varvec{1.043\pm 0.032}\) | |
BM3D | \(0.949\pm 0.033\) | \(1.017\pm 0.027\) | |
Noise | \(\varvec{0.965\pm 0.029}\) | \(1.025\pm 0.032\) | |
MT | \(0.957\pm 0.035\) | \(1.040\pm 0.051\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.958\pm 0.033}\) | \(\varvec{1.049\pm 0.048}\) | |
0 dB | Original | \(0.864\pm 0.107\) | \(1.006\pm 0.090\) |
C.A.D. | \(0.931\pm 0.049\) | \(\varvec{1.066\pm 0.044}\) | |
BM3D | \(\varvec{0.941\pm 0.040}\) | \(1.035\pm 0.044\) | |
Noise | \(\varvec{0.943\pm 0.041}\) | \(1.037\pm 0.038\) | |
MT | \(\varvec{0.936\pm 0.049}\) | \(1.054\pm 0.059\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.935\pm 0.051}\) | \(1.057\pm 0.049\) | |
\(-5\) dB | Original | \(0.771\pm 0.158\) | \(1.021\pm 0.113\) |
C.A.D. | \(0.874\pm 0.096\) | \(\varvec{1.099\pm 0.072}\) | |
BM3D | \(0.850\pm 0.132\) | \(\varvec{1.092\pm 0.088}\) | |
Noise | \(\varvec{0.906\pm 0.070}\) | \(1.066\pm 0.054\) | |
MT | \(0.892\pm 0.081\) | \(1.063\pm 0.061\) | |
MT+\(\mathcal {L}_C\) | \(0.886\pm 0.084\) | \(\varvec{1.091\pm 0.065}\) |