Table 5 The results and their standard errors for motion artefacts. The three experiments differ in the percentage of the center of the k-space left uncorrupted for each corrupted image. 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 | ||
|---|---|---|---|
\(25\%\) | Original | \(\varvec{0.967\pm 0.036}\) | \(1.043\pm 0.058\) |
TV | \(\varvec{0.965\pm 0.036}\) | \(1.044\pm 0.056\) | |
Motion | \(\varvec{0.963\pm 0.033}\) | \(1.017\pm 0.045\) | |
MT | \(\varvec{0.959\pm 0.045}\) | \(1.031\pm 0.055\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.963\pm 0.034}\) | \(\varvec{1.046\pm 0.046}\) | |
\(12.5\%\) | Original | \(0.944\pm 0.064\) | \(1.015\pm 0.094\) |
TV | \(0.938\pm 0.064\) | \(\varvec{1.065\pm 0.091}\) | |
Motion | \(\varvec{0.947\pm 0.045}\) | \(1.020\pm 0.046\) | |
MT | \(0.937\pm 0.065\) | \(1.042\pm 0.069\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.945\pm 0.049}\) | \(1.047\pm 0.050\) | |
\(5\%\) | Original | \(0.908\pm 0.085\) | \(1.015\pm 0.111\) |
TV | \(0.903\pm 0.086\) | \(\varvec{1.1150\pm 0.108}\) | |
Motion | \(\varvec{0.923\pm 0.059}\) | \(1.020\pm 0.048\) | |
MT | \(0.910\pm 0.080\) | \(1.043\pm 0.080\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.921\pm 0.063}\) | \(1.046\pm 0.053\) |