Table 3 The results and their standard errors for bias field correction. The three experiments differ in the selected magnitude of the applied bias fields. 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 | ||
|---|---|---|---|
\(5\%\) | Original | \(0.973\pm 0.018\) | \(1.024\pm 0.016\) |
N4ITK | \(0.972\pm 0.033\) | \(\varvec{1.050\pm 0.093}\) | |
Implicit | \(0.978\pm 0.022\) | \(1.003\pm 0.032\) | |
Bias | \(\varvec{0.986\pm 0.013}\) | \(1.011\pm 0.032\) | |
MT | \(\varvec{0.983\pm 0.020}\) | \(1.019\pm 0.040\) | |
MT+\(\mathcal {L}_C\) | \(0.978\pm 0.020\) | \(\varvec{1.043\pm 0.045}\) | |
\(10\%\) | Original | \(0.957\pm 0.043\) | \(1.022\pm 0.038\) |
N4ITK | \(0.969\pm 0.034\) | \(\varvec{1.058\pm 0.098}\) | |
Implicit | \(0.954\pm 0.044\) | \(1.021\pm 0.045\) | |
Bias | \(\varvec{0.982\pm 0.016}\) | \(1.016\pm 0.037\) | |
MT | \(\varvec{0.979\pm 0.021}\) | \(1.024\pm 0.043\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.976\pm 0.022}\) | \(\varvec{1.044\pm 0.047}\) | |
\(20\%\) | Original | \(0.931\pm 0.066\) | \(1.012\pm 0.064\) |
N4ITK | \(0.965\pm 0.035\) | \(\varvec{1.072\pm 0.103}\) | |
Implicit | \(0.950\pm 0.067\) | \(1.050\pm 0.066\) | |
Bias | \(\varvec{0.981\pm 0.020}\) | \(1.022\pm 0.044\) | |
MT | \(\varvec{0.978\pm 0.025}\) | \(1.030\pm 0.046\) | |
MT+\(\mathcal {L}_C\) | \(0.976\pm 0.023\) | \(1.042\pm 0.049\) |