From: Improving MR image quality with a multi-task model, using convolutional losses
SSIM | VIF | ||
---|---|---|---|
\(\times 2\) | Bicubic | \(0.668\pm 0.146\) | \(1.001\pm 0.160\) |
Zero-filled | \(0.830\pm 0.100\) | \(\varvec{1.007\pm 0.059}\) | |
UniRes | \(0.832\pm 0.191\) | \(0.863\pm 0.266\) | |
Subsampling | \(\varvec{0.848\pm 0.082}\) | \(0.999\pm 0.051\) | |
MT | \(0.843\pm 0.091\) | \(0.998\pm 0.066\) | |
MT+\(\mathcal {L}_C\) | \(0.843\pm 0.084\) | \(\varvec{1.009\pm 0.065}\) | |
\(\times 3\) | Bicubic | \(0.654\pm 0.150\) | \(0.946\pm 0.167\) |
Zero-filled | \(0.782\pm 0.117\) | \(\varvec{1.009\pm 0.071}\) | |
UniRes | \(\varvec{0.809\pm 0.198}\) | \(0.855\pm 0.271\) | |
Subsampling | \(\varvec{0.809\pm 0.098}\) | \(0.967\pm 0.062\) | |
MT | \(0.801\pm 0.104\) | \(0.999\pm 0.078\) | |
MT+\(\mathcal {L}_C\) | \(\varvec{0.805\pm 0.100}\) | \(\varvec{1.010\pm 0.071}\) | |
\(\times 4\) | Bicubic | \(0.634\pm 0.158\) | \(0.835\pm 0.181\) |
Zero-filled | \(0.720\pm 0.142\) | \(0.955\pm 0.101\) | |
UniRes | \(0.743\pm 0.210\) | \(0.851\pm 0.263\) | |
Subsampling | \(\varvec{0.758\pm 0.117}\) | \(0.983\pm 0.069\) | |
MT | \(\varvec{0.756\pm 0.120}\) | \(0.952\pm 0.104\) | |
MT+\(\mathcal {L}_C\) | \(0.754\pm 0.118\) | \(\varvec{0.996\pm 0.088}\) |