Table 3 Segmentation results of IMT translated images on Iseg2017 evaluated by FCN-score. Note that “gm” and “wm” indicate gray matter and white matter respectively. The minor gap between translated-modality images and the target-modality images shows decent generation performance of our framework.
Method | Accuracy | Dice | |||
|---|---|---|---|---|---|
all | gm | wm | gm | wm | |
T1 → T2 | 0.892 | 0.827 | 0.506 | 0.777 | 0.573 |
T2 (real) | 0.920 | 0.829 | 0.610 | 0.794 | 0.646 |
T2 → T1 | 0.882 | 0.722 | 0.513 | 0.743 | 0.569 |
T1 (real) | 0.938 | 0.811 | 0.663 | 0.797 | 0.665 |