Table 1.
Summary results for the synthetic data experiments. The metrics shown in the table are the RMSE between ground truth and registered volumes (Volumes nRMSE), the DICE score between registered masks and the ground truth (Masks DICE), the percentage of times each kidney voxel was classified as such (Percent. kidney classification), the “Total Variation” metric of the concentration curves (TV), the estimation error of the parameters in the tracer kinetic model (FP error and FT error) and the residual of the model fit (Model fit nRMSE). LiMo-MoCo outperformed the other methods for all metrics except for the DICE score of the masks, for which REFVOL was more accurate.
| Method | Volumes nRMSE | Masks DICE | Mask consistency (%) | TV | FP error [ml/100ml/min] | FT error[ml/100ml/min] | Model fit nRMSE |
|---|---|---|---|---|---|---|---|
| No-MoCo | 2.267 ± 0.053 | 0.803 ± 0.295 | 54.8 ± 11.8 | 20.400 ± 2.920 | 79.7 ± 80.6 | 94.7 ± 1.338 | 0.283 ± 0.082 |
| REFVOL | 1.519 ± 0.102 | 0.958 ± 0.056 | 71.8 ± 2.3 | 14.180 ± 2.478 | 94.3 ± 122.0 | 72.5 ± 125.9 | 0.079 ± 0.009 |
| gPCA | 1.632 ± 0.049 | 0.523 ± 0.193 | 72.5 ± 15.0 | 19.704 ± 2.536 | 94.6 ± 114.6 | 109.3 ± 151.9 | 0.249 ± 0.053 |
| LiMo-MoCo | 1.514 ± 0.097 | 0.953 ± 0.051 | 74.1 ± 2.0 | 10.212 ± 1.499 | 75.6 ± 94.3 | 70.6 ± 123.2 | 0.065 ± 0.016 |