Table 1.
State-of-the-art methods for left-ventricular segmentation shown at end-diastole (ED) and end-systole (ES) on the ACDC test set compared to proposed approach.
| Left-ventricle label | Dice similarity coefficient | Hausdorff distance | Ejection fraction | End-diastolic volume | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| ED | ES | ED | ES | Corr. | bias ± std | Corr. | bias ± std | ||||
| val. | val. | mm | mm | val. | % | % | val. | mL | mL | ||
| * | CarSON | 0.967 | 0.929 | 5.656 | 7.676 | 0.990 | 0.252 | 3.183 | 0.996 | 0.762 | 6.672 |
| 1 | Dong et al. (17) | 0.967 | 0.928 | 6.366 | 7.573 | 0.993 | −0.360 | 2.689 | 0.998 | 2.032 | 4.611 |
| 2 | Simantiris and Tziritas (18) | 0.967 | 0.928 | 5.476 | 6.921 | 0.991 | 0.490 | 2.965 | 0.997 | 1.530 | 5.736 |
| 3 | Isensee et al. (19) | 0.964 | 0.912 | 6.180 | 8.386 | 0.990 | −0.476 | 3.114 | 0.997 | 3.746 | 5.146 |
| Myocardium label | Dice similarity coefficient | Hausdorff distance | Left-ventricular mass | End-systolic volume | |||||||
| ED | ES | ED | ES | Corr. | bias ± std | Corr. | bias ± std | ||||
| val. | val. | mm | mm | val. | g | g | val. | mL | mL | ||
| * | CarSON | 0.898 | 0.913 | 8.128 | 9.189 | 0.981 | 1.405 | 10.32 | 0.985 | 1.152 | 9.391 |
| 1 | Dong et al. (17) | 0.904 | 0.923 | 7.014 | 7.328 | 0.987 | −2.547 | 8.28 | 0.988 | −1.984 | 8.335 |
| 2 | Simantiris and Tziritas (18) | 0.891 | 0.904 | 8.264 | 9.575 | 0.992 | −2.904 | 6.46 | 0.983 | −2.134 | 10.11 |
| 3 | Zotti et al. (20) | 0.873 | 0.895 | 8.197 | 8.318 | 0.989 | −2.1 | 7.91 | 0.988 | −1.79 | 8.575 |
Red are the best results for each metric.
Proposed segmentation method.