Table 3.
Segmentation performance derived from different methods. The reported clinical results are represented using the absolute error and relative error between the ground truth and prediction. The best results are shown in bold. SAX2D-M and SAX3D-M represent models that only use the magnitude images as the input. The other methods use both magnitude and velocity images as the input.
| Method | Dice (%) | ASD (mm) | EDV (ml) | ESV (ml) | LVEF (%) | KE (mJ) |
|---|---|---|---|---|---|---|
| SAX2D | 84.44 ± 6.31 | 3.31 ± 1.90 | 19.93(10.39%) | 19.43(25.45%) | 9.44(18.41%) | 0.07(5.61%) |
| RAW2D | 80.02 ± 7.39 | 4.01 ± 1.87 | 29.57(17.97%) | 32.61(44.54%) | 10.92(21.60%) | 0.19(13.93%) |
| SAX3D | 82.81 ± 6.69 | 4.43 ± 4.72 | 22.80(12.80%) | 24.66(32.21%) | 10.06(19.07%) | 0.14(11.61%) |
| RAW3D | 79.79 ± 7.57 | 3.68 ± 1.65 | 24.66(14.32%) | 26.82(35.01%) | 10.41(21.91%) | 0.16(11.61%) |
| SAX2DF | 84.52 ± 6.61 | 3.14 ± 2.35 | 20.29(10.55%) | 17.38(22.22%) | 7.37(13.93%) | 0.09(7.04%) |
| SAX2D-M | 81.39 ± 7.42 | 4.13 ± 2.93 | 26.26(13.45%) | 32.51(45.40%) | 21.04(39.37%) | 0.15(11.83%) |
| SAX3D-M | 80.71 ± 0.09 | 5.29 ± 4.80 | 24.96(13.97%) | 38.41(57.66%) | 18.88(34.86%) | 0.17(13.31%) |