Table 2.
Correlation (R) and mean error for parameters extracted from segmentations, and intra-observer parameter reproducibility
| Parameter | Deep learning | Human performance | ||||
|---|---|---|---|---|---|---|
| Train | Validation | |||||
| R | Error (%) | R | Error (%) | R | Error (%) | |
| Cine LVEF (%) | 0.99 | 1.00 | 0.99 | 1.10 | 0.93 | 3.79 |
| Cine RVEF (%) | 0.84 | 4.12 | 0.76 | 4.05 | 0.70 | 5.77 |
| LGE Scar (%) | 0.91 | 3.69 | 0.78 | 5.39 | 0.60 | 5.97 |
| Native T1 mean values (ms) | 0.99 | 0.69 | 0.98 | 1.00 | 0.83 | 3.24 |
| Post-contrast T1 mean values (ms) | 0.99 | 1.22 | 0.99 | 0.73 | 0.99 | 1.10 |
| Native T2 mean values (ms) | 0.99 | 0.95 | 0.99 | 0.65 | 0.99 | 0.89 |
| AO Flow amplitude (mL) | 0.99 | 3.16 | 0.89 | 5.77 | 0.92 | 6.03 |
The mean errors for the left ventricular ejection fraction (LVEF), the right ventricular ejection fraction (RVEF), and the scar represent absolute errors, while the rest are relative errors
LGE image segmentation