Table 3.
Performance of the different automatic segmentation methods on the research dataset ADNI.
| ADNI | DSC | Volume similarity | Volume error rate | Intraclass correlation | False positive rate | False negative rate |
|---|---|---|---|---|---|---|
| LPA | 0.539 (0.505–0.573) |
0.734 (0.691–0.775) |
0.850 (0.570–1.131) |
0.812 (0.709–0.876) |
0.438 (0.399–0.477) |
0.366 (0.321–0.410) |
| LGA | 0.474 (0.441–0.509) |
0.759 (0.719–0.798) |
0.426 (0.361–0.490) |
0.680 (0.561–0.770) |
0.444 (0.408–0.480) |
0.535 (0.494–0.574) |
| BIANCA | 0.469 (0.430–0.506) |
0.638 (0.588–0.686) |
0.760 (0.609–0.912) |
0.417 (0.249–0.560) |
0.393 (0.349–0.436) |
0.481 (0.428–0.533) |
| SLS | 0.527 (0.495–0.559) |
0.732 (0.696–0.766) |
0.903 (0.729–1.078) |
0.890 (0.507–0.957) |
0.564 (0.531–0.596) |
0.277 (0.239–0.314) |
| W2MHS | 0.351 (0.318–0.385) |
0.603 (0.551–0.654) |
2.219 (1.139–3.299) |
0.292 (0.108–0.456) |
0.539 (0.482–0.594) |
0.569 (0.529–0.608) |
| nicMSlesion(original) | 0.454 (0.419–0.490) |
0.787 (0.746–0.826) |
0.694 (0.382–1.007) |
0.948 (0.924 –0.964) |
0.517 (0.476–0.557) |
0.503 (0.463–0.543) |
| nicMSlesion(retrained) |
0.595 (0.357–0.921) |
0.889 (0.867–0.910) |
0.270 (0.159–0.381) |
0.979 (0.968–0.986) |
0.384 (0.351–0.416) |
0.402 (0.376–0.427) |
| UBO | 0.486 (0.459–0.514) |
0.762 (0.730–0.793) |
0.907 (0.575–1.239) |
0.881 (0.652–0.945) |
0.587 (0.559–0.615) |
0.360 (0.328–0.392) |
For each metric, we present the average and the 95% confidence interval within parentheses. DSC: Dice similarity coefficient. Results in bold indicates the best score for each metric