Table 1.
Lesion Detection Results: Comparison between the different models evaluated. The results represent the mean detection TPF, FPF, DSCd, mean segmentation DSCs, and the mean runtime in minutes when analyzing the 36 MS patients using a leave-one-out cross-validation scheme. The automatic segmentation masks were obtained by thresholding the probability maps at 0.5 (using argmax), and all automatic lesions with a size lower than three voxels were removed.
| Method | TPF | FPF | DSCd | DSCs | Runtime (in minutes) |
|---|---|---|---|---|---|
| SimLearnedDFs | 83.09 ± 21.06 | 9.36 ± 16.97 | 0.83 ± 0.16 | 0.55 ± 0.18 | 8.70 ± 0.09 |
| SepLearnedDFs | 57.77 ± 34.34 | 13.67 ± 21.99 | 0.60 ± 0.31 | 0.39 ± 0.22 | 9.08 ± 0.06 |
| DemonsDFs | 62.06 ± 32.74 | 11.98 ± 23.09 | 0.67 ± 0.29 | 0.42 ± 0.24 | 18.10 ± 0.05 |
| NDFs | 53.99 ± 38.01 | 17.20 ± 26.96 | 0.55 ± 0.35 | 0.37 ± 0.28 | 7.58 ± 0.09 |
| Sweeney et al. (2013) | 59.82 ± 37.59 | 33.59 ± 33.52 | 0.57 ± 0.33 | 0.44 ± 0.26 | 8.36 ± 0.01 |
| Cabezas et al. (2016) | 70.93 ± 34.48 | 17.80 ± 27.96 | 0.68 ± 0.33 | 0.52 ± 0.24 | 18.36 ± 0.02 |
| Salem et al. (2018) | 80.0 ± 27.77 | 21.87 ± 26.26 | 0.76 ± 0.25 | 0.55 ± 0.22 | 18.55 ± 0.02 |
| Schmidt et al. (2019) | 68.66 ± 35.26 | 31.89 ± 36.10 | 0.62 ± 0.34 | 0.40 ± 0.25 | 7.58 ± 0.03 |