Table 4.
Descriptive statistics for each approach. Performance rates for each approach; median (IQR): DC, HD, ASSD, precision, and recall
| Clusterize | ALI | lesion_gnb | LINDA | |
|---|---|---|---|---|
| Image metrics | N = 152 | N = 132 | N = 132 | N = 132 |
| DC | 0.18 (0.31) | 0.4 (0.44) | 0.42 (0.37) | 0.5 (0.61) |
| HD (mm) | 80.89 (36.6) | 62.79 (48.49) | 58.19 (25.22) | 36.34 (42.48) |
| ASSD (mm) | 12.64 (7.68) | 9.58 (13.17) | 8.75 (7.89) | 4.97 (13.98) |
| Precision | 0.11 (0.22) | 0.31 (0.45) | 0.29 (0.33) | 0.6 (0.63) |
| Recall | 0.89 (0.26) | 0.61 (0.51) | 0.8 (0.44) | 0.59 (0.63) |
| Average processing time | 106.43 s for automated clustering + 251.75 s for manual identification | 396.99 + 247.83 s per healthy brain | 246.12 s | 3,843.66 s |
Abbreviations: ALI, automated lesion identification; ASSD, average symmetric surface distance; DC, dice coefficient; HD, Hausdorff's distance; lesionGnb, Gaussian naïve Bayes lesion detection; LINDA, lesion identification with neighborhood data analysis.