Table 2 Segmentation performance in the internal test set.
Fazekas scale | |||||
|---|---|---|---|---|---|
0 (n = 43) | 1 (n = 36) | 2 (n = 48) | 3 (n = 41) | All | |
AUPRC† | 0.69 [0.49–0.84] | 0.82 [0.65–0.92] | 0.90 [0.70–0.95] | 0.81 [0.55–0.95] | |
Hausdorff distance† | 38.5 [21.6–77.3] | 31.7 [18.1–51.8] | 27.2 [18.0–41.4] | 32.2 [18.1–58.5] | |
Average symmetric surface distance† | 2.5 [1.0–8.0] | 1.4 [0.9–2.3] | 1.4 [0.9–3.3] | 1.7 [0.9–4.2] | |
DICE† | 0.61 [0.38–0.76] | 0.74 [0.58–0.84] | 0.81 [0.57–0.88] | 0.73 [0.49–0.87] | |
Jaccard† | 0.45 [0.24–0.61] | 0.59 [0.41–0.72] | 0.69 [0.39–0.78] | 0.58 [0.32–0.77] | |
Recall† | 0.63 [0.35–0.85] | 0.75 [0.59–0.90] | 0.80 [0.57–0.91] | 0.73 [0.37–0.91] | |
Precision† | 0.65 [0.34–0.83] | 0.73 [0.57–0.87] | 0.84 [0.69–0.93] | 0.74 [0.44–0.91] | |
True WMH volume, ml | 0 | 8.2 [3.2–17.7] | 22.9 [10.4–39.2] | 57.3 [31.7–82.4] | 22.3 [0.0–75.2] |
Predicted WMH volume, ml | 0.3 [0.0–1.1] | 8.2 [2.6–14.7] | 23.6 [11.7–37.3] | 54.5 [28.1–74.2] | 21.9 [0.0–72.0] |
Absolute error of the predicted volume, ml | 0.3 [0.0–1.1] | 2.4 [0.1–4.5] | 2.9 [0.1–7.7] | 6.8 [0.1–25.7] | 3.1 [0.0–15.9] |
