Table 3.

List of evaluation metrics and their definitions. The organization of the challenge defined individual lesions for recall and F1 as 3D connected components. N means the number of 3D connected components. $\widehat{d}(X,Y)$ means $\underset{x\in X}{\max }\underset{y\in Y}{\min }d(x,y)$, where d(x, y) is the distance between x and y points. V_ML and V_AL mean the WMH volume computed by manual segmentation and our method, respectively.

	DSC	H95	AVD	Recall	F1
Equation	$\frac{2TP}{FP+FN+2TP}$	$\max \{\widehat{d}(X,Y),\widehat{d}(Y,X)\}$	$\frac{\left|V_{ML}-V_{AL}\right|}{V_{ML}}$	$\frac{N_{TP}}{N_{TP}+N_{FN}}$	$\frac{2N_{TP}}{2N_{TP}+N_{FN}+N_{FP}}$

^*Abbreviations – DSC: the Dice Similarity Coefficient; H95: a modified Hausdorff distance (95th percentile); AVD: the absolute percentage volume difference; Recall: the sensitivity for detecting individual lesions; F1: F1-score for individual lesions; TP: true positive; FN: false negative; FP: false positive.