Table 1.

Nomenclature of the evaluation metrics. The evaluation metrics are listed in the table below and estimated using the following abbreviations: true positive (TP), lesion-wise true positive (LTP), false positive (FP), lesion-wise false positive (LFP), false negative (FN), lesion-wise false negative (LFN), automatic lesion volume (V_a), manual lesion volume (V_m), $d_a^{am}$ and $d_a^{ma}$ are the Euclidean distances between the automatic (a) and the manual (m) lesion surface voxels, and n_a and n_m are the number of surface voxels for each segmentation.

Name	Abbr.	Equation	Units
Dice similarity measure	DSC	$\frac{2\times TP}{FP+FN+2\times TP}$	%
True positive rate or sensitivity	TPR	$\frac{TP}{TP+FN}$	%
	LTPR	$\frac{LTP}{LTP+LFN}$	%
Positive predictive value or precision	PPV	$\frac{TP}{TP+FP}$	%
	LPPV	$\frac{LTP}{LTP+LFP}$	%
Volume difference	VolD	$\frac{\left|V_a-V_m\right|}{V_m}$	%
False positive rate or fall-out	FPR	$\frac{FP}{FP+TP}$	%
Symmetric surface distance	SurfD	$\frac{1}{\left(n_a+n_m\right)}\left({\displaystyle \sum _{i=1}^{n_a}\left|d_a^{\mathit{am}}\right|}+{\displaystyle \sum _{j=1}^{n_m}\left|d_m^{ma}\right|}\right)$	mm