Table 3.

Mean metrics for comparing our method to expert annotators, displayed separately for T1 and T2 mapping images.

	Automated vs expert annotation		Agreement of three experts (interobserver)
	T1	T2	T1	T2
IoU - Jaccard index $\Uparrow$ ($\mathrm{\%}$)	81.7${}^{\mathrm{a}}$	78.39${}^{\alpha }$	84.14${}^{\mathrm{b}}$	86.25${}^{\beta }$
Dice score $\Uparrow$ ($\mathrm{\%}$)	87.8${}^{\mathrm{a}}$	85.02${}^{\alpha }$	89.84${}^{\mathrm{b}}$	91.73${}^{\beta }$
Hausdorff distance $\Downarrow$ (mm)	3.97${}^{\mathrm{a}}$	5.39${}^{\alpha }$	2.89${}^{\mathrm{b}}$	2.41${}^{\beta }$
Mean surface distance $\Downarrow$ (mm)	1.77${}^{\mathrm{a}}$	2.44${}^{\alpha }$	1.34${}^{\mathrm{b}}$	1.00${}^{\beta }$
Number of training samples	378	454	—	—
Number of test samples	116	119	116	119

For each metric and mapping type (T1 or T2), differences are statistically significant ($p<0.001$, also denoted with different lettering in superscripts). $\Uparrow$ indicates higher-the-better metrics, while metrics $\Downarrow$ are lower-the-better.