Table 3.

Lesion detection from smartphone (wide-field) and dermoscopy images.

	One-class	Malignancy	Sub-type
mAP${}_{@0.5}$	0.793	0.755	0.749
mAP${}_{@0.75}$	0.281	0.232	0.250
mAP${}_{@0.5,0.95}$	0.370	0.335	0.337
Recall	0.954	0.956	0.948
IoU	$\mathbf{0}\mathbf{.}{\mathbf{729}}_{\mathbf{(}\mathbf{0.59}\mathbf{,}\mathbf{0.82}\mathbf{)}}$	$0.{726}_{(0.58,0.80)}$	$0.{716}_{(0.59,0.80)}$

Performance is evaluated as the mean Average Precision (mAP) at three different thresholds: 0.5, 0.75 and [0.5, 0.95], recall (sensitivity) and intersection over union (IoU) summarized as median (interquartile range).

Significant values are in [bold].