Table 1.

The performance comparison of test results using background (BG), mandibular canal (MC), global accuracy, class average accuracy, and mean of intersection over union (mIoU).

	BG	MC	Global acc	Class acc	mIoU
2D SegNet*,^	0.96265	0.84278	0.96254	0.90271	0.49116
2D U-Net	0.76764	0.50470	0.76741	0.63617	0.38462
2D U-Net*	0.91744	0.26388	0.91686	0.59066	0.45984
2D U-Net*,^	0.83897	0.42008	0.83859	0.62953	0.42043
2D U-Net (adjacent 2 images)*,^	0.82013	0.54608	0.81988	0.68310	0.41125
3D U-Net	0.99972	0.92738	0.99922	0.95915	0.57721

^*The same number of filters (same as the original SegNet and UNet).

^Pre-trained weights with natural images.