An Improved Mask R-CNN Model for Multiorgan Segmentation

Table 1

Comparisons of the proposed and state-of-the-art multiorgan segmentation methods on the presented dataset.

Methods	Organ	JAC (%)	DICE (%)
Linguraru et al. [47]	Heart	74.9 ± 4.7	71.3 ± 3.1
	Right lung	86.5 ± 2.2	86.3 ± 1.2
	Left lung	85.1 ± 1.3	84.6 ± 0.9
	PTV	82.5 ± 1.6	81.3 ± 2.8
	CTV	80.5 ± 1.2	77.5 ± 1.9
He et al. [48]	Heart	87.5 ± 1.2	86.3 ± 0.7
	Right lung	89.3 ± 1.6	87.3 ± 2.4
	Left lung	90.2 ± 1.9	88.3 ± 1.1
	PTV	86.3 ± 1.7	84.1 ± 2.4
	CTV	85.6 ± 2.3	83.7 ± 3.1
Gauriau et al. [49]	Heart	88.8 ± 0.6	87.5 ± 1.2
	Right lung	91.5 ± 0.9	91.3 ± 0.8
	Left lung	90.3 ± 0.9	90.8 ± 1.5
	PTV	89.1 ± 1.4	86.2 ± 1.7
	CTV	88.7 ± 1.9	87.3 ± 1.5
Original mask R-CNN	Heart	95.1 ± 0.5	94.2 ± 0.7
	Right lung	97.0 ± 0.9	96.2 ± 1.2
	Left lung	96.3 ± 0.4	95.1 ± 0.6
	PTV	94.7 ± 1.1	93.2 ± 0.9
	CTV	94.3 ± 0.5	93.7 ± 0.8
Improved mask R-CNN (ours)	Heart	96.6 ± 1.3	95.1 ± 1.2
	Right lung	98.1 ± 0.5	97.8 ± 0.3
	Left lung	97.6 ± 0.4	96.2 ± 1.4
	PTV	95.3 ± 1.5	95.2 ± 0.7
	CTV	95.8 ± 0.7	94.4 ± 1.2