BA-GCA Net: Boundary-Aware Grid Contextual Attention Net in Osteosarcoma MRI Image Segmentation

Table 3

Performance of models.

Model	Pre	Rec	F1	DSC	IOU	#params	FLOPs
FCN-16s [21]	0.922	0.882	0.900	0.859	0.824	134.3 M	190.35 G
FCN-8s [21]	0.892	0.914	0.902	0.875	0.831	134.3 M	190.08 G
MSFCN [24]	0.881	0.936	0.906	0.874	0.841	23.38 M	1524.34 G
MSRN [25]	0.893	0.945	0.918	0.887	0.853	14.27 M	1461.23 G
FPN [46]	0.914	0.924	0.919	0.888	0.852	48.20 M	141.45 G
U-Net [22]	0.922	0.924	0.923	0.892	0.867	17.26 M	160.16 G
OCR [41]	0.897	0.908	0.901	0.891	0.827	27.35 M	125.67 G
DeepLabV3 [4]	0.926	0.925	0.925	0.909	0.870	39.63 M	170.45 G
UNet++ [59]	0.924	0.924	0.924	0.908	0.868	18.16 M	165.23 G
SVM [60]	0.756	0.764	0.760	0.734	0.702	—	—
DRN [50]	0.916	0.922	0.917	0.909	0.843	17.66 M	76.93 G
DRN + CA [42]	0.918	0.923	0.919	0.910	0.851	18.06 M	77.21 G
Ours (DRN + GCA)	0.927	0.924	0.925	0.913	0.866	18.11 M	77.34 G
Ours (DRN + GCA + STLB)	0.925	0.934	0.929	0.916	0.873	18.47 M	82.67 G
Ours (DRN + GCA + STLB + STB)	0.938	0.937	0.937	0.927	0.880	19.88 M	149.70 G