Rethinking U-Net from an Attention Perspective with Transformers for Osteosarcoma MRI Image Segmentation

Table 4

Quantitative analysis of different segmentation models in MRI images of osteosarcoma.

Model	PRE	REC	IOU	DSC	F1	Parameters	FLOPS
U-Net	0.922 ± 0.09	0.924 ± 0.08	0.867 ± 0.04	0.892 ± 0.04	0.923 ± 0.05	17.26M	160.16G
PSPNet	0.856 ± 0.09	0.888 ± 0.05	0.772 ± 0.04	0.870 ± 0.06	0.872 ± 0.03	49.07M	101.55G
MSFCN	0.881 ± 0.06	0.936 ± 0.03	0.841 ± 0.02	0.874 ± 0.03	0.906 ± 0.05	20.38M	1524.34G
MSRN	0.893 ± 0.03	0.945 ± 0.05	0.853 ± 0.05	0.887 ± 0.03	0.918 ± 0.04	14.27M	1461.23G
FCN-16s	0.922 ± 0.09	0.882 ± 0.06	0.824 ± 0.04	0.859 ± 0.07	0.900 ± 0.08	134.3M	190.35G
FCN-8s	0.941 ± 0.07	0.873 ± 0.05	0.830 ± 0.05	0.876 ± 0.04	0.901 ± 0.04	134.3M	190.08G
FPN	0.914 ± 0.11	0.924 ± 0.07	0.852 ± 0.05	0.888 ± 0.08	0.919 ± 0.07	88.63M	141.45G
UATransNet Residual	0.962 ± 0.03	0.945 ± 0.04	0.922 ± 0.03	0.921 ± 0.04	0.955 ± 0.05	17.9M	161.01G
UATransNet Dense	0.960 ± 0.05	0.941 ± 0.05	0.918 ± 0.02	0.916 ± 0.07	0.950 ± 0.06	18.3M	163.20G