Swin Unet3D: a three-dimensional medical image segmentation network combining vision transformer and convolution

Table 2 Performance comparison of multiple models on the Brats2018 validation dataset and the analysis of the significant differences between the performance of Swin Unet3D on the validation set and the performance of other models using the Wilcoxon sign test

Model name	Params	Params size	Mean dice			Significant difference
	(M)	(MB)	ET	TC	WT	ET	TC	WT
3D U-Net	7.9	15.834	0.704	0.763	0.869	No	Yes	Yes
V-Net	45.6	91.216	0.361	0.528	0.801	Yes	Yes	Yes
UnetR	102	204.899	0.743	0.767	0.869	Yes	Yes	Yes
TransBTS	33.0	65.975	0.707	0.723	0.844	Yes	Yes	No
SwinBTS	15.7	34.411	0.732	0.717	0.863	No	No	No
AttentionUnet	23.6	47.257	0.613	0.550	0.658	Yes	Yes	Yes
Swin Pure Unet3D	33.6	67.163	0.657	0.646	0.797	Yes	Yes	Yes
Swin Unet3D	33.7	67.403	0.716	0.761	0.874	–	–	–

ISSN: 1472-6947