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

Table 1 Performance comparison of multiple models on the Brats2021 validation dataset and 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	0.825	0.844	0.900	Yes	No	Yes
V-Net	45.6	182.432	0.815	0.840	0.751	No	Yes	Yes
UnetR	102	204.899	0.842	0.853	0.905	Yes	No	Yes
TransBTS	33.0	65.975	0.824	0.843	0.889	Yes	Yes	Yes
SwinBTS	35.7	71.394	0.828	0.843	0.896	Yes	Yes	Yes
AttentionUnet	23.6	47.257	0.841	0.851	0.870	No	No	No
Swin Pure Unet3D	33.6	67.163	0.817	0.822	0.885	Yes	Yes	Yes
Swin Unet3D	33.7	67.403	0.834	0.866	0.905	–	–	–

ISSN: 1472-6947