Swin Transformer and the Unet Architecture to Correct Motion Artifacts in Magnetic Resonance Image Reconstruction

Table 1

Comparison of state-of-art motion correction methods.

Year proposed	Method	Parameters	Resizing module	Activation function	Training time (h min)
2018	MoCo-Net	47,120,129	MaxPooling	ReLU	67 20
2019	Namer-Net	893,899	—	ReLU	81 24
2020	Modified-2D-Net	3,363,215	Conv with stride 2	ReLU	23 52
2022	MC-Net	5,496,001	MaxPooling	ReLU, sigmoid	20 44
2022	Stacked-Unet	4,017,966	Average_pooling, GlobalMax_pooling	ReLU	136 49
2023	Mark-Net	1,623,236	MaxPooling	LeakyReLU	23 22
—	Proposed MACS-Net	33,227,907	Patch merging, dual upsampling	GELU, ELU	56 54