Improved Complex Convolutional Neural Network Based on SPIRiT and Dense Connection for Parallel MRI Reconstruction

Table 4

Comparisons of average PSNR and SSIM values of the MR images reconstructed by all the models for the Sagittal-PD dataset from the 3x and 5x accelerated 1DRU, 2DRU, RADU, and 2DPU patterns.

Undersampling; patterns	Reconstruction; models	Threefold AF		Fivefold AF
		PSNR (dB)	SSIM	PSNR (dB)	SSIM
1DRU	SPIRiT	31.976 ± 2.617	0.822 ± 0.045	29.763 ± 2.473	0.754 ± 0.050
	Deepcomplex	36.889 ± 1.590	0.893 ± 0.019	34.342 ± 1.632	0.833 ± 0.027
	DONet	37.308 ± 1.541	0.896 ± 0.018	34.931 ± 1.578	0.831 ± 0.028
	SPIRiT-Net	38.178 ± 1.733	0.910 ± 0.020	35.726 ± 1.804	0.865 ± 0.030
2DRU	SPIRiT	39.310 ± 1.376	0.927 ± 0.013	36.182 ± 1.618	0.884 ± 0.021
	Deepcomplex	40.806 ± 1.169	0.941 ± 0.009	38.430 ± 1.193	0.905 ± 0.014
	DONet	40.798 ± 1.179	0.941 ± 0.009	38.426 ± 1.205	0.904 ± 0.014
	SPIRiT-Net	42.119 ± 1.471	0.951 ± 0.011	39.906 ± 1.498	0.925 ± 0.017
2DPU	SPIRiT	39.759 ± 3.089	0.932 ± 0.013	36.970 ± 1.582	0.886 ± 0.037
	Deepcomplex	40.617 ± 1.219	0.943 ± 0.009	38.342 ± 1.202	0.891 ± 0.015
	DONet	40.897 ± 1.167	0.945 ± 0.008	38.586 ± 1.167	0.908 ± 0.013
	SPIRiT-Net	42.339 ± 1.481	0.955 ± 0.010	40.264 ± 1.544	0.930 ± 0.016
RADU	SPIRiT	39.220 ± 1.165	0.926 ± 0.013	36.575 ± 1.528	0.872 ± 0.021
	Deepcomplex	40.559 ± 1.178	0.938 ± 0.009	37.693 ± 1.293	0.891 ± 0.015
	DONet	40.614 ± 1.714	0.947 ± 0.012	37.749 ± 1.835	0.910 ± 0.021
	SPIRiT-Net	41.809 ± 1.527	0.948 ± 0.012	38.985 ± 1.717	0.912 ± 0.020

The best values are in bold.