Sparse Parallel MRI Based on Accelerated Operator Splitting Schemes

Table 2

Numerical results for the TV-based SparseSENSE model.

	Coil number	Undersampling factor	Performance metric	BOS	AM	SBB	FBOSS	FBOSP
Data 1	32	10	PSNR (dB)	46.87	47.36	49.53	49.61	49.57
			Relative error ()	3.04	2.87	2.24	2.22	2.23
			CPU time (sec.)	325.4	132.3	95.3	55.3	53.0
Data 1	32	6	PSNR (dB)	55.93	56.99	57.62	58.11	57.76
			Relative error ()	10.68	9.49	8.82	8.34	8.68
			CPU time (sec.)	162.7	78.2	45.4	32.2	31.2
Data 1	32	4	PSNR (dB)	61.00	62.41	62.80	63.56	63.56
			Relative error ()	5.98	5.08	4.86	4.45	4.45
			CPU time (sec.)	99.9	48.0	29.3	27.2	26.5
Data 2	15	6	PSNR (dB)	54.22	54.35	57.17	57.25	57.95
			Relative error ()	14.82	14.60	10.55	10.46	9.65
			CPU time (sec.)	199.4	75.1	35.1	29.3	24.7
Data 2	15	4	PSNR (dB)	63.21	65.34	65.93	66.22	66.70
			Relative error ()	5.26	4.12	3.85	3.72	3.50
			CPU time (sec.)	77.7	30.8	16.0	12.7	10.8
Data 3	8	10	PSNR (dB)	40.20	40.21	40.61	40.64	40.68
			Relative error ()	6.06	6.05	5.79	5.77	5.73
			CPU time (sec.)	60.5	24.5	18.0	13.2	10.8
Data 3	8	6	PSNR (dB)	48.86	49.29	50.20	50.13	50.21
			Relative error ()	2.24	2.13	1.92	1.93	1.92
			CPU time (sec.)	51.1	21.4	16.6	11.2	9.8
Data 3	8	4	PSNR (dB)	51.72	51.80	51.78	51.86	51.79
			Relative error ()	1.61	1.60	1.60	1.58	1.60
			CPU time (sec.)	34.0	13.2	10.3	8.5	7.6