Clutter suppression in ultrasound: performance evaluation and review of low-rank and sparse matrix decomposition methods

Table 8 The algorithms with good results on B-mode phantom data

Group	Abbreviation	Time	CNR	Group	Abbreviation	Time	CNR
RPCA	LSADM\(\bullet\)	1.455	3.582	MC	RPCA-GD\(\bullet\)	6.118	3.165
RPCA	L1F	2.595	1.038	MC	ScGrassMC	4.123	1.338
RPCA	DECOLOR	7.015	2.847	LRR	EALM\(\bullet\)	10.899	3.681
RPCA	RegL1-ALM	4.352	3.700	LRR	IALM\(\bullet\)	2.469	3.681
RPCA	GA\(\circ\)	0.032	3.680	LRR	ADM**	0.668	0.024
RPCA	GM\(\circ\)	0.153	3.713	LRR	LADMAP	0.363	3.681
RPCA	MoG-RPCA	4.691	3.359	LRR	FastLADMAP	0.802	3.681
RPCA	noncvxRPCA\(\bullet\)	0.110	3.681	LRR	ROSL	0.421	3.712
RPCA	NSA1\(\bullet\)	1.407	3.602	TTD	3WD\(\bullet\)	1.942	2.964
RPCA	NSA2\(\bullet\)	1.537	3.568	TTD	MAMR	2.784	3.154
RPCA	flip-SPCP-sum-SPG	276	3.695	TTD	RMAMR	6.776	2.289
RPCA	flip-SPCP-max-QN	138	3.695	TTD	ADMM\(\circ\)*	3.627	0.794
RPCA	Lag-SPCP-SPG*	5.010	1.598	NMF	NMF-MU	2.143	3.681
RPCA	Lag-SPCP-QN*	7.219	0.685	NMF	NMF-PG	8.774	3.565
RPCA	FW-T*	0.715	3.073	NMF	NMF-ALS	2.406	3.681
RPCA	BRPCA-MD\(\bullet\)	283	3.724	NMF	NMF-ALS-OBS	2.710	3.681
RPCA	BRPCA-MD-NSS\(\bullet\)	291	3.511	NMF	PNMF	16.815	3.681
RPCA	VBRPCA	4.627	3.692	NMF	ManhNMF	2.292	3.662
RPCA	PRMF	1.522	3.522	NMF	NeNMF	0.066	3.681
RPCA	TFOCS-EC\(\bullet\)	9.131	3.349	NMF	LNMF**	0.204	0.279
RPCA	GoDec	0.095	3.681	NMF	ENMF	13.546	3.681
RPCA	SSGoDec	1.459	3.679	NMF	nmfLS2	0.320	3.681
RPCA	GreGoDec	0.229	3.681	NMF	Semi-NMF	0.154	2.604
ST	GRASTA	1.321	1.156	NMF	Deep-Semi-NMF	0.156	3.681
MC	FPC	49.672	2.454	NMF	iNMF	1.482	3.650
MC	GROUSE**	1.580	0.068	NMF	DRMF\(\bullet\)*	2.461	3.497
MC	IALM-MC	6.992	3.690	TD	HoSVD	0.532	3.681
MC	LMaFit	0.314	3.300	TD	HoRPCA-S-NCX	89.622	3.693
MC	LRGeomCG	0.757	3.723	TD	Tucker-ADAL	258	3.573
MC	MC-NMF\(\circ\)	0.585	3.423	TD	Tucker-ALS	0.130	3.681
MC	OR1MP\(\circ\)	0.096	3.365

The algorithms with \(\circ\) are the 3 new algorithms work on processed data, which are defective on original data. The algorithms with \(\bullet\) are sensitive to structured peak pixels and work after logarithmic processing. Three algorithms with ** get good results on original envelope data but are defective on processed data. The algorithms with * give pure backgrounds

ISSN: 1475-925X