TABLE 1.

List of major MRSI acceleration methods (without combinations). Methods listed in different categories can be combined. We differentiate three ranges of acceleration (~1.5 to 8, low; ~8 to 20, moderate; 20 and above, high) and encoding methods that additionally increase (SNR/t gain) or decrease (SNR/t loss) SNR per time efficiency as compared against a current clinical standard ¹H‐MRSI protocol (Cartesian sampling, T _R = 1500 ms, T _E = 30 ms), respectively

Category	Method	Pros	Cons	Application
Short‐T R/T E	SSFP	‐ highest SNR/t gain ‐ moderate acceleration	‐ T 1/T 2‐weighting ‐ low spectral resolution ‐ poor water and lipid suppression in 1H‐MRSI ‐ banding artifacts/B 0 sensitive	‐ metabolites with long T 2 and short T 1 ‐ hyperpol. 13C MRSI/MRI ‐ 1H‐MRSI possible but restricted to major singlets ‐ preferably <7 T
	Turbo‐spin‐echo	‐ low acceleration	‐ low spectral resolution ‐ T 2‐weighting (in k‐space) ‐ ΔB 1 + sensitive	‐ metabolites with long T 2 ‐ singlets in 1H MRSI ‐ preferably <3 T
	FID‐MRSI	‐ SNR/t gain ‐ moderate acceleration ‐ high SNR ‐ J‐coupled metabolites in phase ‐ ΔB 1 + insensitive ‐ low SAR ‐ low CSDE	‐ T 1‐weighting ‐ trade‐off between speed (T R) and spectral resolution ‐ moderate lipid suppression in 1H‐MRSI	‐ short T 2/J‐coupled metabolites ‐ ultra‐high field ‐ 13C/31P/1H‐MRSI ‐ preferably >1.5 T
Cartesian SSE	EPSI	‐ high acceleration ‐ inherently constant k‐space weighting	‐ some SNR/t loss ‐ limited SBW/spatial resolution ‐ gradient demanding	‐ 13C/31P/1H‐MRSI ‐ preferably <7 T
Non‐Cartesian SSE	Spirals	‐ highest acceleration ‐ any k‐space weighting possible	‐ some SNR/t loss ‐ limited SBW/spatial resolution ‐ gradient demanding	‐ 13C/31P/1H‐MRSI ‐ preferably <7 T
	CRTs	‐ high acceleration ‐ inherent k‐space weighting (optimization possible)	‐ some SNR/t loss ‐ limited SBW/spatial resolution ‐ gradient demanding	‐ 13C/31P/1H‐MRSI ‐ preferably ≥3 T
	Rosettes	‐ can be tailored for either high speed or low gradient stress ‐ inherently weighted k‐space (optimization possible)	‐ some SNR/t loss ‐ moderate SBW/spatial resolution limitation	‐ 13C/31P/1H‐MRSI ‐ preferably ≥7 T
	Radial EPSI	‐ high acceleration ‐ inherent k‐space weighting (fixed)	‐ some SNR/t loss ‐ limited SBW/spatial resolution ‐ gradient demanding	‐ 13C/31P/1H‐MRSI ‐ preferably <7 T
Coherent k‐space undersampling	SENSE	‐ no gradient demands ‐ low acceleration	‐ some SNR/t loss ‐ needs multi‐channel receive coils ‐ needs explicit sensitivity maps ‐ spatial aliasing ‐ motion sensitive	‐ preferably 1H‐MRSI ‐ 13C/31P‐MRSI possible, but difficult to obtain reliable sensitivity maps ‐ preferably ≥3 T/better at UHF
	GRAPPA	‐ no gradient demands ‐ interleaving to reduce motion sensitivity ‐ low acceleration	‐ some SNR/t loss ‐ needs multi‐channel receive coils ‐ spatial aliasing	‐ preferably 1H‐MRSI ‐ 13C/31P‐MRSI possible ‐ preferably ≥3 T/better at UHF
	CAIPIRINHA	‐ no gradient demands ‐ better control of aliasing ‐ interleaving to reduce motion sensitivity ‐ low acceleration	‐ some SNR/t loss ‐ needs multi‐channel receive coils ‐ spatial aliasing	‐ preferably 1H‐MRSI ‐ 13C/31P‐MRSI possible ‐ preferably ≥3 T/better at UHF
Multi‐slice excitation	Multi‐band/SMS	‐ accelerate also in slice direction ‐ low acceleration	‐ some SNR/t loss ‐ needs multi‐channel receive coils ‐ increased SAR/B 1 + ‐ spatial aliasing	‐ preferably 1H‐MRSI, but 13C/31P‐MRSI possible ‐ better at UHF
Incoherent k‐space undersampling	CS	‐ SNR/t gain through regularization ‐ moderate acceleration	‐ sparse data (representation) required ‐ minimum SNR required to work robustly	‐ in spectral domain only for long‐T E 1H‐MRSI or 13C/31P‐MRSI
Prior knowledge based	SLIM/SLOOP/SLAM	‐ SNR/t gain through regularization and spatial averaging ‐ high acceleration	‐ sensitive to bias fields such as B 0 inhomogeneity	‐ 31P‐MRS(I) ‐ potentially hyperpol. 13C‐MRS(I) ‐ spectra from multiple arbitrarily shaped compartments instead of metabolite maps (except for GSLIM)
	SPICE	‐ SNR/t gain through regularization ‐ high acceleration	‐ requires assumptions about spatial and spectral priors; nuisance removal challenging ‐ may lead to spatial averaging ‐ may lead to spectral information loss	‐ preferably sparse well resolved spectra (13C, 31P), but 1H‐MRSI possible
	Super‐resolution reconstruction	‐ resolution increase via pure post‐processing	‐ requires assumptions about spatial priors ‐ no true spatial resolution gain—only smoother appearance of metabolite maps	all
	Spectral‐spatial excitation & IDEAL	‐ replaces time‐consuming spectral encoding by conventional MRI readout	‐ requires good spectral separation and ΔB 0 homogeneity	‐ 13C/31P‐MRSI; good spectral separation required

Abbreviation: ΔB ₁ ⁺—transmit field inhomogeneity.