Noise decorrelation optimizes SNR of GABA-edited MRS data: A comparison of RF coil combination methods

Abstract

Introduction: Determining the best radiofrequency (RF) coil combination method is crucial for maximizing the signal-to-noise ratio (SNR) to detect low concentration metabolites (e.g., {gamma}-aminobutyric acid (GABA)) in magnetic resonance spectroscopy (MRS). We hypothesized that algorithms accounting for noise correlations between coil elements would optimize SNR, given that phased-array coils provide better SNR than surface coils and allow accelerated acquisitions, and methods accounting for noise correlations outperform those assuming no correlations. Methods: We examined six coil combination methods, the latter half accounting for noise correlations: 1) equal weighting; 2) signal weighting; 3) S/N2 weighting; 4) noise-decorrelated combination (nd-comb); 5) whitened singular value decomposition (WSVD); 6) generalized least squares (GLS). We utilized MEGA-PRESS data from 119 participants (mean age: 26.4 {+/-} 1 SD 4.2 years; males/females: 54/65) acquired on 3T GE and Siemens MRI scanners at 11 research sites, obtained from the Big GABA study. We measured the SNR of GABA and N-acetylaspartate (NAA). We also calculated the intersubject coefficients of variation of GABA. Results: There were significant differences in SNR between coil combination methods for both GABA+ and NAA. More specifically, the noise decorrelation methods produced higher GABA+ and NAA SNR than the other approaches, where nd-comb, WSVD, and GLS produced, on average, ~37% and ~34% more SNR than equal weighting, respectively. GLS produced the highest SNR for GABA+ and NAA. The coefficients of variation for GABA+ were generally slightly smaller for the noise decorrelation methods. Conclusion: Noise-decorrelation methods produced higher SNR than other methods, especially GLS, which should be investigated in advanced editing protocols.