Noise decorrelation coil combination optimizes SNR of edited 1H MRS data

Introduction

Determining the optimal radiofrequency (RF) coil combination method for magnetic resonance spectroscopy (MRS) is crucial for maximizing the signal-to-noise ratio (SNR) and reliably detecting low-concentration metabolites, such as γ-aminobutyric acid (GABA). We compared the performances of several previously proposed algorithms using GABA-edited ¹H MRS data. Given that phased-array coils often exhibit noise correlations that reduce SNR, we hypothesized that noise decorrelation algorithms would be most effective.

Methods

We examined six coil combination methods, with the second 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); and 6) generalized least squares (GLS). Each method was applied to 119 GABA-edited MEGA-PRESS datasets acquired on 3 T GE and Siemens MRI scanners across 11 research sites. We estimated the SNR of GABA+ and N-acetylaspartate (NAA) and tested for statistical differences between the six approaches. We also calculated the intersubject coefficients of variation (CVs) of GABA+/creatine (Cr) ratios.

Results

There were significant differences in the SNR of GABA+ and NAA between the methods. Noise decorrelation methods produced higher SNR compared to the other approaches, with nd-comb, WSVD, and GLS yielding, on average, approximately 37 % more GABA+ and 34 % more NAA SNR than equal weighting. GLS yielded the highest SNR for both GABA+ and NAA. The CVs for GABA+/Cr were generally somewhat smaller when using noise decorrelation.

Conclusion

As predicted, noise decorrelation coil combination, particularly GLS, produced optimal SNR for GABA-edited MRS data.