Subtraction artifacts and frequency (Mis-)alignment in J-difference GABA editing

Purpose To compare the repeatability of γ-aminobutyric acid (GABA) measurements using J-difference editing, before and after spectral realignment - a technique which has previously been demonstrated to improve the quality of J-difference GABA spectra. Materials and Methods We performed in vivo measurements in three brain regions (occipital, sensorimotor, and dorsolateral prefrontal cortex [DLPFC]), and analyzed these using alternative alignment approaches to evaluate the impact of alignment on repeatability: "Independent alignment" (aligning each subspectrum independently) and "Pairwise alignment" (aligning each on and off subspectrum as a pair) were compared. Results Pairwise alignment improved the group mean coefficient of variation in all regions; 0.4% in occipital, 1.1% in sensorimotor, and 1.1% in DLPFC. Independent alignment resulted in subtraction artifacts in the majority of cases, and increased the coefficient of variation in the DLPFC by 9.4%. Simulations demonstrate that the GABA quantification error in datasets with high B₀ drift, is 4.5% without alignment, but <1% with optimal alignment. Conclusion Pairwise alignment improves the repeatability of GABA spectroscopy data. However, independently aligning all on and off subspectra can lead to artifacts and worse repeatability when compared with nonaligned data.