Assessing the reliability of hippocampal single-voxel proton magnetic resonance spectroscopy through intra-class effect decomposition

Abstract

The hippocampus is crucial for learning, memory, and stress regulation, but is vulnerable to influences that compromise its structural and metabolic integrity. Proton magnetic resonance spectroscopy (MRS) enables in vivo quantification of hippocampal neurochemistry, yet the absolute and relative reliability of these measures has rarely been systematically evaluated. We assessed hippocampal metabolite estimates acquired with single-voxel MRS using a semi-adiabatic localization by adiabatic selective refocusing (semi-LASER) sequence at 3 T. Nineteen healthy adults (8 male; mean age = 40.9 ± 16.9 years) underwent four measurements across two consecutive days, with one session including an extended acquisition. Absolute reliability was quantified using within-subject coefficients of variation (CVs), and relative reliability was estimated with the Intra-Class Effect Decomposition (ICED) framework, which partitions variance into true-score, day-specific, session-specific, and residual components. Despite good spectral quality and high absolute reliability for several metabolites, relative reliability ranged from poor to good. Good relative reliability was observed only for water-referenced myo-inositol, total choline, and total creatine; no creatine-referenced metabolite reached this threshold. Increasing the number of transients increased signal-to-noise ratio but did not improve relative reliability; only creatine-referenced total N-acetylaspartate showed significant session-specific variance. Thus, high absolute reliability did not necessarily imply high relative reliability. Hippocampal MRS at 3 T may therefore be suitable for group-level or mean-level metabolite estimates, but appears less suitable for correlational, biomarker, or other individual-difference applications that require stable between-person differences. Methodological refinement, including improved voxel placement, motion correction, shimming, and quantification pipelines, may improve the relative reliability of hippocampal MRS.