Region-specific brain metabolite reference estimates and exploratory age-related changes in cognitively intact older adults: a preliminary deep-learning 1H-MRS study
Introduction Normative brain metabolite references are essential for evaluating metabolic change in rehabilitation patients with stroke, traumatic brain injury, or cognitive decline. We provide preliminary, cohort-specific reference estimates in cognitively healthy older adults using proton magnetic resonance spectroscopy (1H-MRS) and examine regional and age-related differences. Methods Twenty healthy older adults (10 males, 10 females; mean age 71.0 ± 9.5 years, range 56–84) underwent single-voxel 1H-MRS at 3T in three regions: dorsal anterior cingulate cortex (DACC), posterior cingulate cortex (PCC), and occipital cortex (OC). A Bayesian deep-learning convolutional neural network quantified 14 metabolite ratios normalized to total creatine (tCr). Because regions were acquired within each participant (20 independent observations), regional differences were evaluated primarily with the within-subject Friedman test (Kendall’s W), with Kruskal–Wallis as a secondary analysis; Benjamini–Hochberg false discovery rate (FDR) correction was applied across the 14 metabolites. Age– and MMSE–metabolite associations used Spearman’s rank correlation with FDR applied both across all 42 tests and within each region. Results Nine of 14 metabolites showed significant regional differences (seven also surviving the primary Friedman test after FDR correction), with glycerophosphocholine (GPC)/tCr showing the largest effect (Friedman W = 0.82; Kruskal–Wallis ε2 = 0.55). γ-aminobutyric acid (GABA)/tCr was highest in the OC, whereas myo-inositol (mI) and choline compounds were highest in the DACC. In exploratory age analyses, the PCC showed negative correlations for aspartate (Asp)/tCr (ρ = –0.61, p = 0.004) and glutathione (GSH)/tCr (ρ = –0.59, p = 0.006), both surviving region-wise FDR correction (q = 0.044), with glutamate (Glu)/tCr nominal (ρ = –0.45) and total N-acetylaspartate (tNAA)/tCr at trend level (ρ = –0.39); the DACC showed positive correlations for glucose (Glc)/tCr (ρ = +0.60) and lactate (Lac)/tCr (ρ = +0.50). No association survived the more conservative correction across all 42 tests. The OC showed no significant age associations. Conclusion This study established preliminary, cohort-specific multi-regional 1H-MRS reference estimates for cognitively intact older adults, with robust regional profiles and large effect sizes. The PCC showed exploratory age-related change—most consistently a coordinated aspartate–glutathione decline—requiring confirmation in larger cohorts, whereas the OC emerged as a candidate metabolically stable region.