Magnetic resonance imaging (MRI) is the primary imaging modality for diagnosing, characterizing, and following multiple sclerosis (MS). However, lesion contrasts on conventional MRI often overlap with other pathological conditions, limiting specificity. Saturation transfer (ST) MRI can provide molecular information related to myelin, proteins, and lipids, but extracting quantitative proton exchange parameters from ST data remains challenging. The authors aimed to extend an artificial intelligence‑assisted ST magnetic resonance fingerprinting (ST‑MRF) framework to enable rapid, multi‑pool quantification of semisolid magnetization transfer (MT) and aliphatic relayed nuclear Overhauser effect (rNOE) contributions relevant to MS pathology.
The reported work modified and extended an existing AI‑boosted ST‑MRF method at 7 tesla to quantify dynamics of three proton pools: semisolid MT and aliphatic rNOE at –3.5 ppm and –1.6 ppm relative to water. The approach combined ST contrast with magnetic resonance fingerprinting and AI reconstruction to produce quantitative maps of proton volume fractions for the targeted pools.
The method was tested both in vitro and in vivo. In vitro validation used lipid phantoms with known lipid concentrations to assess accuracy of the reconstructed proton volume fractions. In vivo experiments were performed longitudinally in a cuprizone mouse model of MS with a cohort size of n = 12. The study tracked changes over the course of cuprizone feeding and compared ST‑MRF biomarkers to conventional water relaxometry measures and to histological outcomes.
Note: The source text does not report full imaging parameter sets, AI model architecture, or exact timing and frequency of imaging sessions beyond the statement that decreases were observed as early as week 4 of cuprizone feeding.
In lipid phantom experiments, the reconstructed proton volume fractions for all three proton pools correlated very strongly with the known lipid concentrations. The reported correlation coefficient exceeded 0.96 (r > 0.96) and the relationship was statistically significant (p < 0.001). These results indicate that the ST‑MRF pipeline can recover quantitative proton pool fractions in controlled samples and support its applicability to tissue measurements where lipid and semisolid macromolecular content are relevant.
Applied longitudinally to a cuprizone model of demyelination (n = 12), the ST‑MRF method detected significant reductions in semisolid MT and rNOE proton volume fractions within the corpus callosum. These decreases reached statistical significance (p < 0.01) and were observed as early as week 4 of cuprizone feeding. The temporal pattern suggests that semisolid and aliphatic rNOE pools are sensitive to early tissue changes in this demyelination model.
The report emphasizes that these ST‑MRF‑derived biomarkers preceded changes detected by conventional water relaxometry, suggesting improved sensitivity of multi‑pool ST quantification for early detection of microstructural or molecular alterations associated with demyelination.
According to the source, changes in ST‑MRF biomarkers occurred earlier than abnormalities detected by conventional water relaxometry measures. Additionally, the ST‑MRF‑based biomarkers were consistent with histological findings reported by the investigators, supporting the biological relevance of the observed reductions in semisolid MT and rNOE proton volume fractions.
The source does not provide granular details on which histological markers were used or the precise histology results, beyond stating agreement between imaging biomarkers and histology.
The authors conclude that rapid, multi‑pool ST‑MRF quantification at 7T is feasible and can produce sensitive biomarkers for MS characterization in a preclinical model. Key takeaways include strong phantom validation (r > 0.96, p < 0.001), early detection of corpus callosum semisolid MT and rNOE reductions (significant by week 4, p < 0.01), and concordance with histology. These findings indicate potential for ST‑MRF to provide molecularly informed metrics of myelin and related macromolecular pools that may offer improved specificity relative to conventional contrasts.
Limitations and missing details in the source abstract: the preprint text summarized here does not report full MRI acquisition parameters, the specific AI model and training details, precise time points imaged beyond the week‑4 observation, or the exact histological assays and quantitative histology results. The authors declared no competing interests and listed funding sources, but additional methodological specifics would be required to assess reproducibility and to guide translation to other preclinical or clinical platforms.
Overall, the work demonstrates a proof of concept that AI‑enhanced, multi‑pool ST‑MRF can quantify semisolid MT and aliphatic rNOE contributions in phantoms and detect early, disease‑relevant changes in a cuprizone model of MS.