Preclinical researchers evaluating noninvasive liver diagnostics have demonstrated that multiparametric 0.4-Tesla low-field magnetic resonance imaging can resolve key pathological features of metabolic dysfunction-associated steatotic liver disease in rats, while exhibiting important stage-dependent limitations. Published in European Radiology Experimental by Kenichiro Okumura, Hibiki Yamauchi, Kazuto Kozaka, and colleagues, the prospective preclinical investigation examined how low-field relaxometric readouts reflect coexisting hepatic steatosis, macrophage-driven inflammation, collagen accumulation, and tissue iron. Across a rigorously characterized dietary cohort of 90 male Wistar rats, transverse relaxometry and apparent proton density fat fraction reliably tracked fat accumulation, whereas longitudinal relaxation time elevation reflected active, high-burden disease progression rather than static residual fibrosis. The authors cautioned that low-field multiparametric measurements require platform-specific calibration and direct human validation before translational implementation, as individual magnetic resonance parameters demonstrated complex sensitivities to overlapping tissue alterations rather than isolated pathological processes.
Metabolic dysfunction-associated steatotic liver disease represents a pervasive clinical challenge requiring quantitative imaging biomarkers that can noninvasively stage fat infiltration, inflammatory activity, and fibrotic remodeling without confounding from tissue iron. While proton density fat fraction has become an established magnetic resonance metric for hepatic steatosis, comprehensive liver relaxometry remains technically demanding because concurrent extracellular matrix deposition, inflammatory tissue water shifts, and iron deposition alter relaxation times in divergent directions. Low-field magnetic resonance imaging operating at sub-Tesla strengths, such as 0.4 Tesla, offers an appealing platform for broadening diagnostic accessibility through less complex hardware, open-magnet physical architectures, and substantially lower installation and operational requirements. However, low-field acquisitions intrinsically suffer from a reduced signal-to-noise ratio, necessitating specialized reconstruction pipelines and rigorous validation against whole-organ histological standards to confirm quantitative integrity.
To evaluate the biological accuracy of 0.4-Tesla imaging, the investigators conducted a prospective cross-sectional study in accordance with the ARRIVE guidelines and institutional animal ethics approval. Healthy male Wistar rats were fed a choline-deficient, L-amino acid-defined, high-fat diet to induce the complete histological spectrum of steatotic liver disease. The experimental cohort comprised 90 analyzed animals allocated across prespecified terminal sacrifice time points without randomization, after excluding two rats that died prior to scanning and 13 unneeded reserve animals. The design included acute dietary induction groups evaluated at day zero, day two, and day four, with six rats per interval. Disease progression cohorts were examined at day seven, day 14, day 28, and day 63, comprising 12 rats at each milestone. Two dietary recovery groups, designated F4N20 and F9N15, each included 12 rats fed the experimental diet for four or nine weeks, respectively, followed by normal chow maintenance for 20 or 15 weeks to specifically assess histological resolution and residual structural damage.
Imaging examinations were conducted under intraperitoneal medetomidine, midazolam, and butorphanol anesthesia on a clinical 0.4-Tesla open scanner, specifically the Gravity MRI system manufactured by FUJIFILM Healthcare. The multiparametric acquisition protocol yielded native longitudinal relaxation time, fat-corrected longitudinal relaxation time, transverse relaxation time, effective transverse relaxation time, and an apparent proton density fat fraction estimate. Longitudinal relaxation mapping utilized a two-dimensional inversion-recovery fast spin-echo sequence with a repetition time of 2,000 milliseconds, an inversion time of 600 milliseconds, and an echo time of 15.0 milliseconds, referenced against an initial fast spin-echo acquisition at an echo time of 15.0 milliseconds representing zero inversion time. Transverse relaxation mapping employed a dual-echo fast spin-echo acquisition with echo times of 15 and 60 milliseconds to calculate an apparent monoexponential value. The authors highlighted that this composite measurement was acquired without water-fat separation or fat suppression, and two echo points prevented voxel-wise goodness-of-fit modeling.
Effective transverse relaxation mapping and apparent proton density fat fraction were derived from a low-flip-angle multi-echo gradient-echo acquisition configured with a repetition time of 250 milliseconds, echo times of 8.9, 17.8, 26.7, and 35.6 milliseconds, and a flip angle of 12 degrees. Calculations relied on a simplified two-component magnitude water-fat model utilizing a fixed nominal chemical shift term without voxel-wise main magnetic field mapping or off-resonance correction, leading the investigators to explicitly designate the resulting values as apparent estimates. To address the inherently low signal-to-noise ratio of the two-point inversion-recovery sequence, commercial image-domain deep-learning reconstruction using SwiftMR version 3.0.13.1 from AIRS Medical was applied specifically to the longitudinal relaxation raw data. Fat-corrected longitudinal relaxation times were subsequently calculated from the reconstructed inversion-recovery datasets using a proton-density-fat-fraction-informed exponential mixing model incorporating a fixed fat longitudinal relaxation time of 200 milliseconds, while other metrics utilized standard system reconstruction.
Following imaging, whole-section digital liver histopathology was performed to provide a comprehensive reference standard that minimizes tissue sampling error relative to conventional needle biopsies. Hematoxylin and eosin staining quantified steatosis as lipid vacuole area fraction, while Sirius Red staining measured hepatic fibrosis through collagen proportionate area. Immunohistochemical labeling for CD68 quantified macrophage-rich inflammatory infiltration, and Perls’ Prussian blue staining measured tissue iron-positive area fraction. Whole-slide images were digitized and segmented using QuPath version 0.5.1 with stain-specific thresholding by pathologists blinded to imaging results. On corresponding magnetic resonance maps, three circular regions of interest were propagated across central liver slices by an independent, blinded observer, demonstrating excellent reproducibility with intraobserver intraclass correlation coefficients spanning 0.99 to 1.00 and interobserver coefficients ranging from 0.95 to 1.00, despite the absence of exact voxel-level coregistration.
Quantitative histological and imaging metrics shifted markedly across the experimental timeline, with all global comparisons demonstrating statistically significant differences at p less than 0.001. During dietary induction and progression, apparent proton density fat fraction rose from a baseline of 7.1 percent with a standard deviation of 1.1 percent at day zero to a peak of 40.9 percent with a standard deviation of 1.7 percent at day 28, before tapering to 28.8 percent with a standard deviation of 2.5 percent at day 63. Histologic steatosis mirrored this trajectory, expanding from 5.5 percent with a standard deviation of 1.9 percent at baseline to 52.5 percent with a standard deviation of 6.7 percent at day 28. Conversely, fibrotic and inflammatory remodeling developed later in disease progression, reaching peak collagen proportionate area of 9.2 percent with a standard deviation of 1.4 percent and peak CD68-positive inflammatory fraction of 7.1 percent with a standard deviation of 0.7 percent at day 63.
Analyses exploring pathological determinants confirmed that apparent proton density fat fraction and transverse relaxation time predominantly tracked steatosis severity. In iron-adjusted partial correlations, lipid vacuole fraction correlated strongly with transverse relaxation time at rho equal to 0.90 and with apparent proton density fat fraction at rho equal to 0.88, both achieving p less than 0.001. Standardized multivariable linear regression adjusting for iron, collagen, and inflammation identified standardized coefficients of 0.91 for lipid vacuole area with transverse relaxation time and 1.00 with proton density fat fraction. When models were conservatively adjusted for experimental timepoint to account for disease co-evolution, transverse relaxation time maintained a statistically significant association with steatosis at rho equal to 0.34 and p equal to 0.002. Variance inflation factors in pathology-only regression remained acceptable at 1.53 for iron, 2.05 for lipid vacuoles, 2.61 for collagen, and 3.74 for CD68, confirming that unadjusted models were not undermined by severe collinearity.
Diagnostic assessment of inflammatory activity revealed divergent behaviors between transverse and effective transverse relaxometry. For detecting high inflammation, defined as a cohort median CD68 area fraction exceeding 2.33 percent, transverse relaxation time yielded an area under the receiver operating characteristic curve of 0.86 with a 95 percent confidence interval of 0.78 to 0.94 at a 74.1-millisecond cutoff, demonstrating 80.0 percent sensitivity and 82.2 percent specificity. Effective transverse relaxation time achieved an area under the curve of 0.80 with a 95 percent confidence interval of 0.71 to 0.89, providing 66.7 percent sensitivity and 91.1 percent specificity. Although their discriminatory curves did not differ significantly at p equal to 0.294, stage-aware sensitivity analyses attenuated the correlation between transverse relaxation time and CD68 to rho equal to 0.07 at p equal to 0.556, indicating that uncorrected transverse relaxation was primarily governed by concurrent fat accumulation rather than specific inflammatory infiltration. In contrast, effective transverse relaxation time retained an independent association with CD68 at rho equal to 0.46 and p less than 0.001 after stage adjustment.
Fibrosis discrimination highlighted critical nuances regarding longitudinal relaxation mapping. Across progression cohorts, native longitudinal relaxation time initially decreased from 272.4 milliseconds with a standard deviation of 3.3 milliseconds at baseline to 246.8 milliseconds with a standard deviation of 5.6 milliseconds at day 28 during steatosis predominance, before surging to 312.0 milliseconds with a standard deviation of 4.8 milliseconds at day 63. Native relaxation mapping discriminated advanced fibrosis thresholds with areas under the curve of 0.77 at collagen area above 2.42 percent and 0.78 above 2.90 percent, outperforming fat-corrected longitudinal relaxation, which yielded corresponding values of 0.62 at p equal to 0.008 and 0.60 at p equal to 0.001. Fat-corrected measurements maintained high specificity between 96.8 percent and 100.0 percent across cutoffs. Crucially, in dietary recovery animals of the F9N15 group, substantial residual fibrosis persisted at a collagen area of 4.7 percent with a standard deviation of 0.4 percent, yet native longitudinal relaxation normalized to 268.0 milliseconds with a standard deviation of 9.2 milliseconds. Consequently, longitudinal relaxation elevation reflected the combined inflammatory, edematous, and cellular stress of active high-burden disease rather than isolated collagen deposition.
Technical characterization of deep-learning reconstruction revealed operational trade-offs that underscore the need for cautious translation. Phantom and healthy rat liver evaluations demonstrated that enabling deep-learning reconstruction introduced a mean longitudinal relaxation time increase of 10.64 milliseconds relative to unenhanced processing, with 95 percent limits of agreement spanning negative 0.31 to positive 21.58 milliseconds. Test-retest descriptors were numerically less favorable with reconstruction enabled, yielding a repeatability coefficient of 15.28 milliseconds compared with 5.71 milliseconds, a within-subject coefficient of variation of 1.99 percent versus 0.77 percent, and an intraclass correlation coefficient of 0.15 versus 0.84, though formal superiority testing was not prespecified. Because biological relaxation shifts between disease stages encompassed only several tens of milliseconds, algorithm-dependent baseline shifts directly affect absolute value interpretation.
Subject of Research: Multiparametric 0.4-Tesla low-field magnetic resonance imaging mapping in a rat model of metabolic dysfunction-associated steatotic liver disease
Article Title: Multiparametric 0.4-T low-field MRI mapping for metabolic dysfunction-associated steatotic liver disease in rats
Article References: Okumura, K., Yamauchi, H., Kozaka, K., Ohno, N., Kitao, A., Yamashita, T., Harada, K., Takeshita, Y., Takamura, T., & Kobayashi, S. (2026). Multiparametric 0.4-T low-field MRI mapping for metabolic dysfunction-associated steatotic liver disease in rats. European Radiology Experimental, 10(1), Article 185. https://doi.org/10.1186/s41747-026-00852-6
Image Credits: AI Generated
DOI: 10.1186/s41747-026-00852-6
Keywords: low-field MRI, metabolic dysfunction-associated steatotic liver disease, relaxometry, T1 mapping, T2 mapping, T2* mapping, proton density fat fraction, deep-learning reconstruction, liver fibrosis, hepatic steatosis, liver inflammation, collagen proportionate area, Wistar rats, digital pathology, preclinical imaging, SwiftMR, CDAHFD diet
