Measuring Brain Stiffness in Normal Adults According to Age and Sex Using Virtual MR Elastography: A Preliminary Study Authors: Jung Hyun Park, Miran Han, Miseon Yeon, Jin Wook Choi Published: 2026 ABSTRACT AbstractPurposeAssessing brain stiffness alterations associated with normal aging, sex-related differences, and regional variations is important from clinical perspective. The study aimed to investigate the changes in brain stiffness according to age and sex using virtual magnetic resonance elastography (vMRE), a noninvasive and novel technique based on diffusion-weighted imaging (DWI).MethodsWe retrospectively reviewed 96 adults (31–80 years, 51 men and 45 women) who underwent MRI as part of routine medical examinations. The shifted apparent diffusion coefficient was calculated from DWI (b = 200 and 1500 sec/mm2) and converted to the DWI-based virtual shear modulus (μdiff). Brain stiffness was measured in 12 brain regions: the cerebrum, cerebral gray/white matter, basal ganglia, thalamus, frontal lobe, parietal lobe, temporal lobe, occipital lobe, cerebellum, hippocampus, and amygdala. Linear regression and multiple-comparison tests were performed to evaluate the effects of age and regional differences.ResultsBrain stiffness exhibited regional differences, with higher shear modulus values observed in deep structures (thalamus and basal ganglia) compared to that in lobar regions (P < 0.01). Among lobar areas, the occipital lobe had the highest stiffness, whereas the temporal lobe had the lowest (P < 0.01). Linear regression revealed a significant negative correlation between age and brain stiffness (−0.06%/year, R2 = 0.09, P = 0.04). The most pronounced decline in elasticity (−0.28%/year, R2 = 0.66, P < 0.01) was observed in cerebral gray matter. Hippocampal stiffness significantly decreased after the fifth decade of life. Sex-related differences were found only in the parietal lobe (P = 0.02) and basal ganglia (P = 0.03).ConclusionOur findings suggest that vMRE can effectively measure the elastic properties of the brain. Brain stiffness decreases with age, but these changes vary across different regions.Keywords: aging, brain stiffness, diffusion-weighted imaging, virtual magnetic resonance elastography Introduction Various microstructural, chemical, and electrophysiological changes occur in the brain with increasing age, affecting its ability to function.1,2 Physical signs of brain aging have been identified through volumetric evaluation and perfusion alterations. Brain volume declines with age at a rate of approximately 5% per decade over 40 years of age, primarily due to the degeneration and death of neurons and oligodendrocytes.3 Thinning of the cortex and a decrease in cerebral blood flow (CBF) exhibit regional predominance.4,5 These anatomical and physiological changes can also be quantitatively measured using MR elastography (MRE), which provides information on tissue viscoelastic properties.6,7 Since the publication of the first brain MRE in 2005, significant advancements have been made in both methodology and clinical applications. MRE has been used to detect biomechanical changes in brain tissue associated with brain tumors and neurological disorders such as Alzheimer’s disease and multiple sclerosis.8–14 For clinical application, it is essential to understand patterns of brain stiffness alterations associated with normal aging, sex-related differences, and regional variations to avoid confounding effects from chronological and geometric factors. While previous studies have consistently reported a decrease in brain stiffness with age, discrepancies remain regarding regional viscoelasticity alterations and sex-related differences.15–21 Although MRE has the potential to provide noninvasive information about neural structure and function under various physiological and pathological conditions, its widespread clinical use is hindered by major limitations. The primary challenge is the lack of consistency across studies, as researchers employ varying methodologies for data acquisition, processing, and analysis. MRE examinations are costly and require dedicated hardware and software, posing a barrier to their widespread adoption.6,7,22 Recently, a diffusion-based approach has been proposed for noninvasive tissue elasticity evaluation in liver fibrosis.23 Le Bihan et al. hypothesized a relationship between tissue water diffusivity and tissue elastic properties in the liver, using a high b-value to increase sensitivity to non-Gaussian diffusion and tissue microstructure effects. They demonstrated that the shifted apparent diffusion coefficient (sADC) strongly correlated with tissue elasticity and that the virtual shear modulus (μdiff) (kPa) calculated from the sADC could effectively estimate liver fibrosis stages without external mechanical vibration.24 The virtual MRE (vMRE) method, based on diffusion, has been applied beyond liver fibrosis.25–29 Studies have assessed lesions in the head and neck, such as pituitary adenomas, cervical metastatic lymph nodes, and salivary gland tumors.30–32 These findings suggest that diffusion-weighted imaging (DWI) may be a viable tool for measuring brain stiffness. Therefore, this preliminary study aimed to investigate changes in brain viscoelastic properties with increasing age and sex using vMRE and to compare the findings with those obtained from conventional MRE. Materials and Methods Study population This retrospective study was approved by the ethics committee of our institution, Ajou University Hospital (AJOUIRB-DB-2022-541), and the requirement for written informed consent was waived. We retrospectively recruited individuals who underwent brain MRI as part of routine health screening examinations. Consecutive subjects without known systemic disease or neurological and psychiatric conditions were included from March 2021 to August 2022. Participants with brain lesions or significant image distortion on conventional images were excluded. Participants were categorized into 10 year age groups from 30 to 80 years, with a target of approximately 20 participants per group. However, because the cohort consisted of individuals undergoing routine health screening examinations, fewer eligible participants were available in the oldest age group. As a result, only 16 participants were included in the 70–80 year age group, resulting in a total sample size of 96 participants. When more than 20 eligible participants were available within an age group, subjects were included consecutively based on the order of examination. MRI acquisition and image analysis All MRI were performed using a single 3T MR scanner (Achieva; Philips Healthcare, Best, the Netherlands) with 3D T1-weighted images, axial T2-weighted images, and axial DWIs acquired at multiple b-values. DWI was performed using a fat-suppressed turbo spin-echo, single-shot echo-planar sequence (EPI) with the following parameters: TR/TE, 4200/72 ms; FOV, 220 × 220 mm2; matrix, 160 × 160; slice thickness, 5 mm without gap; 30 slices; SENSE factor, 2; and b-values of 0, 200, 1000, and 1500 s/mm2. The total image acquisition time for DWI was 1 min 30s. DWI scans were transferred to a workstation and analyzed using specialized vMRE software (EXPRESS; Philips Healthcare Korea, Seoul, Korea). The sADC was calculated using b-values of 200 and 1500 s/mm2, and the virtual shear stiffness maps were derived using Eqs. 1 and 2:23,24,31 where S200 and S1500 are the DWI signal intensities at b-values of 200 and 1500 s/mm2, respectively; μdiff (kPa) represents the diffusion-based virtual shear modulus; and the conversion factors, α = −12740 and β = 14.0. Virtual MRE maps were generated on a voxel-by-voxel basis after automatic removal of skull and background signals from multi–b-value DWI data. The resulting μdiff maps consisted of 30 slices. The μdiff (kPa) of the whole brain was automatically calculated for the entire brain parenchyma by EXPRESS. Regional μdiff values were measured in 12 specific brain regions using freehand ROI analysis, including the cerebrum, cerebral gray (GM) and white matter (WM), basal ganglia, thalamus, frontal, parietal, temporal, and occipital lobes, cerebellum, hippocampus, and amygdala. For most anatomical regions (cerebrum, cerebellum, lobar regions, basal ganglia, thalamus, hippocampus, and amygdala), ROIs were manually drawn across all slices using anatomical landmarks as references. To facilitate anatomical identification and improve ROI placement accuracy, the corresponding b 0 image was used as a background reference when drawing ROIs on the μdiff maps. For cerebral GM and WM measurements, a representative axial slice immediately above the lateral ventricles was selected. On the b 0 image, GM and WM were visually distinguished, and a freehand ROI was initially drawn to define the cerebral region. Extra-cerebral spaces were subsequently removed using semi-automatic segmentation. Within the selected slice, GM and WM regions were then semi-automatically segmented, and the corresponding μdiff values were extracted (Supplementary Fig.). To minimize potential variability associated with manual or semi-automatic ROI placement, ROI drawing and measurements were performed independently by 2 observers: a neuroradiologist (M.H., 16 years of neuroradiology experience) and a radiological researcher with a nursing license (M.Y., 2 years of medical imaging evaluation experience). The mean values of the independently measured μdiff were used for further analysis. Interobserver agreement was evaluated using the intraclass correlation coefficient (ICC). Brain volume was automatically measured using commercially available FDA-approved software (NeuroQuant; CorTechs Labs, San Diego, CA, USA). Statistical analysis All statistical analyses were performed using IBM SPSS Statistics for Windows version 25 (IBM Corporation, Armonk, NY, USA), and P < 0.05 was considered statistically significant. Independent t-tests were performed to evaluate sex-related differences. Multiple comparison tests, including analysis of variance (ANOVA) with post hoc analysis using Bonferroni correction, and linear regression were performed to assess the influence of aging and regional differences. ICCs were measured to determine interobserver agreement across all subjects. An ICC value of less than 0.4 indicated poor agreement, 0.41 to 0.6 indicated moderate agreement, 0.61 to 0.8 indicated substantial agreement, and 0.81 to 1.0 indicated almost perfect agreement. Results Finally, 96 adults with a median age of 54 years (range, 31–80 years; 51 men and 45 women) were enrolled in this study. The μdiff of each anatomic area, as measured by independent readers, showed substantial to almost perfect agreement (0.65–0.97) based on the ICCs. The measurement of the cerebrum ROI showed the highest ICC value of 0.97, whereas the cerebellum exhibited the lowest ROI value of 0.65. The mean brain volume and μdiff, for the whole brain and each anatomical region are presented in Table 1. Differences between sexes were observed only in the parietal lobe (P = 0.02) and basal ganglia (P = 0.03). The μdiff exhibited regional differences, as shown in Fig. 1. The μdiff of WM was higher than that of GM. Higher shear modulus values were observed in the deep structures such as the thalamus and basal ganglia compared to that in the lobar regions (P < 0.01). Among the lobar regions, the occipital lobe had the highest μdiff, whereas the temporal lobe exhibited the lowest value (P < 0.01). There was no significant difference in the μdiff between the frontal and parietal lobes (P = 1.00). Table 1Mean virtual shear modulus (μdiff, kPa) and brain volume for whole brain and each anatomic region, stratified by sexμdiff (kPa)All (n = 96)Men (n = 51)Women (n = 45)P-valueWhole brain4.91 ± 0.144.90 ± 0.144.91 ± 0.140.39Cerebrum, total4.84 ± 0.194.84 ± 0.184.85 ± 0.190.95Cerebrum, WM5.66 ± 0.215.65 ± 0.225.68 ± 0.200.44Cerebrum, GM3.60 ± 0.213.59 ± 0.203.61 ± 0.220.72Frontal lobe4.86 ± 0.204.86 ± 0.204.87 ± 0.200.74Parietal lobe4.90 ± 0.224.85 ± 0.204.96 ± 0.210.02*Temporal lobe4.71 ± 0.204.71 ± 0.224.71 ± 0.180.93Occipital lobe5.08 ± 0.205.05 ± 0.215.11 ± 0.200.12Basal ganglia6.04 ± 0.226.08 ± 0.235.99 ± 0.200.03*Thalamus5.76 ± 0.275.74 ± 0.255.78 ± 0.290.45Amygdala4.88 ± 0.214.89 ± 0.234.87 ± 0.180.58Hippocampus4.74 ± 0.234.72 ± 0.224.76 ± 0.230.43Cerebellum5.95 ± 0.225.97 ± 0.185.93 ± 0.250.37Brain volume (cm3)Cerebrum, total1090.99 ± 113.761144.48 ± 110.721031.55 ± 84.44< 0.01*Cerebrum, WM417.85 ± 50.17442.81 ± 47.68390.11 ± 36.76< 0.01*Cerebrum, GM481.77 ± 50.42500.80 ± 50.97460.62 ± 40.84< 0.01*Cerebellum116.18 ± 14.43122.95 ± 26.42108.66 ± 10.83< 0.01*Data are expressed as mean ± standard deviation. *presents statistically significant difference.GM, gray matter; WM, white matter. Mean virtual shear modulus (μdiff, kPa) and brain volume for whole brain and each anatomic region, stratified by sex Data are expressed as mean ± standard deviation. *presents statistically significant difference. GM, gray matter; WM, white matter. Virtual shear modulus (μdiff, (kPa), of the brain showing regional differences. (a) The μdiff calculated in WM is higher than that in GM but lower than that in deep structures. (b) In lobar areas, the occipital lobe had the highest μdiff value, while the temporal lobe had the lowest (P < 0.01). There was no significant difference between frontal and parietal lobes (P = 1.00).GM, gray matter; WM, white matter. Following volume adjustment, linear regression analysis revealed a significant negative correlation between age and the μdiff of the entire brain (−0.06%/year, R2 = 0.09, P = 0.04). A similar correlation was found in each anatomical area, except the parietal lobe, amygdala, and hippocampus, as shown in Table 2. The cerebral brain stiffness decreased by 0.17% of the μdiff value per year (Fig. 2). The highest decline in stiffness was observed in cerebral GM, with a reduction of 0.28%/year (R2 = 0.66, P < 0.01). Most anatomical areas exhibited a gradual decrease in the μdiff, but the decline in GM stiffness occurred at a relatively early age. In contrast, the μdiff in the hippocampus decreased significantly after the fifth decade of life (Figs. 3 and 4). Table 2Annual decrease in virtual shear modulus (μdiff, kPa), across different brain regionsBrain regions R 2 Annual change (Pa)Annual change (%)P-valueTotal brain0.09−30.06%0.04*Cerebrum, total0.39−90.17%< 0.01*Cerebrum, WM0.40−100.16%< 0.01*Cerebrum, GM0.66−120.28%< 0.01*Frontal lobe0.19−60.12%0.01*Parietal lobe0.01−10.02%0.60Temporal lobe0.12−50.10%< 0.01*Occipital lobe0.12−50.09%< 0.01*Basal ganglia0.22−80.12%< 0.01*Thalamus0.24−100.16%< 0.01*Amygdala0.00−10.02%0.75Hippocampus0.02−20.04%0.16Cerebellum0.14−60.09%< 0.01**presents statistically significant difference.GM, gray matter; WM, white matter. Annual decrease in virtual shear modulus (μdiff, kPa), across different brain regions *presents statistically significant difference. GM, gray matter; WM, white matter. Plot of virtual shear modulus (μdiff, kPa), of the cerebrum showing brain stiffness according to age in males (blue circles) and females (red circles). The μdiff was significantly decreased (−0.17%/year, R2 = 0.39, P < 0.01) with aging in both men and women, with no significant differences in gender. Bar charts showing mean virtual shear modulus (μdiff, kPa) for each brain region with aging. The μdiff values of the parietal lobe and amygdala were relatively preserved. The μdiff of the hippocampus decreased after the fifth decade, whereas that of the cerebral GM significantly decreased from an earlier age.GM, gray matter; WM, white matter. Representative case of a virtual shear modulus (μdiff, kPa), map. White matter and deep structures exhibited relatively high μdiff values. The μdiff values decreased with aging.vMRE, virtual magnetic resonance elastography. Discussion This study applied diffusion-based vMRE to noninvasively evaluate the viscoelastic properties of the brain. The results of this study are consistent with previously reported viscoelastic properties of the brain.17–21 The μdiff, which reflects brain stiffness, decreased with aging. The μdiff of WM was higher than that of GM but lower than that of deep structures. Differences in sex were observed only in the parietal lobe and basal ganglia, without a consistent pattern. The cerebellum showed a higher μdiff than the cerebrum, which contrasts with previous studies that used conventional MRE. The mean μdiff value obtained in this study was higher than the shear modulus values previously reported in conventional MRE studies. However, the absolute shear modulus of the brain measured by conventional MRE also has a wide range —from 0.6 to 13.6 kPa6,7,33—depending on factors such as the transducer type, the vibration frequency, and the post-processing methods used. A gold-standard method for quantifying brain viscoelasticity has not yet been established. To our knowledge, no previous studies have reported the viscoelastic properties of normal brain tissue measured using vMRE, making direct comparison challenging. Nonetheless, previous studies30,31 that applied the same equation and software in pituitary adenomas and cervical lymph nodes have reported μdiff values ranging from 6.9 to 10.5 kPa and from 5.68 to 6.86 kPa, respectively. Considering that tumorous tissues generally exhibit greater stiffness than normal tissues,7,22 the mean whole brain value of 4.9 kPa observed in our study appears to be within a comparable range. Overall, our results followed the general trend of brain stiffness observed with conventional MRE, suggesting the possibility of applying vMRE to the brain, although we could not directly compare the shear modulus from conventional MRE in the same patient group. Although sADC-based vMRE was first proposed in a liver cirrhosis model,23,24 it has been widely applied to assess tissue elasticity in various organs, including the liver, breast, placenta, pituitary gland, cervical lymph nodes, and salivary gland masses.25,27–32 In this study, we used b-values of 200 and 1500 s/mm2 to calculate sADC, which inherently includes both intravoxel incoherent motion (IVIM) effects and non-Gaussian diffusion. This sADC has been theoretically and experimentally optimized to maximize differential sensitivity to tissue microstructures. Therefore, there is no reason to believe that the relationship between tissue elastic properties and microstructure is weaker in brain tissue than in other soft tissues.23–27,34,35 We adopted calibration parameters from a previous liver parenchymal study23,24; however, the conversion factor for transforming sADC into shear modulus may differ across organs and tissue conditions,25 emphasizing the need for brain-specific calibration appropriate calibration factors for the brain tissue. In particular, the intercept term (β = 14.0 kPa) used in the regression equation should not be interpreted as a physiological minimum stiffness of the tissue. In the original liver-based model, both α and β were derived from empirical regression within a limited diffusion range, and the intercept represents a mathematical constant of the fitted model rather than a directly measurable baseline stiffness value.23 Because sADC values in biological tissues do not approach zero under physiological conditions, extrapolation of the intercept beyond the original regression range is not physiologically meaningful. When applied within the observed sADC range of brain tissue in this study, the resulting μdiff values fall within ranges reported by conventional MRE,6,7 supporting the mathematical applicability of the model despite organ-specific differences. Furthermore, unlike liver MRE—standardized through the Quantitative Imaging Biomarkers Alliance (QIBA) Profile36 of the Radiological Society of North America (RSNA)—no standardized protocol currently exists for brain MRE. For example, while QIBA specifies the use and placement of a cylindrical passive disc driver for liver MRE, brain MRE studies have employed various vibration systems such as passive pillow drivers, gravitational transducers, head cradles, and bite bars.6,37–39 This diversity contributes to substantial heterogeneity in stiffness measurement. Establishing a standardized conventional brain MRE protocol is therefore essential before any direct comparison with vMRE. Although deriving brain-specific conversion factors through such comparison in the same patient population would be ideal, it remains impractical at present due to the lack of methodological standardization. Our ongoing research aims to define standardized μdiff values using vMRE in a large population, which are expected to serve as reliable surrogate markers of brain viscoelastic properties and facilitate comparative studies across various disease conditions. Moreover, the inherent limitations of conventional MRE—such as high cost, the need for external vibration equipment, relative invasiveness, long scan times, and lower spatial resolution—make recruiting healthy participants for direct comparison difficult. For these reasons, previously published vMRE studies27–32 conducted in organs other than the liver, including the brain, head and neck, breast, and placenta, have also applied the same conversion factors originally derived from liver studies without organ specific validation. Sex differences of viscoelastic property in brain have shown inconsistent results across previous studies,17,18,20 similar to our observations. Some studies found no significant differences between males and females, whereas others observed sex-related variations that also differed markedly across brain regions. Possible explanations include differences in cortical thickness, GM-WM composition, and CBF.40,41 However, the underlying mechanisms remain unclear, and further investigations with larger cohorts are needed to clarify these findings. It has been demonstrated in previous research19,21,22 that the brain exhibits regionally distinct mechanical properties, which are thought to arise from differences in neuronal tissue composition and microstructural organization. For instance, GM may appear stiffer at the single-cell level due to the abundance of neuronal cell bodies; however, WM shows greater stiffness macroscopically owing to anisotropic organization of densely aligned axonal bundles.19,42 Deep GM structures, including the basal ganglia and thalamus, contain dense cellular architecture with more organized fibers and higher degrees of myelination. In particular, the globus pallidus features a repetitive geometric arrangement of dendrites and iron bound to metalloproteins. The relatively higher stiffness of the putamen may also be associated with its greater CBF, which is supplied directly by branches of the middle and anterior cerebral arteries.19,43,44 This pattern of stiffness—where WM tends to be stiffer than GM but softer than deep GM regions—has been repeatedly demonstrated in prior in vivo and tissue level studies,19,21,22,45 which is also consistent with the results of our study. Similar to sex-related differences, findings regarding lobar differences have varied substantially across studies.7,17,21,46 These discrepancies are likely attributable to differences in analytical methods and study populations, and further validation in large cohorts is warranted. Although studies have reported varying results regarding regional and sex-related differences, there is broad consensus across multiple studies that brain stiffness decreases with age.8,15–18 Our study also demonstrates a decline in brain stiffness with aging. This softening is thought to reflect changes in brain microstructure due to the reduction in neuronal cell number and size associated with aging. Previous reports have shown that the neuron-to-glial cell ratio decreases in healthy aging populations.47–50 An experimental study42 found that glial cells are softer than neuronal cells, supporting the hypothesis that an increasing glial cell proportion contributes to brain softening with age. Such age-related changes in viscoelastic properties of the brain should be taken into account when comparing different disease groups in future studies. This study has some limitations. First, the study population was too small to be representative of the general population. Further studies with larger sample sizes and direct comparisons with conventional MRE findings are required to generalize these results. Second, we used a conversion factor that was originally established for the liver, and further investigation and calibration of the conversion factor specifically for brain tissue are required. Nevertheless, vMRE offers several advantages over conventional MRE, including a shorter scan time, relatively superior spatial resolution, and a simpler analysis process. These advantages make it feasible to conduct large sample size studies or multicenter research, which were previously difficult to achieve. Therefore, the μdiff derived from vMRE has the potential to be used as a surrogate marker in various clinical applications. Conclusion In conclusion, diffusion-based vMRE effectively characterized the viscoelastic properties of the brain and showed aging-related stiffness changes consistent with conventional MRE findings. As the brain ages, it tends to soften, though these changes vary across regions. These results indicate that vMRE is a promising, noninvasive method for assessing brain mechanical properties. However, because the current conversion factors are derived from liver studies and no standardized protocol exists for brain MRE, further calibration and validation through direct comparison with conventional methods are required to enhance the quantitative accuracy and clinical utility of vMRE. Funding Statement This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT [MSIT]) (No. 2022R1F1A1074513). Acknowledgments: We would like to thank Jinwoo Hwang, PhD, a clinical scientist at Philips Korea, for technical support and Editage (www.editage.co.kr) for English language editing. Conflicts of Interest: The authors declare that they have no conflicts of interest. Supplementary Materials: Supplementary Fig. Imaging analysis workflow and ROI segmentation for cerebral GM and WM. (a) Background b 0 image obtained from diffusion-weighted imaging. (b) Automatic removal of the skull and background signals; the removed regions are shown in green. (c) Voxel-wise μdiff color map generated from the diffusion-based vMRE analysis. (d) A representative axial slice immediately above the lateral ventricles was selected on the b 0 image. A freehand ROI (red line) was drawn to define the cerebral region, and extra-cerebral space was removed using semi-automatic segmentation (green areas indicate removed regions). (e) Semi-automatic segmentation (red area) of cortical GM within the selected slice, from which the μdiff values were extracted. (f) Semi-automatic segmentation (red area) of WM within the same slice for extraction of the corresponding μdiff values.GM, gray matter; vMRE, virtual magnetic resonance elastography; WM, white matter. Supplementary Materials Supplementary figures are available online. References