Magnetic resonance imaging-derived vertebral bone quality correlates with computed tomography-derived bone density and paraspinal muscle degeneration: a systematic review and meta-analysis

Protocol registration and reporting framework

This systematic review and meta-analysis was designed and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines to ensure methodological transparency and reproducibility. The study protocol was registered with PROSPERO (registration number: CRD420251238599).

Eligibility criteria

Original studies involving human subjects aged ≥18 years were eligible for inclusion if they reported quantitative associations between: (1) MRI-derived VBQ indices (the VBQ score or closely related variants) and CT-derived vertebral attenuation measured in HU; and/or (2) bone quality metrics (VBQ or HU) and paraspinal muscle characteristics, including measures of muscle size (CSA, FCSA, or relative CSA) and muscle composition (fat infiltration or fat fraction). Studies were considered if they provided extractable quantitative association data, such as correlation coefficients or regression parameters linking these metrics. Non-original articles (literature reviews, editorials, commentaries), animal or cadaveric studies, and studies lacking extractable quantitative data were excluded.

Information sources and search strategy

A comprehensive literature search was conducted in the PubMed, Embase, Scopus, and Web of Science databases for all literature published until December 9, 2025, targeting bone, muscle, and imaging modalities. The following search strings were used for the literature search: ((“vertebral bone quality” OR “endplate bone quality”) OR VBQ OR EBQ OR MVBQ OR “modified vertebral bone quality” OR “MRI-based vertebral bone quality” OR ((vertebr* OR endplate* OR spinal OR spine) AND (“bone quality” OR “bone marrow fat” OR “trabecular bone”) AND (score* OR scoring OR metric*))) AND (“paraspinal muscle*” OR “paravertebral muscle*” OR multifidus OR “erector spinae” OR psoas) AND (MRI OR “magnetic resonance”)) OR ((“Hounsfield unit*” OR HU) AND (MRI OR “MR imaging”) AND (“vertebral bone quality” OR “fat infiltration” OR “myosteatosis” OR CSA)). Reference lists of included studies were also manually screened to identify any additional eligible records.

Study selection

All retrieved records were imported into a reference management database; duplicates were removed prior to screening. Two reviewers (M.A. and A.B.) independently screened titles and abstracts of all studies for relevance, followed by full-text assessment of potentially eligible studies. Any discrepancies were resolved through discussion and mediation with a third reviewer. The PRISMA flowchart for the study selection process is presented in Fig. 1.

Fig. 1

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of study selection. PRISMA 2020 flow diagram illustrating the identification, screening, eligibility assessment, and inclusion of studies evaluating associations between magnetic resonance imaging-derived vertebral bone quality, computed tomography-derived Hounsfield units, and paraspinal muscle characteristics. A total of 25 studies were included in the quantitative synthesis.

Data extraction

Data extraction was performed independently by two reviewers (M.A. and A.B.) using a standardized data collection form. The following variables were extracted: study design, country, sample size, population characteristics, spinal region and vertebral levels analyzed, MRI acquisition details and VBQ methodology, CT acquisition and HU measurement protocols, paraspinal muscle metrics (CSA, FCSA, relative CSA, and fat-related measures), and reported association statistics. Any disagreements in data were resolved by consensus, with input from a third reviewer (F.R.) when necessary.

Risk of bias assessment

We used the Newcastle–Ottawa Scale (NOS) for assessing the methodological quality and risk of bias in the included studies. All evaluations were conducted independently by two reviewers (M.A. and A.B.), with additional input from a third reviewer (F.R.) to resolve any discrepancies. Studies were categorized according to standard NOS criteria across selection, comparability, and outcome domains.

Outcomes

The primary outcomes of interest were quantitative associations between: (1) MRI-derived VBQ (and related indices) and CT-derived HU; (2) VBQ and paraspinal muscle characteristics, including muscle size and fat infiltration; and (3) CT-derived HU and paraspinal muscle characteristics.

Statistical analysis

The meta-analysis was performed using R ver. 4.3.2 (R Core Team, 2023, “meta” package; The R Foundation for Statistical Computing, Vienna, Austria). Correlation coefficients reported in the included studies were transformed into Fisher’s Z values before pooling. Random-effects models were fitted using restricted maximum likelihood, with Hartung-Knapp adjustment applied to generate robust confidence intervals (CIs). Pooled estimates were subsequently back-transformed to Pearson’s correlation coefficients (r) for interpretability. Statistical heterogeneity was assessed using the I² statistic and Cochran’s Q test. For studies reporting Spearman’s rank correlation coefficients, values were converted to Pearson’s “r” using the established approximation r=2×sin (ρπ/6) to allow consistent pooling [7]. To maintain statistical independence, multiple dependent correlation coefficients derived from the same cohort (such as in studies reporting separate correlations for multifidus and erector spinae muscles) were combined into a single composite effect size by averaging Fisher’s Z values. Conversely, when independent subgroup data (e.g., sex-stratified cohorts) were reported, weighted averages were calculated using inverse-variance weighting prior to pooling.

Prespecified subgroup analyses were conducted based on spinal region (cervical vs. lumbar) and muscle metric definitions (CSA vs. FCSA) when sufficient data were available. Publication bias and small-study effects were evaluated using contour-enhanced funnel plots and Egger’s regression test when the number of contributing studies met accepted thresholds.

Regression synthesis using reconstructed individual participant data

To enhance the clinical applicability of the relationship between MRI-derived VBQ and CT-derived HU, individual participant data were reconstructed from published scatter plots using WebPlotDigitizer [8]. Digitized data were validated by comparing them with the original study-level correlation coefficients before pooling. Extracted data points were then aggregated, and both study-specific and pooled simple linear regression models were fitted to estimate the overall VBQ–HU relationship.

Association between MRI-derived VBQ and CT-derived HU

Across all included studies, MRI-derived VBQ demonstrated a moderate inverse association with CT-derived vertebral attenuation. The overall random-effects pooled correlation coefficient was r=−0.55 (95% CI, −0.65 to −0.43), indicating that higher VBQ values were consistently associated with lower HU values. Substantial heterogeneity was observed among the studies included (I²=93.2%).

Subgroup analyses by spinal region yielded similar results. In cervical spine cohorts, the pooled random-effects correlation analysis yielded an “r” value of −0.53 (95% CI, −0.68 to −0.33; I²=88.7%), whereas lumbar spine cohorts demonstrated a pooled correlation “r” of −0.57 (95% CI, −0.73 to −0.37; I²=95.4%). There was no statistically significant difference between the cervical and lumbar spine subgroups (p>0.05), suggesting a consistent VBQ–HU relationship across spinal regions (Fig. 2) [9–17,20–22,28,30,31].

Fig. 2

Association between magnetic resonance imaging-derived vertebral bone quality (VBQ) and computed tomography (CT)-derived Hounsfield units (HU) by spinal region. Forest plot showing pooled correlation coefficients between VBQ and CT-derived HU, stratified by cervical and lumbar spine cohorts. Individual study estimates are displayed with 95% confidence intervals (CIs), and pooled effects are shown using both common-effect and random-effects models. VBQ demonstrated a consistent moderate inverse correlation with HU across spinal regions. Heterogeneity statistics and subgroup difference tests are provided. df, degree of freedom.

Association between VBQ and paraspinal muscle area (CSA/FCSA)

VBQ showed a small but statistically significant negative association with the lumbar paraspinal muscle area. Pooled data from studies reporting CSA and FCSA yielded an overall random-effects correlation “r” of −0.19 (95% CI, −0.35 to −0.02), with moderate heterogeneity (I²=71.8%).

Subgroup analyses based on muscle metric definition demonstrated comparable effect sizes. Studies reporting conventional CSA yielded a pooled correlation “r” of −0.19, while those reporting functional CSA demonstrated a similar association (r=−0.19). No significant subgroup differences were detected, indicating that higher VBQ values were moderately associated with reduced paraspinal muscle size regardless of metric definition (Fig. 3) [6,23–26,29].

Fig. 3

Association between vertebral bone quality (VBQ) and paraspinal muscle cross-sectional area (CSA). Forest plot depicting correlations between VBQ and paraspinal muscle size, stratified by muscle metric definition: CSA and functional cross-sectional area (fCSA). Individual and pooled correlation coefficients (with 95% confidence intervals [CIs]) are shown using random-effects models. Higher VBQ values were associated with modestly smaller paraspinal muscle area, with no significant difference between CSA and fCSA subgroups. df, degree of freedom.

Association between VBQ and paraspinal muscle fat infiltration

Five studies evaluated muscle composition; pooled analysis of data for VBQ exhibited a strong positive association with paraspinal muscle fat infiltration. The pooled random-effects correlation coefficient “r” of 0.63 (95% CI, 0.10–0.88), indicating that higher VBQ values were associated with greater degrees of myosteatosis. However, considerable heterogeneity was observed among these studies (I²=97.7%) (Fig. 4) [6,13,23,25,26].

Fig. 4

Association between vertebral bone quality (VBQ) and paraspinal muscle fat infiltration. Forest plot summarizing correlations between VBQ and paraspinal muscle fat infiltration. Individual study estimates and pooled common-effect and random-effects models are shown. VBQ demonstrated a strong positive association with muscle fat infiltration, indicating greater myosteatosis with poorer VBQ. Substantial between-study heterogeneity was observed. CI, confidence interval.

Association between CT-derived HU and paraspinal muscle metrics

Compared with VBQ, CT-derived HU demonstrated directionally opposite associations with muscle size and composition. In studies evaluating muscle fat infiltration, HU was negatively correlated with fat infiltration (common-effect pooled correlation r=−0.70; 95% CI, −0.75 to −0.64). The corresponding random-effects estimate showed wider uncertainty due to substantial heterogeneity (I²=96.4%).

Regarding muscle size, HU showed a positive association with paraspinal muscle CSA. The pooled random-effects correlation “r” across four studies reporting CSA was 0.40 (95% CI, 0.04–0.67), with high heterogeneity (I²=88.8%). These findings mean that higher vertebral attenuation is associated with greater muscle bulk and lower muscle fat infiltration.

Regression synthesis using reconstructed individual participant data

Individual participant data were reconstructed from scatter plots published in three eligible studies [9–11]. After analysis, the digitized data closely reproduced the originally reported study-level correlation coefficients, supporting the validity of the reconstruction process. Study-specific linear regressions demonstrated consistent inverse relationships between VBQ and HU, with slopes ranging from −33.15 HU to −72.32 HU per 1-point increase in VBQ (r=−0.37 to −0.54). When all reconstructed data points were pooled, a single linear regression equation was derived to compute HU based on VBQ value (r=−0.352): HU=−50.9×VBQ+378.

The pooled regression model revealed a moderate inverse relationship between MRI-derived VBQ and CT-derived HU, serving as an exploratory cross-modality translation model (Fig. 5) [9–11].

Fig. 5

Regression relationship between vertebral bone quality (VBQ) and computed tomography (CT)-derived Hounsfield units (HU) using reconstructed individual participant data. Scatter plot showing reconstructed individual participant data derived from published studies evaluating VBQ and CT-derived HU. Colored points represent individual studies, with study-specific linear regression lines. The dashed line represents the pooled regression model. The derived exploratory equation (HU=−50.9×VBQ+378) demonstrates a moderate inverse relationship between magnetic resonance imaging-based VBQ and CT-based vertebral attenuation.

Risk of bias and publication bias

All included studies had generally good methodological quality, with total NOS scores ranging from 7 to 9 (Table 2, Supplement 1) [6,9–31]. The most common limitations involved incomplete adjustment for potential confounders in the comparability domain.

Table 2

Risk of bias assessment using the NOS

Funnel plots for assessing small-study effects in the VBQ–HU meta-analysis did not reveal significant asymmetry (Egger’s test p=0.216) (Fig. 6). For other outcomes, formal testing was not performed due to the limited number of contributing studies.

Fig. 6

Funnel plots assessing small-study effects. Contour-enhanced funnel plots evaluating potential publication bias and small-study effects for the primary and secondary analyses: (A) vertebral bone quality (VBQ) vs. Hounsfield units (HU), (B) VBQ vs. paraspinal muscle cross-sectional area (CSA), (C) VBQ vs. muscle fat infiltration, (D) HU vs. paraspinal muscle CSA, and (E) HU vs. muscle fat infiltration. No significant asymmetry was observed for the primary VBQ–HU analysis.

VBQ and CT-HU: cross-modality convergence

The most significant finding of this study is the consistent inverse relationship between VBQ and HU observed across spinal regions. The pooled correlation (r=−0.55) indicates moderate-to-strong convergence between MRI- and CT-based assessments of VBQ, with similar effect sizes observed in the cervical and lumbar subgroups. This supports the construct validity of VBQ as a bone-related imaging biomarker, aligning with previous single-center studies demonstrating an inverse association between VBQ and HU [6,9–12].

From a clinical perspective, this finding reinforces the potential value of VBQ as an opportunistic screening tool for osteoporosis assessment derived from routine MRI, particularly in patients who may not otherwise undergo CT-based bone assessment; however, the substantial heterogeneity observed across studies underscores the fact that VBQ and HU are not interchangeable. As emphasized in previous narrative reviews, VBQ represents relative marrow signal intensity; consequently, it is influenced by MRI acquisition parameters, reference region selection, and marrow composition rather than calibrated mineral density alone [32]. Thus, VBQ can identify patients at risk of compromised bone quality, whereas CT-HU is the primary marker for measuring absolute bone strength estimation or detailed osseous planning.

Bone–muscle coupling: VBQ, muscle size, and myosteatosis

A key finding of this meta-analysis is that VBQ aligns more strongly with paraspinal muscle composition than with muscle size. In the analysis, VBQ showed only a small negative correlation with paraspinal muscle CSA/FCSA, whereas the association with muscle fat infiltration was moderate-to-strong and directionally consistent across studies. This indicates that, as an imaging marker, VBQ preferentially reflects qualitative muscle degeneration (myosteatosis) rather than gross reductions in muscle bulk.

This distinction is particularly relevant because in clinical settings, muscle size may be preserved despite substantial fatty infiltration, especially in older or frail patients; however, myosteatosis has been more consistently associated with impaired function, frailty, and adverse surgical outcomes [2,19]. Furthermore, the strong association observed between VBQ and fatty infiltration supports the concept that elevated VBQ identifies a systemic osteosarcopenic phenotype involving fatty changes in the muscle tissue rather than isolated low bone density. This may also explain why VBQ has significant prognostic relevance for mechanical complications, such as cage subsidence, screw loosening, and junctional failure, reported in previous studies focused on surgical outcomes [6,33,34].

CT-HU and paraspinal muscle metrics

The complementary analyses exploring the association between HU and paraspinal muscle characteristics further corroborate the existence of the bone–muscle unit framework. Higher HU values were associated with larger muscle CSA and lower fat infiltration, a relationship opposite in direction to VBQ. This implies that both imaging modalities capture biologically coherent aspects of musculoskeletal health, albeit through different signal mechanisms—HU appears to be more strongly associated with muscle size than VBQ, whereas VBQ demonstrates a stronger association with muscle fat infiltration. This divergence may arise from the fundamental differences between mineral-based attenuation observed on CT versus marrow fat-sensitive signal changes on MRI. Such modality-specific observations suggest that VBQ and HU provide complementary rather than redundant information, underscoring the importance of a combined interpretation to enhance musculoskeletal risk assessment.

Exploratory VBQ–HU translation

To improve interpretability across imaging modalities, we reconstructed individual participant data to formulate an exploratory pooled regression equation correlating VBQ to HU. While this approach allowed visualization of cross-study trends and yielded a plausible translation model, the results should be interpreted cautiously because digitization introduces measurement errors, and heterogeneity in imaging protocols limits generalizability. Nevertheless, the consistency of study-specific slopes supports the hypothesis that marrow-based deterioration captured by VBQ may be parallel to and potentially precede mineral loss reflected by HU.

Limitations

First, most included studies were retrospective and observational, and there was substantial heterogeneity across imaging protocols, VBQ definitions, HU measurement techniques, and muscle assessment methods. Muscle metrics were particularly heterogeneous, which may explain the weaker and more variable associations observed for muscle size than fat infiltration. Second, causality could not be inferred from the reported correlations. Finally, while publication bias was not detected for the primary VBQ–HU analysis, the number of studies was insufficient for robust bias assessment in the secondary analyses.