Paraspinal muscle quality influences outcomes in spine surgery, and intramuscular fat infiltration is an important marker of degeneration. The authors aimed to determine whether simple elliptical Hounsfield unit (HU) measurements placed on routine CT images correlate with MRI-based intramuscular fat estimates (T2-hyperintense area fraction) and to define practical HU screening thresholds usable within standard PACS workflows.
This retrospective cohort included 136 patients who underwent both preoperative CT and MRI within an 8-week interval. The sample comprised 71 men and had a mean age of 61.0 ± 15.4 years. All imaging data were de-identified; institutional review board approval was obtained and individual informed consent was waived.
Elliptical regions of interest were defined on CT images at two lumbar levels: L2/3 and L4/5. HU values were measured for three muscles at each level: erector spinae, multifidus, and psoas major. The ROI approach was intentionally simple so that measurements could be placed quickly using standard picture archiving and communication systems (PACS) without specialized software.
The reference standard for intramuscular fat infiltration was the T2-hyperintense area fraction on MRI. The authors used Otsu thresholding to segment T2-hyperintense areas and to compute a percent area fraction as the surrogate measure of fat infiltration for each muscle at the corresponding levels.
Reproducibility of measurements was assessed. Intraclass correlation coefficients (ICCs) were 0.709 for HU measurements and 0.857 for the MRI T2 fraction, indicating moderate-to-good consistency. Visual grading using the Goutallier system produced a weighted kappa of 0.579.
The association between HU and T2-hyperintense fraction was evaluated using linear mixed-effects (LME) models, accounting for repeated measures and clustering by patient and level. Level-specific receiver operating characteristic (ROC) analyses were performed on the lower HU value side for each muscle-level (n = 136 per muscle-level) to identify HU thresholds corresponding to predefined infiltration criteria of ≥30% and ≥50% T2-fraction.
Across the overall LME, there was a significant inverse relationship: β = -0.880 HU per 1% increase in T2-fraction (95% confidence interval -0.935 to -0.825). The model’s marginal R2 was 0.502, indicating that HU explained a meaningful portion of between-subject variance in T2-fraction.
The slope of the HU–T2 association differed by muscle. Multifidus showed the steepest negative association (β = -1.020), followed by erector spinae (β = -0.753). The psoas major demonstrated a weak correlation (reported R = -0.226) and was excluded from subsequent ROC threshold analyses because of poor discrimination.
ROC analyses were used to identify level- and muscle-specific HU cutoffs for screening. The authors report that HU cutoffs differed by approximately 20 HU between L2/3 and L4/5. The discrimination of HU thresholds was higher when targeting the ≥50% T2-fraction criterion than the ≥30% criterion, indicating more reliable identification of severe fat infiltration.
The key practical finding is that simple elliptical ROI HU measurements on routine CT correlate robustly with MRI-derived T2-hyperintense fat fraction for erector spinae and multifidus. Because the ROI method can be performed within standard PACS, it may enable preoperative or opportunistic screening for paraspinal muscle fat infiltration without additional imaging, specialized software, or extra radiation beyond the CT already obtained for surgical planning.
Level- and muscle-specific thresholds were identified, recognizing anatomical variation between L2/3 and L4/5 and stronger HU–fat relationships in multifidus than erector spinae. The study suggests that using these thresholds could streamline risk stratification or inform perioperative planning where muscle quality is relevant.
The authors excluded psoas major from ROC threshold derivation because of a weak correlation with MRI T2-fraction (R = -0.226). Other limitations commonly relevant to retrospective imaging studies—such as potential selection bias, variability in CT acquisition parameters, and restriction to a single-institution cohort—are implicit; specific additional limitations were not detailed beyond reported metrics. All data are available from the authors on reasonable request.
Elliptical ROI-based HU measurements applied at L2/3 and L4/5 correlate with MRI T2-hyperintense area fraction and can screen for intramuscular fat infiltration in erector spinae and multifidus. Proposed HU thresholds vary by level (approximately 20 HU difference) and perform better for identifying ≥50% infiltration. This approach offers a practical, cost-neutral screening option using standard PACS in preoperative spine patients.