Automated Deep-Learning Quantification of Intramuscular Fat in Lumbar Spine Muscles on Dixon MRI: Validation and Normative Reference Values from 173 Healthy Adults

Automated Deep-Learning Quantification of Intramuscular Fat in Lumbar Spine Muscles on Dixon MRI: Validation and Normative Reference Values from 173 Healthy Adults | Research Square Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Automated Deep-Learning Quantification of Intramuscular Fat in Lumbar Spine Muscles on Dixon MRI: Validation and Normative Reference Values from 173 Healthy Adults Germán Balerdi, Johann Henckel, Anna Di Laura, Alister J. Hart, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6591044/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 31 Mar, 2026 Read the published version in BMC Medical Imaging → Version 1 posted 7 You are reading this latest preprint version Abstract Background. Intramuscular fat (IMF) accumulation in lumbar‑spine muscles is associated with sarcopenia, low‑back pain and poorer surgical outcomes. Dixon MRI allows the measurement of IMF with voxel-wise fat fraction (FF) mapping, but quantitative use in research or clinics is limited by labour-intensive 3D muscle segmentation and by the absence of healthy reference data. We present a lightweight pipeline that requires a small number of labelled images yet delivers accurate segmentation and Dixon-based reference ranges for lumbar spine IMF in asymptomatic adults with diverse physical activity levels. Methods. Twenty-six Dixon scans (44.8 ± 14.3 years old; 12 males / 14 females) were manually annotated for psoas, iliacus, quadratus lumborum and the erector spinae + multifidus complex. A 3D U‑Net was trained with a realistic data augmentation scheme that combines small rigid rotations and non-linear B-spline deformations mimicking real anatomical variability, enhancing generalisation. Mean fat fraction (FF) of each muscle was extracted from Dixon FF images. Segmentation (Dice, relative volume difference, Hausdorff distance) and FF estimation accuracy was assessed with a five‑fold Monte‑Carlo cross‑validation and was compared to a multi-atlas approach. FF agreement with manual measurements was examined using linear regression and Bland–Altman analysis. The trained model then processed the complete study cohort of 173 healthy adults (20–70 years old; sedentary to highly active) to derive healthy subjects FF percentiles. Results. Automated FF strongly correlated with manual derived values (R² = 0.96), showing no systematic bias. The 3D U-Net achieved Dice coefficients of up to 0.936 (erector spinae + multifidus) and demonstrated superior performance over multi-atlas segmentation across all metrics (p < 0.01). The model was also able to capture different characteristic FF values for each muscle, with median (IQR) normative FF values of 21.8% (17.4–26.9%) for erector spinae + multifidus, 18.0% (14.9–22.2%) for quadratus lumborum, 13.4% (10.5–16.3%) for psoas and 11.6% (9.5–13.8%) for iliacus, with significant sex and activity differences (p < 0.01). Conclusion. With fewer than 30 manual annotations, a 3D U-Net model enhanced by anatomically realistic data augmentation, provides a reliable and accurate solution for automated IMF quantification from Dixon MRI. We deliver novel Dixon MRI reference values of lumbar spine muscle composition in healthy adults. Both the automated estimation of FF, and the reference FF values reported here, facilitates the study of sarcopenia, spinal pathologies, and muscle degeneration. MRI Lumbar Spine Muscle Health Deep Learning Intramuscular fat Musculoskeletal Quantitative MRI Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Background Fat infiltration and atrophy of the lumbar spine muscles are associated with sarcopenia and many spinal degenerative conditions, contributing to pain, functional limitation and poorer surgical outcomes [ 1 – 6 ]. Accumulation of intramuscular fat (IMF) (that is, fat within the muscle fascia) reduces muscle contractile capacity, as muscle fibres are replaced by non-contractile tissue [ 7 – 9 ], making quantitative IMF a clinically relevant biomarker. Accurate assessment tools are needed to aid diagnosis, treatment, and monitoring of muscle health [ 10 – 17 ]. Magnetic‑resonance imaging (MRI) is the reference standard for muscle evaluation. Dixon MRI is a quantitative sequence that enables the measurement of proton-density fat fraction (FF) that sensitively capture IMF once individual muscles are accurately segmented [ 18 – 22 ]. The use of quantitative MRI to study muscle health has grown steadily in recent years, both to study myopathies [ 23 – 25 ], as well as sarcopenia and muscle health in healthy subjects with different levels of physical activity [ 22 , 26 – 31 ]. The major obstacle is that 3D, muscle‑specific segmentation remains a labour‑intensive manual task, limiting both clinical adoption and large‑cohort research. Deep learning, particularly convolutional neural networks (CNNs), such as U-Nets [ 32 ], has revolutionised medical image segmentation, achieving high accuracy in various applications [ 33 ]. Previous studies have demonstrated its effectiveness in segmenting musculoskeletal structures, including the psoas and iliacus muscles [ 14 , 34 – 39 ]. However, the lack of large, annotated datasets remains a challenge in training robust deep learning models for muscle segmentation. Due to this, realistic and domain specific data augmentation techniques are required [ 40 , 41 ]. In addition, almost no studies report normative lumbar‑spine FF ranges in healthy adults, even though such baselines are essential for stratifying disease severity and monitoring interventions. This study proposes a novel approach that combines a 3D U-Net with deformation-based data augmentation techniques, needing only 26 manual annotations, to automatically quantify the fat infiltration of the lumbar spine muscles from Dixon MRI. After evaluating its performance with a 5-fold Monte‑Carlo cross‑validation, and comparing it with a multi‑atlas baseline, we deploy the network on 173 asymptomatic adults with different levels of physical activity and provide reference FF values for the psoas, iliacus, quadratus‑lumborum and erector‑spinae + multifidus muscles. Materials and Methods We implemented and evaluated a deep-learning method for fully automated 3D quantification of IMF in the lumbar spine muscles, trained with 26 3D Dixon MRI images from a study looking at muscle health and physical activity. It automatically segments the psoas (P), iliacus (I), quadratus lumborum (QL), and erector spinae and multifidus combined (ES + M) from in-phase Dixon MRI and computes the IMF in each muscle as the mean FF value for each muscle from the Dixon FF image. The method was then applied to the full study cohort to obtain reference IMF values for this muscle group. Study Cohort We analysed 173 Dixon MRI examinations acquired for a study of muscle health and physical activity. The study comprised 87 sedentary volunteers and 86 recreational cyclists who underwent axial Dixon MRI. Overall, 92 participants were female and 81 male, with a mean (SD) age of 43.5 (12.8) years and BMI of 25.7 (5.6) kg/m 2 . The images were acquired in a 3T scanner (Siemens Magneton Vida, Erlangen, Germany) using a body coil and a standardised protocol that includes a Dixon sequence. The scanning protocol consisted of an axial TSE Dixon sequence (slice thickness 1.5 mm, spacing between slices 1.95 mm, repetition time (TR) 4570 ms, echo time (TE) 45 ms, number of excitations 1, number of echoes 14, flip angle 120°), with a field of view (FOV) that covered axially from 1 cm below the lesser trochanter to the origin of the psoas muscle at the level of the L1 vertebra. The voxel size was 0.47 × 0.47 × 1.95 mm3. Manual ground truth A subset of 26 images was manually segmented by an expert operator using Simpleware ScanIP (Version 2024; Synopsys, Inc., Mountain View, USA). Segmented muscles were P, I, QL, and ES + M. The muscle masks included the intramuscular fat located within the fascia compartment. A special case to consider is the posterior region of the ES + M complex, where an epimuscular fat tent is often observed in subjects with fatty atrophy. In our segmentations, we followed the method described in [ 42 ], using the epimyseal border to delineate the posterior boundary, thereby excluding the epimuscular fat [ 43 , 44 ]. Figure 1 shows an example of manual segmentation, and Supplementary Fig. 1 illustrates the exclusion of the epimuscular fat. The twenty-six MRI scans correspond to 12 female and 14 male individuals with a mean (SD) age of 44.8 (14.3) years and a BMI of 26.8 (5.9) kg/m 2 . Seventeen of them were part of the sedentary group and nine were cyclists. Pre-processing and Data Augmentation All images were pre-processed to enhance the consistency of the input data. In this process, all images were rigid registered to a reference image with Simple Elastix using the normalised cross correlation as similarity metric [ 45 ]. The homogenised images were then downsampled to half the size in the axial plane to reduce the memory requirements of the model, which translated into a four-fold reduction in its memory size with respect to the original images, without a significant loss in image quality. As a result, all the images had voxel sizes of 0.94×0.94×1.95 mm 3 , a matrix size of 400x320x145 voxels and had the same geometrical space. This process is depicted in Fig. 2 -A. To improve model generalization, we applied data augmentation techniques to artificially increase the size of the training dataset. In addition to linear augmentation, here we propose a nonlinear deformation-based augmentation that accounts for the large anatomical variability in the abdominal and pelvic regions (Fig. 2 -B): Linear Augmentation: Clockwise and anticlockwise rotations of 5° and 10° were applied to the original and flipped images on the x axis (medial-lateral) to account for asymmetries and variability in the patient positioning. Nonlinear Augmentation: We generated deformed images by registering pairs of original images using B-spline nonlinear transformations to create realistic deformations that simulate anatomical variability. Nonrigid B-spline registration was implemented using SimpleElastix [ 45 ] with normalized cross-correlation (NCC) as similarity metric and the adaptive stochastic gradient descent algorithm with 500 iterations and 2048 samples. The number of iterations was selected to stop early in the registration process, creating virtual anatomies between the two images being registered. Supplementary Fig. 2 shows two examples of this data augmentation process. The data augmentation process was applied only to the training set of each of the trained models. 3D U-Net Architecture We implemented a CNN with a 3D U-Net architecture. The network consists of an encoder-decoder structure with skip connections, as proposed by Ronneberger et al [ 32 ]. The input to the model was the pre-processed 3D MRI Dixon in-phase images, and the output was a multi-label binary mask with hot encoding representing the segmentation of each muscle. The architecture consisted of 4 encoder and 4 decoder layers, with 16 convolutional filters in the first layer. It had a total of 2.5 million parameters. Trained models The 3D U-Net was evaluated using a five-iteration fixed-split Monte Carlo cross validation. In this method, the training and validation sets were randomly sampled five times. For each iteration, 18 3D images were used for training (approximately 70% of the data) and the remaining 8 were used for validation (approximately 30% of the data). This process was repeated five times, ensuring that each 3D image appeared once in the validation set. Data augmentation was applied only to the training subsets, generating a total of 486 augmented 3D images per fold (180 linearly and 306 nonlinearly augmented), as described previously. The models’ performance was evaluated on the validation set of each fold, and the results were averaged across the five folds to ensure robust evaluation of the segmentation method. All training and validation were performed on full 3D in-phase Dixon images. No validation image was used in training, either directly or indirectly through non-linear augmentation. The five cross-validation models were trained using binary cross entropy as loss function and the ADAM optimiser with a learning rate of 0.0001. The model was trained with early stopping to obtain the minimum loss for the validation set. Baseline: Multi-Atlas Segmentation We implemented a multi-atlas segmentation method as a baseline for comparison, as this type of methods works well with low numbers of manually segmented images. The multi-atlas approach was similar to our previous work for the gluteal muscles [ 46 , 47 ]. An 18‑atlas library (subset of the 26 labelled scans) was used in a leave‑one‑out scheme. To segment an image, every atlas in the library is registered to the target image to be segmented, which are then sorted in descending similarity order using the NCC as metric. The labels of the 5 most similar atlases are propagated to the target image space and subsequently fused into a single label for each muscle using majority voting [ 48 ]. For image registration, an affine followed by a non-rigid B-spline registration [ 49 ] was implemented using SimpleElastix [ 45 , 50 ]. NCC was used as a similarity metric and the cost function was minimised using the adaptive stochastic gradient descent algorithm [ 51 ] with 2000 iterations and 2048 samples. A pyramidal scheme of four layers with downsampling factors of 8, 4, 2 and 1 was employed to improve the registration. These parameters have previously been optimized in terms of segmentation performance. All multi-atlas segmentation labels were post-processed to improve the spatial consistency of the labels. Post-processing consisted of applying a soft-tissue intensity mask to remove subcutaneous fat voxels from the labels and then a applying morphological filter to fill holes in the labels. Fat Fraction Measurement We created FF images for each of the 3D scans of the dataset, which is the ratio image between the fat image and the sum of the water and fat Dixon images. For each muscle, the mean FF was then calculated by averaging the voxel values within the corresponding 3D muscle mask. All measurements were performed across the entire muscle volume, ensuring a comprehensive quantification of fat infiltration rather than relying on individual slices or regions. Evaluation Metrics Segmentation performance was evaluated using three complementary metrics: a) the dice similarity coefficient (Dice) to measure the overlap between the predicted segmentation and the ground truth. This metric is the most widely used overall metric for assessing segmentation performance, especially relevant for evaluating agreement in both volume and shape; b) the relative volume difference (RVD) that assesses the accuracy of the volume estimation, calculated as the percentage difference in total volume between the predicted and ground-truth segmentations; and c) the Hausdorff distance (HD), which measures the maximum surface-to-surface distance between the predicted and manual segmentation boundaries, capturing the worst-case discrepancy at the borders and being particularly useful for assessing the accuracy of shape delineation. These metrics were also computed for the multi-atlas method. Each metric offers complementary information: Dice evaluates overall overlap, RVD focuses on volumetric agreement, and HD emphasizes shape and surface accuracy. Note that Dice and RVD can be lower for smaller muscles due to higher relative sensitivity to minor boundary mismatches. To assess the accuracy of the fat fraction measurements, we used a Bland-Altman plot to detect systematic errors and fit a linear regression to automated versus manual scatter plots to assess the strength of the relationship. Both Bland-Altman and linear regressions were performed for the validation set of all cross-validation folds together, including repeated images in the validation set of different models. Metrics were computed on full masks and on masks cropped to the manually annotated axial extent to compensate for limited psoas coverage in some scans. This ensured fair and consistent comparisons between automated and ground truth measurements across all subjects. Statistical Analysis The mean (SD) was computed across folds for all the evaluation metrics. The median (IQR) was computed for the evaluation metrics of each fold of the 3D U-Net and the multi-atlas segmentations. For statistical comparison between the 3D U-Net and the multi-atlas method, we used a Kruskal-Wallis tests, with a significance level of p < 0.05. For the fat infiltration measurements, we computed the median (IQR) for the average of left and right FF values of each muscle. We compared the distribution of FF values between the manual and deep learning-based measurements for each muscle, using Kruskal-Wallis tests. In the full cohort, Type III ANCOVA assessed the contribution of age, sex, BMI and activity group to IMF. Partial eta squared (η²ₚ) were used to quantify the effects sizes of each covariate. Results Segmentation Accuracy Deep learning 3D U-Net segmentation achieved excellent performance in terms of segmentation accuracy when evaluated with a 5-random sample cross validation, which ensures robust assessment across the entire dataset. Table 1 summarises the average metrics for Dice, RVD, and HD across all models for the validation set, for each muscle group. Table 1 Mean (SD) segmentation accuracy metrics across folds for the 3D U-Net cross-validation models. Left Side Right Side P l QL ES + M P l QL ES + M Dice 0.927 (0.004) 0.897 (0.009) 0.871 (0.007) 0.936 (0.007) 0.923 (0.006) 0.885 (0.012) 0.857 (0.015) 0.936 (0.004) RVD [%] -2.6 (2.5) 2.1 (6.4) -2.1 (5.5) -0.8 (2.9) 0.2 (3.6) 0.4 (10.1) 1.2 (8.0) -1.5 (2.3) Hausdorff Distance 7.7 (0.9) 7.7 (1.0) 5.4 (0.5) 8.2 (0.7) 8.8 (1.4) 7.4 (1.3) 5.8 (0.7) 8.5 (1.8) The mean (SD) Dice score across folds was 0.904 for all muscle groups and side together, with the highest performance observed for the complex ES-M with 0.936 (0.005) and the lowest for the QL with 0.864 (0.011). In terms of RVD, the mean (SD) ranged from − 2.6% (2.5%) for the left psoas to 2.1% (6.4%) for the left iliacus, demonstrating reliable volume estimation. Because segmentation errors occurred primarily where psoas and iliacus merge, we also evaluated the composite iliopsoas. Dice and RVD improved to 0.920 (0.004) and 0.4% (4.9%), respectively. The HD was lowest for QL, being 5–9 mm for all muscles, indicating accurate boundary delineation. The lower HD but simultaneously lower Dice were due to the small size of QL. No significant differences were found in performance between sides. Comparison with Baseline The 3D U‑Net outperformed the multi‑atlas method on every metric (all p < 0.001). Figure 3 displays Dice (A) and RVD (B) distributions. The deep‑learning boxes pool 40 validation segmentations (8 scans × 5 folds), while multi‑atlas results are based on the 26 manually labelled scans. Figure 4 visually compares the segmentation performance of both methods in four representative subjects with varying levels of BMI and degrees of muscle fat infiltration. The deep learning method demonstrates superior capability in accurately segmenting regions where the multi-atlas approach struggled, primarily due to the significant anatomical variability observed in the lumbar spine muscles. Additionally, Fig. 4 -D specifically highlights the case with the lowest segmentation accuracy for the deep learning method. This instance, which belongs to the outliers for the quadratus lumborum (QL) muscle in Fig. 3 , clearly illustrates complete fatty infiltration in the subject's right QL muscle, which poses challenges for accurate automated segmentation. Fat Fraction Measurement Accuracy Automated and manual FF values were highly correlated with R² = 0.98 for the training set, and R² = 0.96 for the validation set, demonstrating excellent agreement and indicating robust generalisation capability of the deep learning models. Figure 5 -A shows a scatter plot comparing FF values derived from automated deep learning segmentation and manual segmentation in all lumbar spine muscles. The Bland–Altman analysis (Fig. 5 -B) showed negligible bias (0.1%) and narrow 95% limits of agreement (–2.0% to + 2.4%). Per‑muscle Bland–Altman plots (Supplementary Fig. 3) confirmed excellent agreement for psoas and ES + M, with slightly wider limits for iliacus and QL, and a single QL outlier (muscle with fatty atrophy in Fig. 4 -D). No systematic bias was observed for any muscle group, reinforcing the reliability of the deep learning approach. Moreover, no statistically significant differences were found for the FF distributions between automated and manual measurements for any of the muscle groups (p = 0.77 for P, p = 0.23 for I, p = 0.46 for QL, and p = 0.74 for ES + M). The median (IQR) FF values obtained from deep learning were 15.0% (12.9–16.5%) for P, 10.6% (8.8–12.7%) for I, 12.0% (10.4–13.6%) for QL, and 24.1% (19.0–27.6%) for ES + M; the corresponding values obtained from manual segmentation were 15.6% (12.9–16.1%), 10.0% (8.8–11.7%), 11.6% (10.6–13.3%) and 23.7% (19.0–26.6%), respectively. Fat fraction reference values and determinants in healthy adults Using the trained U‑Net on the entire cohort (N = 173), the median (IQR) FF values were 15.6% (13.5–17.7%) for P, 11.6% (10.0–13.8%) for I, 14.3% (12.7–16.8%) for QL, and 21.8% (17.9–27.8%) for ES + M respectively, showing distinct FF distributions across muscles, illustrated in Fig. 6 with violin plots. ANCOVA analysis revealed that age had the largest effect size on FF for QL (Partial η²=0.40) and ES + M (Partial η²=0.34), while being in the sedentary or physically active groups for P (Partial η²=0.40) and I (Partial η²=0.45). Sex was a significant contributor for FF for I, QL and ES + M, but with an intermediate effect. Collectively, these findings suggest that age and physical activity substantially contribute to the variance of the lumbar spine fat fraction, whereas BMI and sex appear less influential in this model. Table 2 reports the full results for the ANCOVA analysis for each muscle. Table 2 Results of the Type III ANCOVA analysis examining determinants of intramuscular fat fraction for each muscle. Muscle Significant Covariates F (df, 166) p Partial η² Psoas Group 37.31 (3, 166) p < 0.01 0.403 Age (years) 24.96 (1, 166) p < 0.01 0.131 BMI 23.49 (1, 166) p < 0.01 0.124 Iliacus Group 45.47 (3, 166) p < 0.01 0.451 Age (years) 37.63 (1, 166) p < 0.01 0.185 Sex 21.10 (1, 166) p < 0.01 0.113 BMI 47.17 (1, 166) p < 0.01 0.221 Quadratus Lumborum Group 16.63 (3, 166) p < 0.01 0.231 Age (years) 110.47 (1, 166) p < 0.01 0.400 Sex 20.55 (1, 166) p < 0.01 0.110 BMI 21.91 (1, 166) p < 0.01 0.118 Erector Spinae + Multifidus Group 18.72 (3, 166) p < 0.01 0.253 Age (years) 83.88 (1, 166) p < 0.01 0.336 Sex 64.42 (1, 166) p < 0.01 0.280 Table 3 presents muscle‑specific median (IQR) fat‑fraction reference values grouped by physical‑activity level (sedentary vs active) and by sex. Supplementary Fig. 4 visualises the FF distributions for these groups, panel A for activity group and panel B for sex. Table 3 Median (IQR) intramuscular FF in percentage by physical activity group and by sex. Muscle Sedentary (n = 87) Active (n = 86) Females (n = 92) Males (n = 81) Psoas 17.5 (16.0–19.1) 13.9 (12.8–15.2) 15.7 (13.3–17.9) 15.5 (13.7–17.0) Iliacus 13.7 (11.9–15.3) 10.2 ( 9.4–11.3) 11.9 (10.1–14.1) 11.3 ( 10.0–13.4) Quadratus Lumborum 16.2 (13.8–17.9) 13.2 (12.2–15.0) 15.0 (12.7–17.7) 14.0 (12.7–16.2) Erector Spinae + Multifidus 25.7 (19.7–29.6) 19.2 (17.4–22.8) 25.6 (19.4–29.8) 19.0 (17.2–23.3) Discussion This study demonstrates that a lightweight 3D U‑Net, trained with only 26 manual labels augmented by anatomically realistic deformations, can segment four lumbar spine muscles on Dixon MRI with high accuracy and automatically quantify FF, a measure of IMF, with manual level precision. The network not only outperformed a multi‑atlas baseline in all segmentation metrics but also allowed us to produce muscle specific reference FF ranges in a large healthy cohort, revealing strong effects of age and physical activity. Segmentation Performance Our results show that the 3D U-Net achieves high Dice similarity coefficients across all lumbar spine muscle groups, with the best performance for the P and ES + M muscles (Dices of 0.928 and 0.936 respectively) and slightly lower performance for the QL (Dice = 0.862). The relative volume difference (RVD) and Hausdorff distance metrics also confirm the robustness of our method. These results align with prior studies in musculoskeletal segmentation, which have reported that deep learning approaches outperform traditional methods in accuracy and reproducibility [ 14 , 16 , 38 , 52 ]. The slight performance decrease for the QL could be attributed to its irregular shape, and its relatively smaller size, making accurate segmentation more challenging. Another region where segmentations were prone to errors was where P and I join near their insertion, affecting mainly I dice scores. This region is particularly difficult to segment even for experts. In the electronic supplementary material, segmentations with and without this issue are available. Critical to this success was the nonlinear, muscle aware augmentation, which injected realistic anatomical variability and mitigated over fitting, a common issue when public lumbar muscle datasets are unavailable. This is particularly important in musculoskeletal MRI, where subject-specific anatomical differences can significantly affect segmentation performance. Our model achieved similar or better performance that other deep learning methods with larger manually segmented datasets [ 14 , 38 ]. Our approach demonstrates that, even with a limited dataset, a deep learning method can be successfully trained by implementing advanced, domain-specific, augmentation techniques. Comparison with Multi-Atlas Segmentation The multi‑atlas approach, a popular choice when labels are scarce, struggled with inter‑subject differences and high IMF cases. Our 3D U‑Net delivered markedly higher Dice scores and lower HD while reducing inference time from minutes to seconds, supporting real‑time use in large‑scale studies or clinical workflows. Fat Fraction Measure Accuracy and Clinical Relevance Accurate segmentation is only valuable if biomarkers, such as FF, are equally reliable. Mean FF within a muscle mask quantifies IMF, which is key biomarker for muscle health, sarcopenia, and spinal pathology [ 7 , 21 , 53 – 56 ]. Automated FF correlated tightly with manual measurements (validation R² = 0.96, bias ≈ 0%), with 95% limits of agreement within ± 2.4%. No muscle‑specific systematic errors emerged, even for outlier cases with complete fatty replacement of QL, underscoring the pipeline’s robustness. Reliable, fully automated IMF quantification makes longitudinal and large-scale assessments feasible. It offers an objective way to track age-related decline, disease progression, and responses to rehabilitation or pharmacological interventions, all settings in which small changes in FF can be clinically meaningful. Fat fraction reference values and determinants in healthy adults The deep‑learning pipeline captured all inter‑muscle and inter‑subject differences with the same reliability as manual segmentation. In the 173 subject cohort, physiological heterogeneity among lumbar muscles was also confirmed. The ES + M complex showed the highest baseline FF and iliacus the lowest, mirroring patterns reported for gluteal muscles [ 54 ]. Multivariable analysis identified age and a sedentary lifestyle as the strongest predictors of elevated FF. Sex added a moderate contribution, and BMI was least influential. These reference values provide new 3D normative data from a large cohort of healthy adults, contributing to the growing body of literature on lumbar muscle composition [ 57 , 58 ], and extending available 2D CSA-based values reported in previous studies [ 59 , 60 ]. When comparing the ES + M reference values to prior studies, it is important to consider differences in segmentation conventions. Specifically, there is ambiguity in the inclusion of posterior epimuscular fat tissue located between the epimyseal border and the lumbosacral fascia. In our study, we excluded this tissue based on anatomical considerations that define it as external to the erector spinae and multifidus compartments [ 42 ]. However, some published studies include this region in their segmentation of the paraspinal muscles, which may lead to higher reported FF values [ 43 , 44 ]. Because these conclusions were drawn from a large cohort spanning a wide activity spectrum, these muscle‑specific reference distributions fill a gap in the literature and can aid future sarcopenia or spine‑degeneration studies. Clinical and research implications Fully automated, reproducible IMF quantification enables longitudinal monitoring of spinal muscle health without the labour and variability inherent to manual labelling. Potential applications include early sarcopenia screening, risk stratification before spine surgery, and objective evaluation of exercise or pharmacologic interventions, where small FF changes are clinically meaningful. Limitations While our deep learning model demonstrates quantitative results, several limitations should be noted. First, the training dataset comprises only 26 manually labelled scans of healthy individuals, although diverse, only includes healthy individuals and remains relatively small compared to large-scale imaging repositories (which are not available for muscle imaging). While nonlinear augmentation broadened anatomical diversity, a substantially larger set that includes pathological cases will be essential for full generalisability. Second, all images were acquired with an identical Dixon protocol on a single 3T scanner. Therefore, external validation across vendors, field strengths, resolutions and pulse‑sequence variants is a key next step and a major challenge in quantitative muscle MRI, where multicentre datasets are scarce and Dixon implementations vary substantially across platforms. Third, although our goal was not to introduce a new deep learning architecture or method, other newer methods than a 3D U-Net could potentially achieve even better performance. Lastly, our segmentation approach excluded the posterior epimuscular fat located between the epimyseal border and the lumbosacral fascia. While this choice aligns with prior anatomical definitions, it may limit direct comparison with studies that include this tissue in paraspinal muscle measurements. Conclusion This study demonstrates that a 3D U-Net model, trained on just 26 manual labels and enhanced by anatomically realistic, nonlinear data augmentation, can accurately segment lumbar‑spine muscles and quantify muscle-specific FF from Dixon MRI. Applying the network on 173 asymptomatic adults spanning sedentary to highly active lifestyles, we generated novel normative FF reference ranges for these muscles in adults (ages 20–70) of both sexes. By reducing reliance on manual segmentation, this method enables large-scale investigations of sarcopenia, spine disorders, and muscle degeneration, with valuable applications in both research and clinical contexts. Declarations Ethics approval and consent to participate The study was approved by the UCL Research Ethics Committee (REC) [Number 13823 /001] and all procedures were conducted in accordance with the Declaration of Helsinki and its later amendments. All subjects provided written informed consent. Consent for publication Not applicable. Competing Interests German Balerdi, Johann Henckel, Anna Di Laura, Alister Hart, and Martin A. Belzunce declare no competing interests. Funding This research study was funded by the Arthroplasty for Arthritis Charity, the Trustees of the London Clinic Charity, the Maurice Hatter Foundation, the RNOH Charity, the Rosetrees Trust, and the Stoneygate Trust. Author Contribution M.A.B. conceived the study, supervised the analysis, and wrote the main manuscript draft. G.B. implemented the deep learning pipeline, conducted the data analysis, and contributed to writing. J.H. and A.D.L. conceived the study, coordinated subject recruitment and imaging, and reviewed the manuscript. A.J.H. conceived the study, provided clinical oversight, resources, and critical manuscript revisions. All authors reviewed and approved the final manuscript. Acknowledgement This research study was funded by the Arthroplasty for Arthritis Charity, the Trustees of the London Clinic Charity, the Maurice Hatter Foundation, the RNOH Charity, the Rosetrees Trust, and the Stoneygate Trust and supported by researchers at the National Institute for Health Research University College London Hospitals Biomedical Research Centre. Data Availability Part of the data is available from the corresponding author on reasonable request. References D’Hooge R, Cagnie B, Crombez G, et al. Increased intramuscular fatty infiltration without differences in lumbar muscle cross-sectional area during remission of unilateral recurrent low back pain. Man Ther. 2012;17:584–8. https://doi.org/10.1016/j.math.2012.06.007 . Köhli P, Schönnagel L, Hambrecht J, et al. The relationship between paraspinal muscle atrophy and degenerative lumbar spondylolisthesis at the L4/5 level. Spine J. 2024;24:1396–406. https://doi.org/10.1016/J.SPINEE.2024.03.016 . Guven AE, Schönnagel L, Camino-Willhuber G, et al. Relationship between facet joint osteoarthritis and lumbar paraspinal muscle atrophy: a cross-sectional study. J Neurosurg Spine. 2024;41:360–8. https://doi.org/10.3171/2024.4.SPINE231018 . Wang Z, Zhao Z, Li Z, et al. Fatty Infiltration in Paraspinal Muscles: Predicting the Outcome of Lumbar Surgery and Postoperative Complications. World Neurosurg. 2024;190:218–27. https://doi.org/10.1016/J.WNEU.2024.07.074 . Yun JH, Lee DG. Physical Functional Ability and Quantitative Assessment of the Multifidus Muscle of the Lumbar Spine in the Elderly. Diagnostics (Basel Switzerland). 2023;13. https://doi.org/10.3390/DIAGNOSTICS13142423 . Shen M, Shen Z, Yang G, et al. The Differences on the Fatty Infiltration of Paraspinal Muscles between Single- and Multiple-level Intervertebral Disc Degeneration in Patients with Lumbar Disc Herniation. Orthop Surg. 2024;16:1999–2010. https://doi.org/10.1111/OS.14101 . Coen PM, Goodpaster BH. Role of intramyocelluar lipids in human health. Trends Endocrinol Metab. 2012;23:391–8. Hamrick MW, McGee-Lawrence ME, Frechette DM. Fatty Infiltration of Skeletal Muscle: Mechanisms and Comparisons with Bone Marrow Adiposity. Front Endocrinol (Lausanne). 2016;7:1. https://doi.org/10.3389/fendo.2016.00069 . Addison O, Marcus RL, Lastayo PC, Ryan AS. (2014) Intermuscular fat: A review of the consequences and causes. Int. J. Endocrinol. 2014. Pezolato A, De Vasconcelos EE, Defino HLA, et al. Fat infiltration in the lumbar multifidus and erector spinae muscles in subjects with sway-back posture. Eur Spine J. 2012;21:2158–64. https://doi.org/10.1007/s00586-012-2286-z . Mandelli F, Nüesch C, Zhang Y, et al. Assessing Fatty Infiltration of Paraspinal Muscles in Patients With Lumbar Spinal Stenosis: Goutallier Classification and Quantitative MRI Measurements. Front Neurol. 2021;12:656487. https://doi.org/10.3389/FNEUR.2021.656487/BIBTEX . Antony J, McGuinness K, Welch N, et al. An interactive segmentation tool for quantifying fat in lumbar muscles using axial lumbar-spine MRI. IRBM. 2016;37:11–22. https://doi.org/10.1016/j.irbm.2015.10.004 . Engstrom CM, Fripp J, Jurcak V, et al. Segmentation of the quadratus lumborum muscle using statistical shape modeling. J Magn Reson Imaging. 2011;33:1422–9. https://doi.org/10.1002/jmri.22188 . Wesselink EO, Elliott JM, Coppieters MW, et al. Convolutional neural networks for the automatic segmentation of lumbar paraspinal muscles in people with low back pain. Sci Rep. 2022;12:13485. https://doi.org/10.1038/s41598-022-16710-5 . Burian E, Rohrmeier A, Schlaeger S, et al. Lumbar muscle and vertebral bodies segmentation of chemical shift encoding-based water-fat MRI: The reference database MyoSegmenTUM spine. BMC Musculoskelet Disord. 2019;20:1–7. https://doi.org/10.1186/s12891-019-2528-x . Li H, Luo H, Liu Y. Paraspinal muscle segmentation based on deep neural network. Sens (Switzerland). 2019;19. https://doi.org/10.3390/s19122650 . Weber KA, Abbott R, Bojilov V, et al. Multi-muscle deep learning segmentation to automate the quantification of muscle fat infiltration in cervical spine conditions. Sci Rep 2021. 2021;111 11:1–15. https://doi.org/10.1038/s41598-021-95972-x . Dixon WT. Simple proton spectroscopic imaging. Radiology. 1984;153:189–94. https://doi.org/10.1148/radiology.153.1.6089263 . Wren TAL, Bluml S, Tseng-Ong L, Gilsanz V. Three-Point Technique of Fat Quantification of Muscle Tissue as a Marker of Disease Progression in Duchenne Muscular Dystrophy: Preliminary Study. Am J Roentgenol. 2008;190:W8–12. https://doi.org/10.2214/AJR.07.2732 . Marcon M, Berger N, Manoliu A, et al. Normative values for volume and fat content of the hip abductor muscles and their dependence on side, age and gender in a healthy population. Skeletal Radiol. 2016;45:465–74. https://doi.org/10.1007/s00256-015-2325-z . Belzunce MA, Henckel J, Di Laura A, et al. Similarities and differences in skeletal muscle and body composition between sexes: an MRI study of recreational cyclists. BMJ Open Sport Exerc Med. 2023;9:e001672. https://doi.org/10.1136/BMJSEM-2023-001672 . Belzunce MA, Henckel J, Laura A, Di, et al. Mid-life cyclists preserve muscle mass and composition: a 3D MRI study. BMC Musculoskelet Disord. 2023;24:209. https://doi.org/10.1186/s12891-023-06283-3 . Paoletti M, Pichiecchio A, Piccinelli SC, et al. Advances in quantitative imaging of genetic and acquired myopathies: Clinical applications and perspectives. Front Neurol. 2019;10:78. Maggi L, Moscatelli M, Frangiamore R, et al. Quantitative Muscle MRI Protocol as Possible Biomarker in Becker Muscular Dystrophy. Clin Neuroradiol. 2020;1–10. https://doi.org/10.1007/s00062-019-00875-0 . Yin L, Xie Z, ying, Xu Hyan, et al. T2 Mapping and Fat Quantification of Thigh Muscles in Children with Duchenne Muscular Dystrophy. Curr Med Sci. 2019;39:138–45. https://doi.org/10.1007/s11596-019-2012-8 . Belzunce MA, Henckel J, Di Laura A, Hart A. Intramuscular fat in gluteus maximus for different levels of physical activity. Sci Rep 2021. 2021;111 11:1–10. https://doi.org/10.1038/s41598-021-00790-w . Belzunce MA, Henckel J, Di Laura A, et al. Similarities and differences in skeletal muscle and body composition between sexes: an MRI study of recreational cyclists. BMJ Open Sport Exerc Med. 2023;9:e001672. https://doi.org/10.1136/BMJSEM-2023-001672 . Weber KA, Abbott R, Bojilov V, et al. Multi-muscle deep learning segmentation to automate the quantification of muscle fat infiltration in cervical spine conditions. Sci Rep 2021. 2021;111 11:1–15. https://doi.org/10.1038/s41598-021-95972-x . Bas J, Ogier AC, Le Troter A et al. (2020) Fat fraction distribution in lower limb muscles of patients with CMT1A. Neurology 10.1212/WNL.0000000000009013. https://doi.org/10.1212/wnl.0000000000009013 Willis TA, Hollingsworth KG, Coombs A, et al. Quantitative Muscle MRI as an Assessment Tool for Monitoring Disease Progression in LGMD2I: A Multicentre Longitudinal Study. PLoS ONE. 2013;8:e70993. https://doi.org/10.1371/journal.pone.0070993 . Kemnitz J, Baumgartner CF, Eckstein F, et al. Clinical evaluation of fully automated thigh muscle and adipose tissue segmentation using a U-Net deep learning architecture in context of osteoarthritic knee pain. Magn Reson Mater Phys Biol Med. 2019;1–11. https://doi.org/10.1007/s10334-019-00816-5 . Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Springer; 2015. pp. 234–41. Jaudet C, Weyts K, Lechervy A, et al. The Impact of Artificial Intelligence CNN Based Denoising on FDG PET Radiomics. Front Oncol. 2021;11:3136. https://doi.org/10.3389/FONC.2021.692973/BIBTEX . Balerdi G, Henckel J, Di Laura A, et al. Automatic Segmentation Technique for Lumbar Spine Muscle Evaluation from MRI Images. IFMBE Proc. 2024;106:80–7. https://doi.org/10.1007/978-3-031-61960-1_8 . Hemke R, Buckless CG, Tsao A, et al. Deep learning for automated segmentation of pelvic muscles, fat, and bone from CT studies for body composition assessment. Skeletal Radiol. 2019. https://doi.org/10.1007/s00256-019-03289-8 . Sakamoto M, Hiasa Y, Otake Y et al. (2019) Automated segmentation of hip and thigh muscles in metal artifact contaminated CT using CNN. https://doi.org/101117/122521440 11050:124–9. https://doi.org/10.1117/12.2521440 Baur D, Bieck R, Berger J, et al. Analysis of the paraspinal muscle morphology of the lumbar spine using a convolutional neural network (CNN). Eur Spine J. 2022;31:774–82. https://doi.org/10.1007/S00586-021-07073-Y/FIGURES/4 . Fitzpatrick JA, Basty N, Cule M, et al. Large-scale analysis of iliopsoas muscle volumes in the UK Biobank. Sci Rep 2020. 2020;101 10:1–10. https://doi.org/10.1038/s41598-020-77351-0 . De Martino E, Hides J, Elliott JM, et al. Lumbar muscle atrophy and increased relative intramuscular lipid concentration are not mitigated by daily artificial gravity after 60-day headdown tilt bed rest. J Appl Physiol. 2021;131:356–68. https://doi.org/10.1152/JAPPLPHYSIOL.00990.2020/ASSET . /IMAGES/LARGE/JAPPLPHYSIOL.00990.2020_F005.JPEG . Gadermayr M, Li K, Müller M et al. (2019) Domain-Specific Data Augmentation for Segmenting MR Images of Fatty Infiltrated Human Thighs With Neural Networks. https://doi.org/10.1002/jmri.26544 Gadermayr M, Heckmann L, Li K, et al. Image-to-Image Translation for Simplified MRI Muscle Segmentation. Front Radiol. 2021;1:664444. https://doi.org/10.3389/FRADI.2021.664444/BIBTEX . Crawford RJ, Cornwall J, Abbott R, Elliott JM. Manually defining regions of interest when quantifying paravertebral muscles fatty infiltration from axial magnetic resonance imaging: a proposed method for the lumbar spine with anatomical cross-reference. BMC Musculoskelet Disord. 2017;18:25. https://doi.org/10.1186/S12891-016-1378-Z . Anstruther M, Rossini B, Zhang T, et al. PILLAR: ParaspInaL muscLe segmentAtion pRoject - a comprehensive online resource to guide manual segmentation of paraspinal muscles from magnetic resonance imaging. BMC Musculoskelet Disord. 2023;24:909–. https://doi.org/10.1186/S12891-023-07029-X/FIGURES/3 . Berry DB, Padwal J, Johnson S, et al. Methodological considerations in region of interest definitions for paraspinal muscles in axial MRIs of the lumbar spine. BMC Musculoskelet Disord. 2018;19:135–. https://doi.org/10.1186/S12891-018-2059-X/FIGURES/6 . Marstal K, Berendsen F, Staring M, Klein S. (2016) SimpleElastix: A User-Friendly, Multi-lingual Library for Medical Image Registration. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, pp 574–582. Belzunce MA, Henckel J, Fotiadou A, et al. Automated multi-atlas segmentation of gluteus maximus from Dixon and T1-weighted magnetic resonance images. Magn Reson Mater Phys Biol Med. 2020;33:677–88. Belzunce MA, Henckel J, Fotiadou A, et al. Automated measurement of fat infiltration in the hip abductors from Dixon magnetic resonance imaging. Magn Reson Imaging. 2020;72:61–70. https://doi.org/10.1016/j.mri.2020.06.019 . Aljabar P, Heckemann RA, Hammers A, et al. Multi-atlas based segmentation of brain images: Atlas selection and its effect on accuracy. NeuroImage. 2009;46:726–38. https://doi.org/10.1016/J.NEUROIMAGE.2009.02.018 . Rueckert D, Sonoda LI, Hayes C, et al. Nonrigid registration using free-form deformations: application to breast MR images. IEEE Trans Med Imaging. 1999;18:712–21. https://doi.org/10.1109/42.796284 . Klein S, Staring M, Murphy K, et al. elastix: A Toolbox for Intensity-Based Medical Image Registration. IEEE Trans Med Imaging. 2010;29:196–205. https://doi.org/10.1109/TMI.2009.2035616 . Vandervoort E, Sossi V. An analytical scatter correction for singles-mode transmission data in PET. IEEE Trans Med Imaging. 2008;27:402–12. https://doi.org/10.1109/TMI.2007.909851 . Kim H, Bin, Kim HS, Kim SJ, Yoo J, Il. Spine muscle auto segmentation techniques in MRI imaging: a systematic review. BMC Musculoskelet Disord. 2024;25:1–8. https://doi.org/10.1186/S12891-024-07777-4/FIGURES/3 . Lareau-Trudel E, Troter A, Le, Ghattas B et al. (2015) Muscle quantitative MR imaging and clustering analysis in patients with facioscapulohumeral muscular dystrophy type 1. PLoS One 10:. https://doi.org/10.1371/journal.pone.0132717 Sanz-Requena R, Martínez-Arnau FM, Pablos-Monzó A, et al. The role of imaging biomarkers in the assessment of sarcopenia. Diagnostics. 2020;10. https://doi.org/10.3390/diagnostics10080534 . Otto LAM, Pol W, Schlaffke L, et al. Quantitative MRI of skeletal muscle in a cross-sectional cohort of patients with spinal muscular atrophy types 2 and 3. NMR Biomed. 2020;33. https://doi.org/10.1002/nbm.4357 . Arrieta C, Urrutia J, Besa P, et al. Automatic quantification of fat infiltration in paraspinal muscles using T2-weighted images: An OsiriX application. Biomed Signal Process Control. 2020;57:101793. https://doi.org/10.1016/J.BSPC.2019.101793 . Wesselink EO, Pool-Goudzwaard A, De Leener B, et al. Investigating the associations between lumbar paraspinal muscle health and age, BMI, sex, physical activity, and back pain using an automated computer-vision model: a UK Biobank study. Spine J. 2024;24:1253–66. https://doi.org/10.1016/j.spinee.2024.02.013 . Berry DB, Rodriguez-Soto AE, Englund EK, et al. Multiparametric MRI characterization of level dependent differences in lumbar muscle size, quality, and microstructure. JOR Spine. 2020;3:e1079. https://doi.org/10.1002/JSP2.1079 . Vitale J, Sconfienza LM, Galbusera F. Cross-sectional area and fat infiltration of the lumbar spine muscles in patients with back disorders: a deep learning-based big data analysis. Eur Spine J 2023. 2023;331 33:1–10. https://doi.org/10.1007/S00586-023-07982-0 . Schlaeger S, Inhuber S, Rohrmeier A, et al. Association of paraspinal muscle water–fat MRI-based measurements with isometric strength measurements. Eur Radiol 2018. 2018;292 29:599–608. https://doi.org/10.1007/S00330-018-5631-8 . Additional Declarations No competing interests reported. Supplementary Files SupplementaryDataSegmentationExamples.zip SupplementaryMaterial.docx Cite Share Download PDF Status: Published Journal Publication published 31 Mar, 2026 Read the published version in BMC Medical Imaging → Version 1 posted Editorial decision: Revision requested 20 Mar, 2026 Editor assigned by journal 20 Mar, 2026 Reviews received at journal 15 Feb, 2026 Reviewers agreed at journal 09 Feb, 2026 Reviewers invited by journal 29 Jan, 2026 Submission checks completed at journal 29 Jan, 2026 First submitted to journal 29 Jan, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information

Related documents

• Chemical Shift Encoding-based Water-fat Magnetic Resonance Imaging in Spinal Tumors#875405
• Toward an Automated Method of Abdominal Fat Segmentation of MR Images#1029843
• Longitudinal Quantitative Neuromuscular MRI in Neuropathic Patients#880570
• MRIHeadScan: An Automated Pipeline for Quantifying Structural Brain Asymmetry and Statistical Rarity#336395
• Validation of AI Lumbar Spine Anatomy Measureing Function#887365