All participant data were acquired with a 3T MRI scanner (Magnetom Prisma, Siemens, Germany) in a supine position. An 18-channel body matrix coil was placed on the participants’ legs, as well as an integrated 32-channel spine coil for image acquisition. T1-weighted and m-Dixon scans were acquired from the participants’ dominant leg. The T1-weighted scan covered the entire length of the shank, extending from the distal end of the femur to the ankle, covering 260 transverse slices. The scan protocol used a 2D turbo spin echo (TSE) with a field of view (FOV) of 262 × 350 mm and an acquisition matrix of 336 × 448, reconstructed to 672 × 896. Repetition time/echo time (TR/TE) was 11.7/5.29 ms with a flip angle of 10°, one signal average (NSA), and a total scan time of 462 s, split into two sequences of 231 s each. The mDixon scans covered 192 transverse slices and focused on the muscle bellies of LG, MG, and SOL. A 2-point 3D multi-echo mDixon fast field echo (FFE) sequence, with a FOV of 284 × 350 mm, an acquisition matrix of 260 × 320, and a slice thickness of 2 mm, without a gap was acquired. The scan parameters were TR/TE1/TE2 = 4.33/1.35/2.58 ms, flip angle = 9°, one NSA, and a total scan time of 97 s (D'Souza et al., 2020; Bolsterlee et al., 2019).

An experienced examiner used a combination of semi-automated (Sashimi Segmentation V1.1 (D'Souza et al., 2020)) and manual segmentation (ITK-SNAP v3.8.0, NIH, United States) to delineate the boundary of the MG, LG, and SOL muscles on all slices of the T1-weighted image, when the respective muscle was visible on the scan. The segmentations from Sashimi were imported into ITK-SNAP for inspection, as the automated tool occasionally misclassified boundaries in regions of low contrast or high intramuscular FF. Manual adjustments were made as needed to ensure anatomical accuracy. Following this, a 3D reconstruction was performed in ITK-SNAP to visually assess the combined segmentations of the three muscles. In regions where muscle boundaries were directly adjacent, fat voxels located at the interface were assigned to the muscle whose segmentation label enclosed the voxel. As segmentation masks were manually corrected and non-overlapping, each fat voxel was uniquely assigned to a single muscle region.

All statistical analyses were performed using Python’s SciPy and scikit-posthocs, with visualization via Matplotlib and Seaborn. Pearson’s correlation coefficient assessed the relationship between mean FF and FF pixel percentage along the muscle length. Normality and homogeneity of variances of the Delaunay mean edge lengths and volumes of each muscle were tested using the Shapiro-Wilk and Levene’s tests. If assumptions were met, a two-way ANOVA (p < 0.05) assessed age and sex effects, followed by Tukey’s HSD for post hoc comparisons. If assumptions were violated, the Friedman test was used, followed by the Kruskal–Wallis test and Dunn’s post hoc test with Bonferroni correction. Considering the small sample size, effect sizes (Hedges’ g) were assessed to directly quantify the magnitude of differences, allowing interpretation regardless of statistical significance or sample size limitations.

As displayed in Figures 1A–C, the overall intramuscular FF pixel percentages were greater in the older compared to the young adults. In MG, the older male group had the greatest intramuscular FF pixel percentage, with values increasing from distal to proximal regions, and stabilizing as they approached the proximal region. In LG, the largest values were seen in the older female group (Figure 1B), and in general there was a steeper increase in intramuscular FF pixel percentage from distal to proximal regions than seen in MG. In the SOL muscle (Figure 1C), the largest values were observed in the older male group with smaller fluctuations in values across the muscle length observed in each group compared to MG and LG. In MG and LG, the greatest values occurred in the proximal muscle region in most groups except for the older female group (Table 1). In the SOL muscle, the greatest values occurred in the distal region but shifted to the proximal region for older groups.

The mean intramuscular FF distributions were generally similar to the intramuscular FF pixel percentage in the muscles studied, except for the proximal region of MG muscle for older groups and the distal region of SOL muscle for the young female group (Figures 1D–F; Table 1). In the young male group, there was moderate to strong correlations in all three muscles, indicating a consistent relationship between the two parameters (Table 2). Similarly, the young female group also exhibited strong correlations in the MG and LG muscles, but a poor correlation in the SOL muscle, indicating that mean intramuscular FF within specific regions of the muscle does not align consistently with the extent of spatial coverage represented by the intramuscular FF pixel percentage. In older males and females, mean intramuscular FF strongly correlated with intramuscular FF pixel percentage in LG and SOL, but only moderately in MG.

The contour bandwidths, visible in Figure 1, are detailed in Table 3. Larger bandwidths indicate greater spatial variability along the muscle. In general, larger values were observed for intramuscular FF pixel percentage compared to mean intramuscular FF in all three muscles in each group. For most of the comparisons between young vs. older age groups, older participants showed the largest bandwidth. The only exception was in the SOL for young males, with a slightly lower value (0.43 vs. 0.36). Interestingly, while the mean intramuscular FF bandwidth values were moderate in size in older participants, the difference in intramuscular FF pixel percentage bandwidth values between young and older groups were larger. In terms of the inter-muscle comparisons, the LG muscle displayed the largest bandwidth, indicating the largest variability of the two parameters along the muscle length.

In general, 3D intramuscular FF distribution was similar in each of the muscles for participants within the same age and sex group (see Supplementary File S2). There were however distinct differences in the patterns observed between the three muscles and the participant groups. In young participants, the MG and LG showed a similar pattern with branch-like structures of intramuscular FF voxels through the muscle volume; compared to the LG muscle, MG showed more branches (Figures 2A,B). For the SOL of the young participants, a broader distribution of intramuscular FF voxels with several distinct high-density clusters were observed (Figure 2C). These clusters were dispersed throughout the muscle volume, indicating multiple focal points of IMF accumulation rather than a single branch-like structure. In the older participant, the intramuscular FF voxels appeared more densely clustered (Figures 2D–F), especially for the MG and SOL muscles (Figures 2D,F). More complex and greater branch-like structures were observed in these two muscles, with more densely packed intramuscular FF voxels appearing to be more widely distributed through the muscle volume.

Significant main effects of age were observed for both the Delaunay mean edge length and volume within each muscle, with young participants having longer edge lengths than older participants (Figure 3; Table 4), suggesting sparser intramuscular FF distribution with larger gaps between clusters. Specifically, age-only paired comparisons for the mean edge length showed significantly longer values for young vs. older males (MG: mean difference = 0.64 mm, p < 0.001, g = 1.953; LG: p < 0.001, g = 1.615) but similar lengths between young and older females (MG: mean difference = 0.33 mm, p = 0.2, g = 1.475; LG: p = 0.75, g = 1.416). Only the SOL muscle showed significantly larger values for both males (mean difference = 0.69 mm, p = 0.002, g = 1.696) and females (mean difference = 0.81 mm, p = 0.005, g = 2.181). The mean volume paired comparisons revealed the same differences as the mean edge length in males (MG: mean difference = 1.40 mm³, p < 0.001,g 2.056; LG: p = 0.002, g = 1.396; SOL: p = 0.006, g = 1.443) and females (MG: mean difference = 0.80 mm³, p = 0.13, g = 1.432; LG: p = 0.42, g = 1.597; SOL: p = 0.04, g = 1.721).

Paired age-sex group comparisons showed some significant differences for mean edge length (MG: (young female vs. older male with mean difference = 0.53 mm, p = 0.006, g = 2.451; young male vs. older female with mean difference = 0.44 mm, p = 0.03, g = 1.267); LG: young female vs. older male with p = 0.01, g = 2.737; SOL: young female vs. older male with mean difference = 0.97 mm, p < 0.001, g = 3.000) and mean volumes (MG: (young female vs. older male with mean difference = 1.09 mm³, p = 0.008, g = 2.239; young male vs. older female with mean difference = 1.11 mm³, p = 0.009, g = 1.479); SOL: young female vs. older male with p = 0.004, g = 2.291). However, it is important to note that age was the dominant factor for these differences, as the effects of sex and age × sex interaction were not significant in the muscles studied (Table 4). Beyond p-values, effect size analysis using Hedges’ g further demonstrated strong age-related differences in both mean edge length and volume. Across all muscles, Hedges’ g values for age effects in males ranged from 1.40 to 2.06, and in females from 1.43 to 2.18, indicating large effects of age on intramuscular fat distribution patterns. In contrast, sex effects were small to negligible within age groups (Hedges’ g < 0.4 in all cases), and interaction effects remained nonsignificant.

Comparisons between the CSR expectation line and the K values were shown in Figure 4. The largest inter-participant variations occurred in young males. Among all group-muscle combinations, the SOL muscle in the young males exhibited the largest maximum clustering distance, up to 17.5 a.u. This suggests clustering occurs up to this distance, beyond which intramuscular FF clusters become randomly distributed. The inter-muscle comparison showed SOL > MG > LG in clustering strength, a consistent order across participant groups. The L(r) function in Figure 5 provided a transformed view of Ripley’s K function that centers the expectation of CSR at zero. Positive values of L(r) across most distances indicated clustering of intramuscular FF beyond CSR. SOL consistently showed the highest L(r) values, particularly in older adults, indicating stronger and more extended spatial clustering. The older male participants (solid lines) generally exhibited the highest peaks in L(r), indicating dense local intramuscular FF clusters. In contrast, young participants (dotted lines) had lower L(r) values, suggesting more uniform cluster distributions. Notably, in most cases, L(r) values decreased beyond a distance of 75 a.u., approaching or dipping below zero, indicating a transition from clustering to random or even regular distributions beyond that scale.

The region-growing method incorporates both local intramuscular FF values and their spatial relationships to classify intramuscular FF regions (Rosea et al., 2024). By using spatial connectivity, this method ensures that intramuscular FF voxels are classified in a manner consistent with their neighbouring tissue, providing spatial coherence and reducing the influence of noise or isolated artifacts. Spatially connected voxels that form contiguous clusters are more likely to represent true IMF deposits, as they exhibit similar tissue properties—such as uniform fat content—rather than being isolated voxels that may result from partial volume effects or mixtures of muscle, tendon, and fat (Soret et al., 2007). Additionally, maintaining spatial coherence is a critical advantage over simple thresholding approaches (Maddocks et al., 2014; Young et al., 2015). The method has previously been used to identify intramuscular FF regions (Wong et al., 2020) but was limited to just a few slices rather than the whole muscle volume.

As the region growing algorithm was applied to images from healthy participants, the voxels considered predominantly muscle were identified and removed to leave only FF clusters. This approach may require refinements if applied to clinical data, where the FF values are likely to be much higher and tracking change over time would be valuable. The current method used a data-driven approach to dynamically threshold, resulting in slightly larger values in older participants (See methods). This could have led to smaller IMF masks. However, the greater intramuscular FF pixel percentage values (Figure 1) and more densely clustered intramuscular FF (Figure 2) seen in older adults suggests the differences are unlikely to be an artifact of the threshold values. It should however also be noted that the use of slice-specific F F median as the threshold could lead to inconsistencies in voxel classification across slices. However, F F median reflects the local muscle baseline that can adapt to both anatomical variations and participant-specific differences. A sensitivity analysis with ± 5 variations of F F median threshold was conducted (see Supplementary File S1). Only moderate increases in the total number of intramuscular FF voxels were found with reduced threshold but the spatial distribution remained unchanged (see Supplementary File S1), supporting the robustness of the slice-dependent thresholding method.

In the MG and LG, and to a smaller extent the SOL, there was an increase in intramuscular FF pixel percentage along the distal-proximal muscle length in the older group, with the distance to values observed in the younger groups also increasing from distal-to-proximal regions (Figure 1). This indicates regional variation in accumulation of IMF may occur because of aging (Pinel et al., 2021). What drives this variation is however unclear. Possible explanations include differences in fibre types along the muscle length. Previous work has suggested that faster fibre types may be preferentially expressed in proximal regions of MG (Hodson-Tole et al., 2013) and it is posited that muscle degeneration, particularly in Type II muscle fibers, leads to increased fat storage (Goodpaster and Wolf, 2004). Additionally, variation in intramuscular pressure distributions during contraction could also influence IMF accumulation, given that adipocytes have been shown to be mechanosensitive (Blade et al., 2024).

We found only two previous reports of variation in mean intramuscular FF levels along the muscle length, both in patient populations. While one study observed a distal-to-proximal decrease (Riem et al., 2024), the other larger study, found more heterogeneous patterns (Heskamp et al., 2022). In contrast, our findings in healthy participants show the highest mean intramuscular FF in proximal muscle regions (Figure 1; Table 1), suggesting different mechanisms may drive IMF infiltration in disease-related versus age-related changes. (Addison et al., 2014b). These differences indicate potential for intramuscular FF quantification to distinguish between age- and disease-related changes in muscle characteristics. Therefore, spatial variations in intramuscular FF values could be further exploited to provide more robust and sensitive markers of skeletal muscle health. Such markers would be useful not only as outcome measures in clinical trials but would also have value in diagnosis and monitoring progressive and degenerative diseases to support more personalized decision-making in clinical care.

We observed branch-like structures for intramuscular FF clusters, however, without direct vascular imaging such as magnetic resonance angiography, this interpretation remains speculative. There was a noticeable difference in the 3D intramuscular FF distribution between the three muscles studied (Figure 2, Supplementary File S2). In MG and LG, clear and large branching structure were observed, while SOL showed much more diffuse intramuscular FF clusters. The differences between muscles may be related to differences in muscle fibre architecture (more variables in SOL muscles with multiple architectural compartments compared to MG and LG muscles) (Edgerton et al., 1975) and/or by differences in fibre type composition: the SOL is predominantly composed of slow-twitch fibers (Gollnick et al., 1974). These fibers have a higher capillary density, allowing them to effectively use oxygen and fatty acids without requiring additional vascular adaptations to support fat storage and metabolism (Saltin and Gollnick, 1983). As a result, FF may accumulate around the capillaries rather than the larger, primary vascular branches. In contrast, MG and LG contain more of a mix of slow- and fast-twitch fibers (Edgerton et al., 1975), and tend to accumulate more extramyocellular lipids (EMCL) especially with aging (Schrauwen-Hinderling et al., 2006). However, a limitation of FF quantification using the mDixon method is that it does not distinguish between intramyocellular lipids (IMCL) and EMCL. EMCL has been shown to exhibit orientation-dependent chemical shift dispersion due to the anisotropic magnetic susceptibility of the surrounding muscle tissue (Khuu et al., 2009). This orientation effect can influence the accuracy of the mDixon method for characterization of intramuscular FF.