ABSTRACT
Background
Differentiating myogenic from neurogenic neuromuscular diseases (NMDs) can be clinically challenging. While quantitative muscle MRI (qMRI) with proton density fat fraction (PDFF,%) quantifies fat replacement, it misses micro‐spatial patterns linked to underlying pathology. This study investigates whether qMRI with 3D radiomic texture analysis (TA) might improve differentiation between myogenic and neurogenic diseases, using myotonic dystrophy type 1 (DM1) and Charcot–Marie‐Tooth neuropathy type 1A (CMT1A) as proof‐of‐concept models.
Methods
Thirty‐three adults with DM1, 33 with CMT1A, and 33 matched healthy controls were included. qMRI on a 3T Philips Achieva system using a 6‐point Dixon sequence generated PDFF(%) maps of the lower limbs, and a convolutional neural network performed 3D segmentation of 28 lower limb muscles. We extracted macroscopic features, including muscle volume, asymmetry, and disto‐proximal gradients, alongside micro‐spatial radiomic features (entropy, contrast, homogeneity) to quantify tissue heterogeneity.
Results
Both patient cohorts exhibited higher PDFF(%) in all lower limb muscles compared to controls (p < 0.001). DM1 predominantly involved the posterior compartment, while CMT1A targeted the anterolateral compartment with significantly steeper disto‐proximal fat gradients (p < 0.05). TA revealed higher entropy and contrast, and lower homogeneity in CMT1A compared to DM1, reflecting a more reticular pattern of fat infiltration vs. the confluent pattern in DM1.
Conclusion
In this proof‐of‐concept study, 3D radiomic texture analysis of PDFF(%) maps revealed distinct spatial patterns of fat replacement in DM1 and CMT1A. Integrating radiomic and conventional qMRI features may enhance the non‐invasive distinction between DM1 and CMT1A and warrants further investigation in a broader range of NMDs.
Keywords: 3D muscle segmentation, Charcot–Marie‐tooth type 1A, myotonic dystrophy type 1, Radiomics, texture analysis
In this proof‐of‐concept study, 3D radiomic texture analysis of quantitative muscle MRI (proton density fat fraction (PDFF, %) maps) distinguished the myogenic disease DM1 from the neurogenic disease CMT1A. Compared with DM1, CMT1A showed higher entropy, contrast, and lower homogeneity, reflecting a reticular vs. confluent pattern of fat infiltration. Integrating these micro‐spatial radiomic features with macroscopic qMRI measures, including atrophy, asymmetry, and disto‐proximal gradients, may enhance the non‐invasive distinction between these myogenic and neurogenic diseases.
1. Introduction
Myopathies and hereditary peripheral neuropathies are heterogeneous genetic disorders with distinct pathomechanisms but overlapping clinical phenotypes characterized by progressive muscle weakness and atrophy [1, 2]. Electromyography (EMG) and serum creatine kinase (CK) levels are commonly used diagnostically but are not always conclusive, making differentiation challenging [3, 4]. Muscle biopsy can provide additional information but is invasive and not consistently discriminative. Definitive diagnosis often relies on molecular genetic testing, although the diagnostic rate is lower in certain hereditary peripheral neuropathies, such as distal hereditary motor neuropathies (dHMN) [5, 6]. Consequently, patients can experience prolonged diagnostic trajectories, delaying appropriate disease monitoring and access to emerging, disease‐specific therapies [4].
Muscle Magnetic Resonance Imaging (MRI) is a valuable non‐invasive tool in the diagnostic workup of neuromuscular disorders (NMDs), enabling characterization of disease‐specific patterns of muscle involvement [7, 8, 9]. Previous studies have described distinct micro‐spatial patterns of fat replacement in myogenic and neurogenic diseases. A reticular pattern of fat replacement and the presence of “muscle islands” was suggested to be characteristic of neurogenic diseases, whereas more homogeneous, moth‐eaten patterns were features of myogenic conditions [7, 8, 10]. However, the evaluation of these patterns relied on qualitative gradings, using observer‐dependent visual descriptions, and semi‐quantitative scoring systems such as the Mercuri scale [11].
While quantitative MRI (qMRI) with proton density fat fraction (PDFF, %) has been shown to be more sensitive than qualitative or semi‐quantitative scoring [12, 13], current qMRI analyses focus on macroscopic measures such as mean PDFF (%), muscle volume, and asymmetry [14, 15]. These measures already provide substantial information but may not fully capture the micro‐spatial pattern of fat replacement. Radiomic texture analysis addresses this limitation by extracting quantitative features that characterize micro‐spatial intensity distributions [16]. By quantifying the heterogeneity of tissue characteristics, it offers significantly higher sensitivity than standard visual assessments [17, 18]. Radiomic texture analysis has shown promise in skeletal muscle imaging for differentiating NMD subtypes [19, 20] and characterizing patterns of fat infiltration [15, 18, 21].
To move beyond qualitative descriptions of fat replacement patterns, we applied 3D radiomic texture analysis of PDFF (%) maps to determine whether micro‐spatial fat replacement patterns differentiate myogenic from neurogenic diseases. As comparative disease models, we selected Myotonic Dystrophy type 1 (DM1) and Charcot–Marie‐Tooth neuropathy type 1A (CMT1A), among the most prevalent myogenic and neurogenic conditions with predominantly distal involvement [22, 23], to assess both macroscopic and micro‐spatial muscle MRI features of the lower limbs.
2. Methods
2.1. Participants and Study Design
We included patients with genetically confirmed adult‐onset DM1 or CMT1A, and matched healthy controls. Patient eligibility criteria were (i) age ≥ 18 years; (ii) presence of distal lower limb muscle weakness; and (iii) ambulatory status; use of assistive devices was permitted. Individuals with contraindications to MRI were excluded from participation. This study was approved by the Ethics Committee Research UZ/KU Leuven (s67475;s67476) and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent. Age, sex, body mass index (BMI), and disease duration were recorded at the time of the MRI scan. Disease duration was defined as the interval between self‐reported symptom onset and study visit.
2.2. MRI Acquisition and Analysis
MRI data were acquired using a 3 T scanner (Achieva, Philips Medical Systems, The Netherlands). Participants were positioned supine, feet‐first. Axial T1‐weighted images of the lower limbs were acquired for anatomical reference (four stacks, TR 546 ms, TE 20 ms, 30 slices, 6 mm slice thickness, 1 mm interslice gap, FOV 455 × 301 mm2, voxel size 0.8 × 0.8 mm2). Muscle PDFF (%) was quantified using a 3D 6‐point Dixon sequence (TR/TE/ΔTE = 10/1.5/1.0 ms, flip angle 3°, 130 slices, slice thickness 2 mm, FOV 450 × 394 × 260 mm3, matrix 348 × 304 × 130, voxel size 1.2 × 1.2 × 2 mm3). Five image stacks encompassed proximal (anterior superior iliac spine to tibial plateau) and distal (patella to lateral malleolus) lower limbs. Dixon images were processed using an in‐house AI tool based on a convolutional neural network (CNN) [24]. Fat, water, in‐phase, and out‐of‐phase images were preprocessed to stitched PDFF (%) maps of the proximal and distal lower limbs. The CNN subsequently predicted 3D whole‐muscle segmentations of 28 individual lower limb muscles (18 proximal and 10 distal) [24]. Segmentations were visually inspected and corrected if necessary using ITK‐Snap. Mean PDFF (%) of each muscle and volume‐weighted mean for each muscle group were extracted using a MATLAB script (MathWorks, Natick, USA). In the distal lower limb, muscle groups were defined as anterolateral compartment (tibialis anterior, extensor digitorum longus, and peroneus) and posterior compartment (medial and lateral gastrocnemius, soleus, tibialis posterior, flexor digitorum longus, flexor hallucis longus, and popliteus). In the proximal lower limb, muscles were grouped into quadriceps (rectus femoris, vastus lateralis, vastus intermedius, and vastus medialis), adductor (adductor longus, adductor brevis, and adductor magnus), and hamstring (biceps femoris long head and short head, semitendinosus, and semimembranosus).
2.2.1. MRI Macroscopic Features
Total muscle volume (cm3) was calculated by multiplying the segmented voxels by the voxel volume. Muscle volumes were normalized to the participants' height squared (cm3/m2) and standardized as Z‐scores relative to the matched healthy controls. Muscles with Z‐scores ≤ −2.0 or ≥ 2.0 were considered atrophic or hypertrophic, respectively [25]. To evaluate asymmetry in both legs in fat replacement, the mean PDFF (%) was extracted for the right and left leg independently. An absolute left–right PDFF difference ≥ 10% was considered asymmetric [13].
To evaluate the disto‐proximal fat distribution, axial slice coordinates were normalized to a relative scale of 0–100% (0% = proximal, 100% = distal) to account for inter‐individual differences in muscle length. Mean PDFF (%) was extracted for each axial slice along the anatomical z‐axis. A Generalized Additive Model (GAM) was used to generate a smoothed representation of the gradient [12]. The disto‐proximal gradient was then defined as the rate of change in PDFF (%) relative to the normalized length, calculated via linear regression.
2.2.2. MRI Micro‐Spatial Features
To evaluate the micro‐spatial texture in muscle fat replacement, 3D radiomic texture analysis features were extracted from the PDFF (%) maps using the PyRadiomics library [26]. Prior to extraction, MRI data underwent B‐spline interpolation to isotropic voxels (1.2 × 1.2 × 1.2 mm3) consistent with the Image Biomarker Standardization Initiative guidelines [27, 28]. To ensure biological relevance, gray‐level discretization was performed using a fixed bin width of 1.0% PDFF (%). This resulted in 100 discrete intensity bins, capturing the full physiological range of fat infiltration (0–100%). Three Gray‐Level Co‐occurrence Matrix (GLCM) features (contrast, joint entropy, and homogeneity) were selected a priori for their correspondence to qualitative patterns previously described in NMDs. Specifically, contrast measures local intensity variation. High contrast identifies sharp transitions between preserved muscle fibers and localized fat deposits, serving as a marker for the spatial granularity [14, 15, 21]. Joint Entropy is a measure of the randomness (disorder). High entropy characterizes the micro‐spatial complexity and chaotic mixture of fat and muscle fibers [15, 18, 21]. Finally, homogeneity measures the local uniformity. Higher values represent a smooth, contiguous tissue texture where neighboring voxels have similar PDFF (%) levels [14, 15, 21]. All radiomic texture features were extracted from these quantitative PDFF (%) maps. The T1‐weighted images were used only for qualitative illustration.
2.3. Statistical Analysis
Statistical analyses were performed using R software (version 4.5.2). Continuous variables were analyzed using the Kruskal–Wallis test, with pairwise post hoc comparisons performed via Wilcoxon rank‐sum tests. Categorical variables were summarized as frequencies and percentages and compared using Fisher's exact test. To evaluate differences in tissue texture, radiomic features were analyzed using Analysis of Covariance (ANCOVA) models. Disease group was included as a fixed effect, and the mean PDFF (%) of the corresponding muscle was included as a covariate. Radiomic features were log‐transformed prior to modeling to satisfy assumptions. Results are reported as estimated marginal means ± standard error (SE). Associations between CTG repeat length and disease duration with mean PDFF (%), entropy, homogeneity, and contrast were assessed with Spearman rank correlations. All p‐values were adjusted using the Benjamini–Hochberg (BH) procedure to control the False Discovery Rate (FDR). Statistical significance was defined as an adjusted p < 0.05.
3. Results
3.1. Participant Demographics and Clinical Characteristics
A total of 99 participants were included, 33 patients with adult‐onset DM1, 33 adults with CMT1A, and 33 matched healthy controls. The demographics of each group are presented in Table 1. Age and sex did not differ significantly between groups. BMI was significantly different between patients and healthy controls, but not between the DM1 and CMT1A groups (p = 0.4). Disease duration did not differ significantly between the patient groups (p = 0.1).
TABLE 1.
Participant demographics, clinical and MRI characteristics.
| Patients with DM1 (n = 33) | Patients with CMT1A (n = 33) | Healthy controls (n = 33) | |
|---|---|---|---|
| Female (n) | 20 (61%) | 21 (64%) | 20 (61%) |
| Age (y) | 41.8 (±10.5) | 43.8 (±12.6) | 41.9 (±10.8) |
| Age at symptom onset (y) | 30 [18–40] | 18 [2–45] | — |
| Disease duration (y) | 12 [1–37] | 17 [2–49] | — |
| CTG repeats (n) | 365 [100–1000] | — | — |
| BMI (kg/m2) | 26.6 [16.5–34.2] | 26.3 [18.3–34.0] | 22.3 [18.4–29.4] |
| Ambulation (n) | |||
| Independent | 27 (82%) | 26 (79%) | 33 (100%) |
| Orthoses | 3 (9%) | 4 (12%) | — |
| Unilateral support | 0 (0%) | 2 (6%) | — |
| Bilateral support | 2 (6%) | 1 (3%) | — |
| Intermittent wheelchair use | 1 (3%) | 0 (0%) | — |
| Muscle PDFF (%) | |||
| Anterolateral compartment | 25.3 (±15.8) | 26.3 (±13.6) | 8.1 (±1.1) |
| Posterior compartment | 37.2 (±22.4) | 17.7 (±13.9) | 6.0 (±1.2) |
| Total distal lower limb | 33.7 (±20.3) | 19.5 (±13.4) | 6.5 (±1.1) |
| Quadriceps | 16.7 (±15.7) | 9.2 (±1.7) | 6.8 (±1.4) |
| Adductors | 10.4 (±11.4) | 8.4 (±2.9) | 5.3 (±1.0) |
| Hamstrings | 14.2 (±10.4) | 12.8 (±3.2) | 7.3 (±1.4) |
| Total proximal lower limb | 14.3 (±9.9) | 11.7 (±2.6) | 7.7 (±1.5) |
| Asymmetric PDFF (%) involvement (n) | |||
| Distal lower limb muscles | 42 (13%) | 37 (11%) | (0%) |
| Proximal lower limb muscles | 42 (7%) | 12 (2%) | (0%) |
| Muscle atrophy (n) | |||
| Distal lower limb muscles | 59 (18%) | 62 (19%) | (0%) |
| Proximal lower limb muscles | 13 (2%) | 18 (3%) | (0%) |
| Muscle hypertrophy (n) | |||
| Distal lower limb muscles | 19 (6%) | 21 (6%) | (0%) |
| Proximal lower limb muscles | 13 (2%) | 23 (4%) | (0%) |
Note: Data are shown as n (%), mean (± SD) or median [min–max].
Abbreviations: BMI, Body Mass Index; CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; CTG, cytosine‐thymine‐guanine; DM1, Myotonic dystrophy type 1; kg, kilograms; m, meter; n, number; PDFF, Proton density fat fraction; SD, Standard deviation; y, years.
3.2. Fat Replacement, Asymmetry and Muscle Volume
Although all muscles showed statistically higher PDFF (%) than controls (Table 1), the mean PDFF (%) of individual muscles is provided in Table S1. Within the patient populations, the DM1 group exhibited significantly higher PDFF (%) in the soleus (46.3% vs. 15.3%, p < 0.001), gastrocnemius medialis (52.9% vs. 26.2%, p < 0.001), gastrocnemius lateralis (31.0% vs. 15.0%, p = 0.01), and flexor hallucis longus (39.5% vs. 20.1%, p < 0.001) compared to the CMT1A group. In the proximal lower limb muscles, the PDFF (%) of the vastus lateralis (21.4% vs. 9.0%, p = 0.01) and vastus intermedius (19.0% vs. 8.0%, p = 0.02) of the DM1 cohort was significantly higher compared to the CMT1A cohort (Table S1).
The distal lower limb muscles showed a high disease burden in both conditions. In DM1, 59 muscles (18%, from 23 patients) were atrophic, 19 (6%, from 11 patients) hypertrophic, and 42 (13%, from 20 patients) asymmetric. Similarly, in CMT1A, 62 muscles (19%, from 20 patients) were atrophic, 21 (6%, from 11 patients) hypertrophic, and 37 (11%, from 19 patients) asymmetric (Table 1). Compared with controls, 5 of 10 distal lower limb muscles showed significant atrophy in both DM1 and CMT1A (Figure 1). In CMT1A, the highest prevalence of atrophy was observed in the tibialis anterior (36%), tibialis posterior (27%), and extensor digitorum longus (24%). In contrast, DM1 showed the greatest atrophy in the gastrocnemius medialis (49%), flexor hallucis longus (46%), and soleus (33%) (Figure 1). Direct comparisons between patient groups revealed disease‐specific patterns in which the gastrocnemius medialis was more frequently atrophic in DM1 than in CMT1A (49% vs. 9%, p < 0.01) and the tibialis posterior was more frequently affected in CMT1A (27% vs. 0%, p < 0.01). Regarding asymmetry, CMT1A patients exhibited significantly higher rates in the gastrocnemius medialis (30% vs. 0%, p = 0.008) and peroneus longus (24% vs. 0%, p = 0.02) compared with controls (Figure 2). In contrast, the DM1 group showed no significant differences compared with either CMT1A or controls (Figure 2). In the proximal lower limb muscles, atrophy, hypertrophy, and asymmetry were infrequent in both patient groups. In DM1, 13 muscles (2%, from 8 patients) were atrophic, 13 (2%, from 6 patients) hypertrophic, and 42 (7%, from 15 patients) asymmetric (Table 1). In CMT1A, 18 muscles (3%, from 5 patients) were atrophic, 23 (4%, from 10 patients) hypertrophic, and 12 (2%, from 8 patients) asymmetric. No significant differences were observed between groups or relative to controls for these measures, except for the tensor fasciae latae, which showed significantly greater hypertrophy in CMT1A compared with DM1 (27% vs. 0%, p = 0.03) (Figures 1 and 2).
FIGURE 1.
Muscle volumetric Z‐score: Quantifying atrophy and hypertrophy. Standardized muscle volumes (Z‐scores) of patients with DM1 and CMT1A are shown relative to matched healthy controls. Each data point represents an individual patient's deviation from the control mean. The horizontal line at Z = 0 represents the healthy cohort, and the dashed lines at Z = ±2 indicate thresholds for significant clinical deviation. Values below −2 indicate significant muscle atrophy, while values above 2 indicate hypertrophy. Central markers and error bars represent the group means and 95% confidence intervals. Asterisks show significant FDR‐adjusted p‐values (p < 0.05). CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; SD, Standard deviation. Muscle Abbreviations: AB, Adductor brevis; AL, Adductor longus; AM, Adductor magnus; BFB, Biceps femoris brevis; BFL, Biceps femoris longus; EDL, Extensor digitorum longus; FDL, Flexor digitorum longus; FHL, Flexor hallucis longus; GL, Gastrocnemius lateralis; GM, Gastrocnemius medialis; Gmax, Gluteus maximus; Gmed, Gluteus medius; Gmin, Gluteus minimus; GRA, Gracilis; PEC, Pectineus; PER, Peroneus; POP, Popliteus; RF, Rectus femoris; SAR, Sartorius; SM, Semimembranosus; SOL, Soleus; ST, Semitendinosus; TA, Tibialis anterior; TFL, Tensor fasciae latae; TP, Tibialis posterior; VI, Vastus intermedius; VL, Vastus lateralis; VM, Vastus medialis.
FIGURE 2.
Heatmap of prevalence of asymmetry. Heatmaps displaying the percentage of patients in the DM1 and CMT1A groups exhibiting significant asymmetry (10% difference in PDFF) in muscle fat replacement. Each cell and corresponding number represents the prevalence (%) within the specific cohort. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; SD, Standard deviation; For muscle abbreviations, see legend Figure 1.
3.3. Disto‐Proximal Profiles of Muscle Fat Replacement
In distal lower limb muscles, CMT1A exhibited significantly steeper disto‐proximal gradients compared to healthy controls in 9 of 10 distal lower limb muscles (Table 2). In DM1, distal gradients were only significantly increased compared with controls in 4 of 10 distal lower limb muscles (Table 2). Comparisons between patient groups demonstrated significantly higher distal gradients in 4 of the 10 distal lower limb muscles in CMT1A compared to DM1, mainly in the anterolateral compartment (TA: 14.9% vs. 8.6%, p = 0.04; PER: 35.4% vs. 6.6%, p < 0.001; EDL: 20.9% vs. 3.2%, p < 0.001; FDL: 13.5% vs. 7.8%, p < 0.001). In the proximal lower limb muscles, CMT1A showed significantly higher disto‐proximal gradients than controls in 8 of 18 muscles (Table 2). Conversely, some proximal lower limb muscles in CMT1A showed significantly lower gradients compared with controls, indicating relatively greater proximal than distal fat replacement in 3 of the 18 muscles (Table 2). In DM1, significantly higher proximal gradients compared with controls were observed in 10 of 18 muscles (Table 2). Between‐group comparisons further showed that DM1 exhibited significantly higher disto‐proximal gradients than CMT1A in 4 proximal lower limb muscles (VI: 36.0% vs. 5.9%, p < 0.001; VM: 10.7% vs. 2.2%, p = 0.03; AB: 2.9% vs. −2.1%, p = 0.01; Gmin: 1.2% vs. −15.3%, p < 0.001).
TABLE 2.
Distal to proximal fat replacement gradients.
| Muscle | Gradient (%) | p‐values | ||||
|---|---|---|---|---|---|---|
| DM1 | CMT1A | HC | CMT1A vs. HC | DM1 vs. HC | CMT1A vs. DM1 | |
| TA | 8.6 ± 18.5 | 14.9 ± 14.0 | 0.9 ± 2.9 | < 0.001 | 0.07 | 0.04 |
| PER | 6.6 ± 18.9 | 35.4 ± 29.9 | −3.3 ± 3.3 | < 0.001 | 0.03 | < 0.001 |
| EDL | 3.2 ± 17.9 | 20.9 ± 20.0 | −1.0 ± 3.1 | < 0.001 | 0.11 | < 0.001 |
| FDL | 7.8 ± 8.7 | 13.5 ± 10.2 | 6.6 ± 3.4 | < 0.001 | 0.90 | < 0.001 |
| FHL | 28.3 ± 27.0 | 28.8 ± 22.2 | 4.3 ± 3.1 | < 0.001 | < 0.001 | 0.88 |
| SOL | 20.0 ± 24.1 | 24.1 ± 25.2 | 4.0 ± 2.7 | < 0.001 | < 0.001 | 0.44 |
| GM | 10.4 ± 29.5 | 22.2 ± 28.3 | 0.7 ± 1.5 | < 0.001 | 0.13 | 0.13 |
| GL | 4.4 ± 15.7 | 6.0 ± 17.4 | −3.0 ± 1.8 | 0.03 | 0.03 | 0.78 |
| TP | 4.5 ± 10.8 | 8.3 ± 16.5 | 0.2 ± 2.5 | < 0.001 | 0.06 | 0.13 |
| POP | −0.2 ± 3.9 | −0.1 ± 7.6 | −0.3 ± 1.7 | 0.20 | 0.62 | 0.16 |
| RF | −1.0 ± 3.4 | −1.6 ± 3.5 | −2.6 ± 2.1 | 0.32 | 0.08 | 0.54 |
| VM | 10.7 ± 19.4 | 2.2 ± 3.2 | 1.0 ± 1.2 | 0.03 | < 0.001 | 0.03 |
| VI | 36.0 ± 29.8 | 5.9 ± 9.6 | 1.0 ± 2.1 | < 0.001 | < 0.001 | < 0.001 |
| VL | 1.1 ± 9.6 | −1.3 ± 3.3 | −3.2 ± 2.5 | 0.02 | < 0.001 | 0.13 |
| SAR | 20.2 ± 26.9 | 8.9 ± 13.8 | 0.0 ± 2.0 | < 0.001 | < 0.001 | 0.43 |
| AB | 2.9 ± 19.2 | −2.1 ± 1.8 | −0.8 ± 1.4 | < 0.001 | 0.81 | 0.01 |
| AL | −0.1 ± 14.6 | −0.6 ± 2.5 | −0.9 ± 1.8 | 0.63 | 0.63 | 0.63 |
| AM | 9.1 ± 23.0 | 4.5 ± 5.4 | 1.3 ± 2.0 | 0.02 | 0.07 | 0.53 |
| GRA | 11.2 ± 19.5 | 3.7 ± 4.1 | 1.1 ± 2.1 | < 0.001 | < 0.001 | 0.80 |
| PEC | −0.7 ± 2.7 | −1.8 ± 3.2 | −0.5 ± 2.7 | 0.24 | 0.96 | 0.24 |
| BFB | 6.3 ± 17.5 | 5.0 ± 5.4 | 4.0 ± 2.2 | 0.79 | 0.55 | 0.62 |
| BFL | 2.0 ± 10.3 | −1.2 ± 3.0 | 0.4 ± 2.9 | 0.07 | 0.70 | 0.24 |
| ST | 11.1 ± 25.2 | −0.4 ± 3.6 | −0.3 ± 1.9 | 0.85 | 0.85 | 0.85 |
| SM | 2.9 ± 6.1 | 3.6 ± 5.2 | −0.2 ± 1.6 | < 0.001 | 0.01 | 0.17 |
| TFL | 9.4 ± 13.7 | 10.3 ± 8.9 | 1.8 ± 4.5 | < 0.001 | < 0.001 | 0.18 |
| Gmax | 11.8 ± 17.6 | 9.3 ± 9.6 | 7.5 ± 6.4 | 0.81 | 0.81 | 0.92 |
| Gmed | 1.1 ± 4.8 | −1.1 ± 5.6 | 2.8 ± 3.3 | < 0.001 | 0.02 | 0.07 |
| Gmin | 1.2 ± 26.9 | −15.3 ± 16.5 | −3.5 ± 6.4 | < 0.001 | 0.41 | < 0.001 |
Note: Data are presented as mean ± standard deviation. The gradient represents the change in PDFF (%) based on a linear regression. A positive gradient indicates a distal‐dominant fat distribution, while a negative gradient indicates a proximal‐dominant distribution. Significant p‐values are marked in bold.
Abbreviations: AB, Adductor brevis; AL, Adductor longus; AM, Adductor magnus; BFB, Biceps femoris brevis; BFL, Biceps femoris longus; CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; EDL, Extensor digitorum longus; FDL, Flexor digitorum longus; FHL, Flexor hallucis longus; GL, Gastrocnemius lateralis; GM, Gastrocnemius medialis; Gmax, Gluteus maximus; Gmed, Gluteus medius; Gmin, Gluteus minimus; GRA, Gracilis; HC, Healthy controls; PEC, Pectineus; PER, Peroneus; POP, Popliteus; RF, Rectus femoris; SAR, Sartorius; SM, Semimembranosus; SOL, Soleus; ST, Semitendinosus; TA, Tibialis anterior; TFL, Tensor fasciae latae; TP, Tibialis posterior; VI, Vastus intermedius; VL, Vastus lateralis; VM, Vastus medialis.
3.4. Texture Features
Figure 3 illustrates the distinct micro‐spatial pattern of fat replacement on T1‐weighted images in representative cases of both patient cohorts. Entropy and contrast were increased across all muscles in patients (all p < 0.001; Tables S2 and S3), while homogeneity was significantly reduced (Table S4).
FIGURE 3.
Representative T1‐weighted MRI of fat replacement patterns of CMT1A and DM1 patients. Axial T1‐weighted MRI of the distal lower limbs demonstrating the qualitative, phenotypic patterns that correspond to the quantitative radiomic features identified in this study. (A, B) illustrate a reticular pattern typical of neurogenic changes in CMT1A patients, corresponding to the high entropy and contrast observed in the texture analysis. (C, D) illustrate a more homogeneous, confluent pattern characteristic of myogenic changes in DM1 patients, reflecting higher homogeneity. While these T1‐weighted MRI examples highlight clear phenotypic extremes, radiomic texture analysis provided an objective quantification of these differences across the entire cohort. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; MRI, Magnetic resonance imaging.
Direct comparison between patient groups revealed a consistent directional pattern, with CMT1A exhibiting higher entropy and contrast and lower homogeneity than DM1 (Figure 4, Tables S2–, S4). In distal lower limb muscles, both entropy and contrast were significantly higher in CMT1A than in DM1, particularly in the gastrocnemius medialis (entropy: 9.7 vs. 8.8, p = 0.03; contrast: 107.8 vs. 68.2, p < 0.05) and tibialis posterior (entropy: 8.3 vs. 7.6, p < 0.01; contrast: 71.0 vs. 43.0, p < 0.01). Homogeneity was significantly lower in the extensor digitorum longus (0.989 vs. 0.991, p = 0.04), gastrocnemius medialis (0.988 vs. 0.991, p = 0.04), tibialis posterior (0.992 vs. 0.995, p < 0.001), and popliteus (0.991 vs. 0.993, p < 0.01).
FIGURE 4.
Micro‐spatial fat replacement pattern DM1 vs. CMT1A based on radiomic texture analysis. Heatmap of t‐values comparing radiomic texture features between DM1 and CMT1A, adjusted for PDFF (%). This analysis identifies micro‐spatial differences in muscle tissue “texture” that differentiate the two diseases while correcting for mean PDFF (%). Blue indicates higher texture feature values in DM1, while green indicates higher values in CMT1A. Asterisks show Benjamini–Hochberg (BH) adjusted significance (* p < 0.05, **p < 0.01, ***p < 0.001). CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1. For muscle abbreviations, see Figure 1.
A similar pattern was observed in proximal lower limb muscles (Figure 4, Tables S2–, S4). Entropy was significantly higher in CMT1A in the gracilis (7.8 vs. 7.5, p < 0.05), biceps femoris longus (8.0 vs. 7.6, p < 0.01), semitendinosus (8.3 vs. 7.6, p < 0.001), semimembranosus (8.7 vs. 8.1, p = 0.02), tensor fasciae latae (9.7 vs. 9.3, p = 0.04), gluteus maximus (9.3 vs. 8.9, p = 0.03), and gluteus minimus (8.8 vs. 8.1, p = 0.01). Additionally, contrast was higher in the biceps femoris longus (37.4 vs. 26.6, p = 0.03), semitendinosus (52.1 vs. 36.1, p = 0.02), and gluteus minimus (56.5 vs. 39.5, p < 0.05). Consistently, homogeneity was significantly lower in CMT1A in the sartorius (p < 0.001), biceps femoris brevis (p = 0.02), semitendinosus (p < 0.001), gluteus maximus (p = 0.02), and gluteus minimus (p < 0.001). No significant correlations were observed between CTG repeat length and any imaging‐derived parameter in the DM1 cohort, but longer disease duration correlated with higher mean PDFF (%) and with greater textural disruption (higher entropy and contrast, lower homogeneity) in multiple muscles. In CMT1A, correlations were directionally consistent but did not reach significance after correction for multiple comparisons (Table S5).
4. Discussion
This study demonstrates that the integration of micro‐spatial qMRI features may significantly enhance the objective differential diagnosis of DM1 and CMT1A as proof of concept and warrants further investigations in other myogenic and neurogenic diseases. Although both DM1 and CMT1A showed higher PDFF (%) compared with matched controls, radiomic texture analysis revealed distinct micro‐spatial alterations in the two conditions. These findings indicate that the spatial distribution of fat replacement might provide diagnostic information. Furthermore, these micro‐spatial patterns were complemented by disease‐specific signatures in muscle involvement, asymmetry, and disto‐proximal gradients.
4.1. Radiomic Texture Analysis Captures Micro‐Spatial Differences Between Adult‐Onset DM1 and CMT1A
The micro‐spatial distribution of fat replacement in neurogenic conditions shows a reticular pattern that may progress to muscle islands, whereas myopathies demonstrate a more diffuse appearance [7, 8, 10]. Studies describing these patterns have relied on qualitative visual evaluation on T1‐weighted images. Esteller et al. reported a reticular pattern in neurogenic cohorts [10], and Payá et al. confirmed that this pattern strongly characterizes neurogenic disease in contrast to the more homogeneous appearance observed in myopathies [7]. Radiomic texture analysis offers a quantitative approach to capture these visual differences [14, 17]. Only one previous study applied texture analysis in CMT1A, demonstrating increased textural heterogeneity compared to healthy controls using T1‐weighted MRI [18]. However, radiomics features derived from conventional T1‐weighted or STIR imaging are inherently influenced by scanner‐ and protocol‐dependent intensity variations. Burian et al. [21] and Dieckmeyer et al. (2021) extended texture analysis to PDFF (%) images in healthy controls, validating PDFF‐based radiomics and showing that texture features reflect structural information beyond mean fat fraction [15, 21]. By performing radiomics on quantitative PDFF (%) maps, our approach provides a methodologically sound framework for quantitative texture analysis. Recent studies have highlighted the growing utility of machine learning and radiomics in evaluating NMDs. Nagawa et al. (2021) extracted radiomic features using machine learning from STIR images; their study successfully predicted disease subgroups and autoantibody status in idiopathic inflammatory myopathies, offering a valuable, non‐invasive alternative to muscle biopsies [19]. Similarly, Wei et al. (2023) demonstrated that machine learning‐based radiomics could differentiate limb‐girdle muscular dystrophy R2 (LGMDR2) from immune‐mediated necrotizing myopathy (IMNM) with significantly higher accuracy than standard visual MRI evaluations [20]. To our knowledge, no prior study has applied 3D PDFF‐based texture analysis to differentiate DM1 and CMT1A. In our study, both cohorts demonstrated increased entropy and contrast and reduced homogeneity compared with controls, reflecting disruption of normal muscle architecture. Notably, CMT1A exhibited significantly greater textural heterogeneity than DM1. This high‐entropy, high‐contrast profile is consistent with the reticular fat replacement pattern characteristic of neurogenic disease, whereas DM1 showed a more homogeneous texture, aligning with the diffuse myopathic infiltration. These findings are a proof of concept that PDFF‐based 3D radiomic texture analysis might objectify qualitative imaging patterns and capture complementary micro‐spatial information beyond conventional mean fat fraction assessment, supporting its potential role in objective disease differentiation.
4.2. Disto‐Proximal Gradients Quantify Length‐Dependent Pathology
The disto‐proximal fat gradients provided quantitative evidence of the length‐dependent pathology. CMT1A exhibited significantly steeper distal gradients across nearly all lower limb muscles compared to both controls and DM1. This is consistent with the established length‐dependent pathophysiology of CMT1A. Conversely, DM1 demonstrated higher disto‐proximal gradients in several proximal thigh muscles compared to CMT1A, reflecting a distinct, disease‐specific trajectory of fat infiltration that does not strictly adhere to the neurogenic length‐dependent model [29, 30]. However, it should be noted that both diseases studied have a distal‐predominant involvement, and CMT1A in particular is length‐dependent. This interpretation should therefore be limited to distal and length‐dependent disease models and not extrapolated to the full spectrum of neurogenic and myogenic disorders.
4.3. Disease‐Specific Signatures: Compartmental Involvement, Atrophy, and the Role of Asymmetry
Consistent with prior reports, DM1 predominantly affected the posterior distal compartment, whereas CMT1A mainly involved the anterolateral compartment, with relative sparing of proximal lower limb muscles in both groups [29, 31, 32, 33]. Importantly, these distributions reflect disease‐specific rather than inherently myogenic or neurogenic patterns. Other neurogenic conditions may preferentially involve the posterior compartment, and certain myopathies may show predominant anterolateral involvement [7, 10, 34]. This underscores the importance of integrating multiple qMRI features. Beyond the specific muscles affected, muscle volume and asymmetry analysis revealed further discriminatory features. Both patient groups exhibited significant distal atrophy, yet the specific muscles affected closely mirrored their respective fat replacement patterns, with profound atrophy in the anterolateral compartment for CMT1A and the posterior compartment for DM1. Additionally, our quantitative approach detected significant asymmetry in the gastrocnemius medialis and peroneus longus of the CMT1A cohort, whereas DM1 showed no significant difference in the individual muscles compared to controls. These findings suggest that evaluating the involvement of individual muscles is more informative for the differential diagnosis, as global measures can mask disease‐specific involvement. Payá et al. (2026) reported that asymmetric muscle involvement and hypertrophy were more common in distal myopathies than in hereditary neuropathies [7]. These differences reflect the specific disease models used in our study. While myopathies as a broader category, such as facioscapulohumeral muscular dystrophy (FSHD), can present with asymmetry [13, 35], DM1 typically manifests with a relatively symmetric fat replacement [32]. Because macroscopic measures like asymmetry and atrophy can overlap between myogenic and neurogenic diseases, they are insufficient for the differentiation of neurogenic and myogenic diseases.
4.4. Strengths and Limitations
This study has several strengths. DM1 and CMT1A are well‐characterized, genetically confirmed distal myogenic and neurogenic disorders, providing a robust framework for systematic MRI comparison. The use of the automated 3D segmentation pipeline enabled analysis of 28 individual muscles, and the inclusion of matched healthy controls provided reliable reference values.
However, the limitations of our study should be considered. First, DM1 and CMT1A were investigated as proof‐of‐concept models, and generalizability to other myogenic and neurogenic conditions remains to be established. Moreover, DM1 is more heterogeneous than CMT1A, and the muscle MRI phenotype in DM1 can vary accordingly [32]. Inclusion was therefore restricted to genetically confirmed adult‐onset DM1. Second, the modest sample size may have limited statistical power, particularly for subgroup analyses, and larger cohorts are needed to confirm robustness. Third, in DM1, disease duration correlated with PDFF (%) and texture features. As the texture comparisons were adjusted for PDFF (%) and disease duration did not differ between cohorts, this does not account for the between‐group differences, but longitudinal studies are needed to confirm progression. Fourth, we utilized a hypothesis‐driven selection of three a priori GLCM features specifically chosen to mathematically quantify the reticular and diffuse visual patterns. Future studies with larger cohorts should evaluate additional feature classes, such as run‐length or size‐zone matrices. Finally, the texture features in patients with low or overlapping PDFF (%) values were not evaluated. Texture differences were nonetheless detected between patients and healthy controls, indicating a discriminative signal at low fat fractions. Dedicated analyses in early‐disease cohorts are required to confirm this in patients.
5. Conclusion
This proof‐of‐concept study demonstrates that quantitative muscle MRI captures complementary imaging features that may aid in distinguishing CMT1A from adult‐onset DM1. By quantifying the micro‐spatial patterns, 3D texture analysis might offer diagnostic information that extends beyond the traditional qualitative assessment. These micro‐spatial markers are complemented by macroscopic features, including disease‐specific patterns of muscle atrophy, asymmetry, and disto‐proximal gradients. Future studies should determine whether integrating radiomic and conventional qMRI features improves the non‐invasive differentiation of a wider range of neurogenic and myogenic disorders and evaluate their potential use in automated diagnostic pipelines.
Author Contributions
Veerle Goosens: writing – review and editing. Kobe Bamps: writing – review and editing, formal analysis. Lotte Huysmans: methodology, writing – review and editing, formal analysis. Louise Iterbeke: conceptualization, investigation, writing – original draft, methodology, visualization, writing – review and editing, formal analysis, data curation, project administration. Ronald Peeters: writing – review and editing, methodology. Kristl G. Claeys: conceptualization, writing – original draft, methodology, visualization, writing – review and editing, formal analysis, data curation, project administration, supervision. Patrick Dupont: writing – review and editing, methodology, formal analysis. Frederik Maes: writing – review and editing, methodology.
Funding
This study was supported in part by the Flemish Government under the Onderzoeksprogramma Artificiële Intelligentie (AI) Vlaanderen program.
Ethics Statement
This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee Research UZ/KU Leuven (s67475 and s67476). Written informed consent was obtained from all participants prior to inclusion in the study.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Mean PDFF (%) of all individual muscles.
Comparison of mean proton density fat fraction (PDFF, %) across distal and proximal lower limb muscles in patients with DM1, CMT1A, and HC. Data are presented as mean ± standard deviation (SD). Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. AB, Adductor brevis; AL, Adductor longus; AM, Adductor magnus; BFB, Biceps femoris brevis; BFL, Biceps femoris longus; CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; EDL, Extensor digitorum longus; FDL, Flexor digitorum longus; FHL, Flexor hallucis longus; GL, Gastrocnemius lateralis; GM, Gastrocnemius medialis; Gmax, Gluteus maximus; Gmed, Gluteus medius; Gmin, Gluteus minimus; GRA, Gracilis; HC, Healthy controls; PEC, Pectineus; PER, Peroneus; POP, Popliteus; RF, Rectus femoris; SAR, Sartorius; SD, Standard deviation; SM, Semimembranosus; SOL, Soleus; ST, Semitendinosus; TA, Tibialis anterior; TFL, Tensor fasciae latae; TP, Tibialis posterior; VI, Vastus intermedius; VL, Vastus lateralis; VM, Vastus medialis.
Table S2: Muscle texture entropy analysis.
Comparison of entropy across lower limb muscles in patients with DM1, CMT1A, and healthy controls (HC). Entropy serves as a measure of tissue complexity and randomness; higher values indicate a more disordered, “marbled,” or reticular pattern of fat infiltration, while lower values suggest a more predic uniform tissue structure. Data are presented as Estimated Marginal Means ± Standard Error (SE), derived from ANCOVA models adjusted for fat fraction (PDFF %). Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S3: Muscle texture contrast analysis.
Comparison of contrast across lower limb muscles in patients with DM1, CMT1A, and healthy controls (HC). Contrast quantifies local intensity variations, representing the “sharpness” and frequency of transitions between muscle tissue and fat islets. Higher contrast values typically reflect a heterogeneous infiltration pattern where distinct boundaries between muscle and fat are maintained. Data are presented as Estimated Marginal Means ± Standard Error (SE), derived from ANCOVA models adjusted for PDFF %. Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S4: Muscle texture homogeneity analysis.
Comparison of estimated marginal means for the texture feature “Homogeneity” across lower limb muscles in patients with DM1, CMT1A, and healthy controls. Homogeneity values quantify the local uniformity of image signal, where higher values (closer to 1.0) represent more confluent or uniform tissue structures and lower values indicate increased local signal variation. Data are presented as adjusted Mean ± Standard Error (SE), derived from ANCOVA models adjusted for PDFF %. Pairwise comparisons are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S5: Spearman correlations between disease duration and quantitative MRI features in DM1 and CMT1A.
Values are Spearman correlation coefficients (ρ) between disease duration and each feature. Bold indicates statistical significance after BH correction for multiple comparisons (adjusted p < 0.05). Positive ρ for PDFF, entropy and contrast, and negative ρ for homogeneity indicate greater fat replacement and textural disruption with longer disease duration. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; PDFF, proton density fat fraction; For muscle abbreviations, see legend Table S1.
Acknowledgements
The authors are grateful to the patients and healthy controls for their participation in this study and to Stefan Ghysels, Kris Byloos, and Guido Putzeys for their support with the MRI assessments.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- 1. Savarese M., Jokela M., and Udd B., “Distal Myopathy,” in Handbook of Clinical Neurology, vol. 195 (Elsevier BV, 2023), 497–519, 10.1016/B978-0-323-98818-6.00002-9. [DOI] [PubMed] [Google Scholar]
- 2. Kuhlenbåumer G., Staegbauer F., Ringelstein E. B., and Young P., Hereditary Peripheral Neuropathies, 1st ed. (Steinkopff, 2005), 10.1007/3-7985-1586-7. [DOI] [Google Scholar]
- 3. Milone M., “A Pattern Recognition Approach to Myopathy,” Continuum (Minneap Minn) 31, no. 5 (2025): 1244–1269, 10.1212/cont.0000000000001611. [DOI] [PubMed] [Google Scholar]
- 4. Rosenberg A., Tian C., He H., Ulm E., Collins Ruff K., and Nagaraj C., “An Evaluation of Clinical Presentation and Genetic Testing Approaches for Patients With Neuromuscular Disorders,” American Journal of Medical Genetics. Part A 191, no. 11 (2023): 2679–2692, 10.1002/ajmg.a.63356. [DOI] [PubMed] [Google Scholar]
- 5. Frasquet M. and Sevilla T., “Hereditary Motor Neuropathies,” Current Opinion in Neurology 35, no. 5 (2022): 562–570, 10.1097/WCO.0000000000001087. [DOI] [PubMed] [Google Scholar]
- 6. Bansagi B., Griffin H., Whittaker R. G., et al., “Genetic Heterogeneity of Motor Neuropathies,” Neurology 88, no. 13 (2017): 1226–1234, 10.1212/WNL.0000000000003772. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Payá M., Portela S., Sivera R., et al., “Muscle MRI Contributes to the Differential Diagnosis Between Distal Myopathies and Distal Hereditary Motor Neuropathies,” European Journal of Neurology 33, no. 1 (2026): e70456, 10.1111/ene.70456. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Astrea G., Morrow J. M., Manzur A., et al., “Muscle “Islands”: An MRI Signature Distinguishing Neurogenic From Myopathic Causes of Early Onset Distal Weakness,” Neuromuscular Disorders 32, no. 2 (2022): 142–149, 10.1016/j.nmd.2021.11.003. [DOI] [PubMed] [Google Scholar]
- 9. Díaz‐Manera J., Llauger J., Gallardo E., and Illa I., “Muscle MRI in Muscular Dystrophies,” Acta Myologica 34, no. 2–3 (2015): 95–108. [PMC free article] [PubMed] [Google Scholar]
- 10. Esteller D., Morrow J., Alonso‐Pérez J., et al., “Muscle Magnetic Resonance Imaging of a Large Cohort of Distal Hereditary Motor Neuropathies Reveals Characteristic Features Useful for Diagnosis,” Neuromuscular Disorders 33, no. 10 (2023): 744–753, 10.1016/j.nmd.2023.08.010. [DOI] [PubMed] [Google Scholar]
- 11. Mercuri E., Talim B., Moghadaszadeh B., et al., “Clinical and Imaging Findings in Six Cases of Congenital Muscular Dystrophy With Rigid Spine Syndrome Linked to Chromosome 1p (RSMD1),” Neuromuscular Disorders 12, no. 7–8 (2002): 631–638, 10.1016/s0960-8966(02)00023-8. [DOI] [PubMed] [Google Scholar]
- 12. De Wel B., Huysmans L., Depuydt C. E., et al., “Histopathological Correlations and Fat Replacement Imaging Patterns in Recessive Limb‐Girdle Muscular Dystrophy Type 12,” Journal of Cachexia, Sarcopenia and Muscle 14, no. 3 (2023): 1468–1481, 10.1002/jcsm.13234. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Mul K., Vincenten S. C. C., Voermans N. C., et al., “Adding Quantitative Muscle MRI to the FSHD Clinical Trial Toolbox,” Neurology 89, no. 20 (2017): 2057–2065, 10.1212/WNL.0000000000004647. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. De Certaines J. D., Larcher T., Duda D., et al., “Application of Texture Analysis to Muscle MRI: 1‐What Kind of Information Should Be Expected From Texture Analysis?,” EPJ Nonlinear Biomedical Physics 3 (2011): 3, 10.1140/s40366-015-0017-1. [DOI] [Google Scholar]
- 15. Dieckmeyer M., Inhuber S., Schlaeger S., et al., “Texture Features of Proton Density Fat Fraction Maps From Chemical Shift Encoding‐Based Mri Predict Paraspinal Muscle Strength,” Diagnostics 11, no. 2 (2021): 239, 10.3390/diagnostics11020239. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Mayerhoefer M. E., Materka A., Langs G., et al., “Introduction to Radiomics,” Journal of Nuclear Medicine 61, no. 4 (2020): 488–495, 10.2967/JNUMED.118.222893. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Herlidou S., Rolland Y., Bansard J. Y., Le Rumeur E., and De Certaines J. D., “Comparison of Automated and Visual Texture Analysis in MRI: Characterization of Normal and Diseased Skeletal Muscle,” Magnetic Resonance Imaging 17, no. 9 (1999): 1393–1397, 10.1016/s0730-725x(99)00066-1. [DOI] [PubMed] [Google Scholar]
- 18. Lee J. H., Yoon Y. C., Kim H. S., Kim J. H., and Choi B. O., “Texture Analysis Using T1‐Weighted Images for Muscles in Charcot‐Marie‐Tooth Disease Patients and Volunteers,” European Radiology 31, no. 5 (2021): 3008–3017, 10.1007/s00330-020-07435-y. [DOI] [PubMed] [Google Scholar]
- 19. Nagawa K., Suzuki M., Yamamoto Y., et al., “Texture Analysis of Muscle MRI: Machine Learning‐Based Classifications in Idiopathic Inflammatory Myopathies,” Scientific Reports 11, no. 1 (2021): 9821, 10.1038/s41598-021-89311-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Wei P., Zhong H., Xie Q., et al., “Machine Learning‐Based Radiomics to Differentiate Immune‐Mediated Necrotizing Myopathy From Limb‐Girdle Muscular Dystrophy R2 Using MRI,” Frontiers in Neurology 14 (2023): 14, 10.3389/fneur.2023.1251025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Burian E., Becherucci E. A., Junker D., et al., “Association of Cervical and Lumbar Paraspinal Muscle Composition Using Texture Analysis of Mr‐Based Proton Density Fat Fraction Maps,” Diagnostics 11, no. 10 (2021): 1929, 10.3390/diagnostics11101929. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Wenninger S., Montagnese F., and Schoser B., “Core Clinical Phenotypes in Myotonic Dystrophies,” Frontiers in Neurology 9 (2018): 303, 10.3389/fneur.2018.00303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Reilly M. M., Murphy S. M., and Laurá M., “Charcot‐Marie‐Tooth Disease,” Journal of the Peripheral Nervous System 16 (2011): 1–14, 10.1111/j.1529-8027.2011.00324.x. [DOI] [PubMed] [Google Scholar]
- 24. Huysmans L., De Wel B., Claeys K. G., and Maes F., “Automated MRI Quantification of Volumetric Per‐Muscle Fat Fractions in the Proximal Leg of Patients With Muscular Dystrophies,” Frontiers in Neurology 14 (2023): 14, 10.3389/fneur.2023.1200727. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Andrade C., “Z Scores, Standard Scores, and Composite Test Scores Explained,” Indian Journal of Psychological Medicine 43, no. 6 (2021): 555–557, 10.1177/02537176211046525. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Van Griethuysen J. J. M., Fedorov A., Parmar C., et al., “Computational Radiomics System to Decode the Radiographic Phenotype,” Cancer Research 77, no. 21 (2017): e104–e107, 10.1158/0008-5472.CAN-17-0339. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Lerski R. A., de Certaines J. D., Duda D., et al., “Application of Texture Analysis to Muscle MRI: 2 – Technical Recommendations,” EPJ Nonlinear Biomedical Physics 3, no. 1 (2015): 2, 10.1140/epjnbp/s40366-015-0018-0. [DOI] [Google Scholar]
- 28. Zwanenburg A., Vallières M., Abdalah M. A., et al., “The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High‐Throughput Image‐Based Phenotyping,” Radiology 295, no. 2 (2020): 328–338, 10.1148/radiol.2020191145. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Bas J., Ogier A. C., Le Troter A., et al., “Fat Fraction Distribution in Lower Limb Muscles of Patients With CMT1A A Quantitative MRI Study,” Neurology 94, no. 14 (2020): e1480–e1487, 10.1212/WNL.0000000000009013. [DOI] [PubMed] [Google Scholar]
- 30. Fortanier E., Hostin M. A., Michel C., et al., “One‐Year Longitudinal Assessment of Patients With CMT1A Using Quantitative MRI,” Neurology 102, no. 9 (2024): e209277, 10.1212/WNL.0000000000209277. [DOI] [PubMed] [Google Scholar]
- 31. Kornblum C., “Myotonic Dystrophies,” in Neuromuscular Imaging (Springer, 2013), 279–293, 10.1007/978-1-4614-6552-2_22. [DOI] [Google Scholar]
- 32. Garibaldi M., Nicoletti T., Bucci E., et al., “Muscle Magnetic Resonance Imaging in Myotonic Dystrophy Type 1 (DM1): Refining Muscle Involvement and Implications for Clinical Trials,” European Journal of Neurology 29, no. 3 (2022): 843–854, 10.1111/ene.15174. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Kim Y. J., Kim H. S., Lee J. H., Yoon Y. C., and Choi B. O., “Magnetic Resonance Imaging‐Based Lower Limb Muscle Evaluation in Charcot‐Marie‐Tooth Disease Type 1A Patients and Its Correlation With Clinical Data,” Scientific Reports 12, no. 1 (2022): 16622, 10.1038/s41598-022-21112-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Venturelli N., Tordjman M., Ammar A., Chetrit A., Renault V., and Carlier R. Y., “Contribution of Muscle MRI for Diagnosis of Myopathy,” Revue Neurologique (Paris) 179, no. 1–2 (2023): 61–80, 10.1016/j.neurol.2022.12.002. [DOI] [PubMed] [Google Scholar]
- 35. Heskamp L., Ogier A., Bendahan D., and Heerschap A., “Whole‐Muscle Fat Analysis Identifies Distal Muscle End as Disease Initiation Site in Facioscapulohumeral Muscular Dystrophy,” Communications Medicine 2, no. 1 (2022): 155, 10.1038/s43856-022-00217-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Mean PDFF (%) of all individual muscles.
Comparison of mean proton density fat fraction (PDFF, %) across distal and proximal lower limb muscles in patients with DM1, CMT1A, and HC. Data are presented as mean ± standard deviation (SD). Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. AB, Adductor brevis; AL, Adductor longus; AM, Adductor magnus; BFB, Biceps femoris brevis; BFL, Biceps femoris longus; CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; EDL, Extensor digitorum longus; FDL, Flexor digitorum longus; FHL, Flexor hallucis longus; GL, Gastrocnemius lateralis; GM, Gastrocnemius medialis; Gmax, Gluteus maximus; Gmed, Gluteus medius; Gmin, Gluteus minimus; GRA, Gracilis; HC, Healthy controls; PEC, Pectineus; PER, Peroneus; POP, Popliteus; RF, Rectus femoris; SAR, Sartorius; SD, Standard deviation; SM, Semimembranosus; SOL, Soleus; ST, Semitendinosus; TA, Tibialis anterior; TFL, Tensor fasciae latae; TP, Tibialis posterior; VI, Vastus intermedius; VL, Vastus lateralis; VM, Vastus medialis.
Table S2: Muscle texture entropy analysis.
Comparison of entropy across lower limb muscles in patients with DM1, CMT1A, and healthy controls (HC). Entropy serves as a measure of tissue complexity and randomness; higher values indicate a more disordered, “marbled,” or reticular pattern of fat infiltration, while lower values suggest a more predic uniform tissue structure. Data are presented as Estimated Marginal Means ± Standard Error (SE), derived from ANCOVA models adjusted for fat fraction (PDFF %). Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S3: Muscle texture contrast analysis.
Comparison of contrast across lower limb muscles in patients with DM1, CMT1A, and healthy controls (HC). Contrast quantifies local intensity variations, representing the “sharpness” and frequency of transitions between muscle tissue and fat islets. Higher contrast values typically reflect a heterogeneous infiltration pattern where distinct boundaries between muscle and fat are maintained. Data are presented as Estimated Marginal Means ± Standard Error (SE), derived from ANCOVA models adjusted for PDFF %. Pairwise comparisons between groups are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S4: Muscle texture homogeneity analysis.
Comparison of estimated marginal means for the texture feature “Homogeneity” across lower limb muscles in patients with DM1, CMT1A, and healthy controls. Homogeneity values quantify the local uniformity of image signal, where higher values (closer to 1.0) represent more confluent or uniform tissue structures and lower values indicate increased local signal variation. Data are presented as adjusted Mean ± Standard Error (SE), derived from ANCOVA models adjusted for PDFF %. Pairwise comparisons are provided with FDR‐adjusted p‐values. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; For muscle abbreviations, see legend Table S1.
Table S5: Spearman correlations between disease duration and quantitative MRI features in DM1 and CMT1A.
Values are Spearman correlation coefficients (ρ) between disease duration and each feature. Bold indicates statistical significance after BH correction for multiple comparisons (adjusted p < 0.05). Positive ρ for PDFF, entropy and contrast, and negative ρ for homogeneity indicate greater fat replacement and textural disruption with longer disease duration. CMT1A, Charcot–Marie‐Tooth neuropathy type 1A; DM1, Myotonic dystrophy type 1; HC, Healthy controls; PDFF, proton density fat fraction; For muscle abbreviations, see legend Table S1.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.