Magnetic Resonance’s Impact on Diagnosis, Prognostication, and Therapeutic Approaches in Neuromuscular Disorders

April 27, 2021
Kenneth Bender, PharmD, MA

NeurologyLive, April 2021, Volume 4, Issue 2

Advances in MRI and MRS techniques and protocols enhance diagnosis of neuromuscular disorders and provide prognostic and intervention efficacy biomarkers.

IN ADDITION TO FACILITATING DIAGNOSIS, magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS) are increasingly used in neuromuscular disorders to detect and predict deterioration and to evaluate efficacy of disease-modifying interventions.

Developments and innovative applications of the technologies are revealing disease progression before it manifests in functional testing. These technologies are also providing quantifiable biomarker end points for clinical trials that are challenged to detect efficacy of interventions for rare, slowly progressing conditions in small cohorts of likely phenotypic variability.

“Now, an MR scanner is no longer viewed as a simple camera. [It’s seen] as an advanced scientific and clinical tool that not only takes ‘pictures,’ but also, via recently developed methodologies, can report on disease evolution, lean muscle tissue volume, fraction of muscle replaced by fat, metabolism, inflammation, and contractility of muscle,” wrote Julia Dahlqvist, MD, and colleagues in their recent review published in the Annals of Neurology.1 Dahlqvist is a member of the Department of Neurology, Copenhagen Neuromuscular Center, of Denmark’s Copenhagen University.

In the review, Dahlqvist and colleagues point to Duchenne muscular dystrophy (DMD) as one of the conditions that is more accurately and usefully characterized with the advancing MRI technology. They note studies that have correlated muscle histology and MRI changes, and they point out advances that have been made since the initial visual grading systems were introduced for semiquantification of fat infiltration in muscle.

“Visual grading systems are satisfactory for cross-sectional characterization of disease severity, but not to monitor subtle changes over time as needed in clinical trials,” Dahlqvist and colleagues wrote.

The Dixon imaging technique, utilizing multiecho-chemical-shift encoded water-fat imaging, is now the most frequently used method to quantify fat deposition in muscular dystrophy. In contrast to semiquantitative scoring, the Dixon technique provides the fat fraction as a continuous variable. This allows for greater sensitivity in detecting small differences in fat infiltration, according to another review by Doris Leung, MD, PhD, of the Center for Genetic Muscle Disorders, Hugo W. Moser Institute at Kennedy Krieger Institute, and the Department of Neurology, Johns Hopkins School of Medicine, Baltimore, Maryland.2

Others rely on proton spectroscopy to measure increases in fat fraction, especially in younger patients in the early stages of the disease.3 Dixon imaging is advantageous in providing spatial resolution, acknowledged Krista Vandenborne, PhD, PT, of the Department of Physical Therapy, University of Florida in Gainesville, “but MRS is, in our view, a more sensitive and robust method for quantifying fat fraction,” she commented to NeurologyLive®.

Muscle edema or inflammation often precedes fat replacement in many muscular dystrophies, according to Dahlqvist and colleagues. They note the utility of both relaxation time (T2) mapping and of sensitive, fat-saturated sequences—most often, short tau inversion recovery (STIR) to visualize muscle edema—and find T2 relaxation time mapping superior to STIR in quantifying the level of edema.

In addition to progressive replacement of muscle by fat, often accompanied by and preceded by edema, Dahlqvist and colleagues note that fibrosis commonly occurs in connective tissue, resulting in the muscle stiffness characteristic of many neuromuscular conditions, especially muscular dystrophies. They find potential for assessing muscle fibrosis in other technologies that include MR elastography, magnetization transfer, and sodium imaging (MRI imaging conducted with nonproton nuclei, such as 23Na).

Determining MR Biomarkers for DMD

The ImagingDMD study investigated the potential of MRI and MRS to be used in lower and upper extremity muscles to reveal biomarker end points for clinical trials of therapeutics for DMD. Vandenborne and colleagues characterized longitudinal progression of lower extremity muscle MRI/MRS biomarkers, examined their relationship with functioning over time, and ascertained their validity for predicting clinically meaningful sentinel events.3

“Although a body of literature exists [that establishes] quantitative MR measures as high-quality biomarkers for DMD, a high burden of proof is required to establish MR biomarkers as secondary end points or surrogate outcomes,” Vandenborne and colleagues indicated.

The largest MR natural history study in DMD to date, the ImagingDMD study includes data collected over 10 years and more than 100 participants completing at least 4 years of data collection. “The ImagingDMD study’s long duration has been important for 2 reasons,” Vandenborne explained to NeurologyLive®. “First, it has allowed us to evaluate the predictive value of MR biomarkers— measuring early MR markers of disease involvement, then prospectively monitoring each individual’s functional ability to quantify changes over time,” Vandenborne said. “Second, it has allowed us to capture the natural history of DMD and evaluate patterns within and [among] individuals over a wide range of ages and disease severities.”

The investigators monitored 160 participants with DMD at 3 study sites. The initial inclusion criteria required participants to be able to walk at least 100 meters and climb up 4 stair steps, but the criteria later expanded to include nonambulatory individuals. At the baseline visit, participants underwent an MRI and MRS examination of the lower leg and thigh, followed by clinical assessments of ambulatory function. Participants returned annually for follow-up MR and functional and medical history data collection, for up to 7 years.

“The disease trajectory can vary considerably [among] individuals and muscles, independent of genetic mutation or corticosteroid treatment,” Vandenborne commented. “Recognizing this variability is critically important in designing clinical trials, matching study arms, and ensuring that [a] study is properly powered.” Vandenborne and colleagues reported that vastus lateralis fat fraction (FF) measured by MRS, vastus lateralis MRI T2, and biceps femoris long head MRI T2 biomarkers were the fastest progressing biomarkers over time in this primarily ambulatory cohort. Over the study period, biomarker values tended toward a nonlinear, sigmoidal trajectory.

The investigators found that the lower extremity biomarkers predicted functional performance over the subsequent 12 and 24 months; additionally, the magnitude of change in an MRI/MRS biomarker over time was related to magnitude of change in function.

“The longitudinal study has allowed us to establish that the current FF in the leg muscles is highly predictive of future functional changes,” Vandenborne said. “We can predict, based on someone’s MR data, the likelihood that they will maintain their walking ability over the coming year or years.”

The Utility of Sodium MRI

With availability of high-field MRI systems, sodium MRI is increasingly being utilized for studies of muscle dystrophies such as DMD as well as muscular channelopathies. In a recent review, Marc-André Weber, MD, MSc, of the Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology, at University Medical Center Rostock, in Germany, and colleagues found that 23Na MRI has demonstrated “a muscular Na+ overload” in DMD.4

“This permanent Na+ overload in all DMD patients may be osmotically relevant and may play a role in the development of the muscle edema that was present and persisted in all studied DMD patients,” Weber and colleagues posit.

The prevailing theory of muscle edema in DMD, according to Weber et al, is that it represents the initial inflammatory stage of the ongoing muscle degeneration. They note that increased Na+ concentrations can be present even without apparent fatty degenerative changes or increased T2 relaxation times. The possibility that increased sodium concentrations are relevant to pathogenesis has prompted studies examining effects of diuretic drugs in cell models of DMD and in exploratory human trials.

“This demonstrates that 23Na MRI can be a very sensitive technique for the analysis of muscular diseases,” Weber and colleagues conclude.

Artificial Intelligence and ML Quicken MRI

The capacity to reveal minute distinctions across innumerable tissue slice images represents both the value of MRI technology and the challenge of correct assessment and interpretation. The application of algorithms and of machine learning, however, promises more efficient and rapid elucidation of clinically significant findings. In a recent proof-of-concept study, an active contour-evolution algorithm was successfully employed to enable semiautomated quantification of muscle MRI results. It was “reliable and time-efficient in determination of muscle volumes of neuromuscular patients,” the investigators related.5

In this study, Madlaine Müller, of the Department of Neurology at University Hospital Aachen in Germany, and colleagues described the results they achieved when performing 3D semiautomated segmentation of standardized muscle MRI datasets to analyze whole-muscle volumes of thighs and lower legs in 65 patients with established neuromuscular diagnoses (from molecular genetics, histological examination, and/or clinical tests).

The investigators reported similar determination of muscle volume for all subjects—patients with either neuropathy or myopathy, and controls—whether measured by semiautomatic algorithm assist or by manual segmentation (semiautomated: 2613 cm3; manual: 2594 cm3). There was, however, significant time efficiency with the semiautomated approach across patient groups (29.7 [±2.7] vs 399.3 [±9.2] minutes per patient; P <.0001).

Recognizing this variability [in disease] is critically important in designing clinical trials, matching study arms, and ensuring a study is properly powered.
— KRISTA VANDENBORNE, PHD, PT

In addition to demonstrating the efficiency in ascertaining muscle volume, the investigators are working on a protocol for semiautomated segmentation of the FF. Müller et al posit, however, that their work in improving efficiency in measuring remaining muscle volume could help detect disease progression in long-term studies “more accurately, or even before the fat fraction shows any significant differences.” Machine Learning Models and MRI While algorithms that improve efficiencies in building MRI datasets will help speed results, technology that boosts proper interpretation of these data could improve the relevance of the data to diagnosis and patient care. Machine learning (ML) platforms are being developed to enhance the utility of MRI in untangling complex, and often overlapping, patterns to differentiate among neuromuscular disorders.

In the genomics realm, next-generation sequencing has increased diagnostic efficiency in neuromuscular disorders, but it has limitations. These could be alleviated at least in part by enhanced MRI-aided diagnostics, according to Jordi Díaz-Manera, MD, PhD, of the Neuromuscular Disorders Unit, Neurology Department, Hospital de Santa Creu i de Sant Pau, Barcelona, Spain, and colleagues.

“The candidate gene sometimes does not fit with the phenotype, potential disease-causing variants in more than 1 gene can be found, or variants of unknown effect can be identified,” Díaz-Manera and colleagues point out. “In all these situations, clinical data and results of complementary tests continue to be of great value to make sense of the results obtained.”

Díaz-Manera and colleagues sought to develop ML-enhanced MRI as an informatic tool to help the diagnostic process for neuromuscular disorders.6 They used a “random forest” methodology to analyze the data and to find a model that would best distinguish among the different disorders. “Random forest is a [ML] tool capable of fitting large datasets and performing both classification and regression tasks,” Díaz-Manera and colleagues explain.

The ML program was trained on almost 1000 MRI datasets from the lower limbs of patients with 10 different diseases, against previously confirmed diagnoses. Two thousand different models were generated, and the best performed with 95.7% accuracy with an overall sensitivity of 92.1% and specificity of 99.4%.

The investigators then tested the ML diagnostics program on scans from 20 new, undiagnosed patients, against the diagnoses rendered by 4 neuromuscular MRI experts who had also not previously encountered them. The model outperformed each of the 4 experts.

“The results were impressive,” declared Jasper Morrow, MBChB, PhD, of the Department of Neuromuscular Diseases, Queen Square UCL Institute of Neurology, London, England, and Maria Sormani, PhD, of the Department of Health Sciences, University of Genoa, and IRCCS Osperdale Policlinico, San Martino, both in Italy, in commentary accompanying the published study.“7

This is an exciting proof-of-concept study applying ML to pattern assessment of lower-limb muscle MRI involvement in muscular dystrophies,” Morrow and Sormani wrote.

The Future of MRI and MRS

Vandenborne foresees continued acceleration of technological capabilities. Citing the 10-year ImagingDMD study as a pivotal example, she anticipates that the advances will yield increasing returns to patients and practitioners. The ImagingDMD team is particularly focused on helping transition-sensitive MR biomarkers into clinical trials, with the goal of accelerating therapeutic development.

“At this point, an MR Imaging Biomarker Steering Committee, consisting of industry stakeholders, private foundations, and other DMD community representatives, [is] helping us prepare a fit-for-purpose application to [submit to] both the FDA and the European Medicines Agency,” Vandenborne explained.

She continued that the 10-year ImagingDMD data will serve as the backbone for this initiative. They will be the foundation for creating an MR biomarker clinical trial simulation tool that could offer its users a way to simulate hypothetical drug effects on disease progression characteristics for particular subpopulations of patients with DMD. Additionally, Vandenborne said, it will help with investigators’ selection of clinical trial design characteristics.

“This tool will be open and made publicly available in order to facilitate easy access, broad use, and high impact,” Vandenborne said.

REFERENCES
1. Dahlqvist JR, Widholm P, Dahlqvist Leinhard O, Vissing J. MRI in neuromuscular diseases: an emerging diagnostic tool and biomarker for prognosis. Ann Neurol. 2020;88(4):669-681. doi:10.1002/ana.25804
2. Leung DG. Advancements in magnetic resonance imaging–based biomarkers for muscular dystrophy. Muscle Nerve. 2019;60(4):347-360. doi:10.1002/mus.26497 
3. Barnard AM, Willcocks RJ, Triplett WT, et al. MR biomarkers predict clinical function in Duchenne muscular dystrophy. Neurology. 2020;94(9):e897-e909. doi:10.1212/WNL.0000000000009012
4. Weber M-A, Nagel AM, Kan HE, Wattjes MP. Quantitative imaging in muscle diseases with focus on non-proton MRI and other advanced MRI techniques. Semin Musculoskelet Radiol. 2020;24(4):402-412. doi:10.1055/s-0040-1712955
5. Müller M, Dohrn MF, Romanzetti S, et al. Semi-automated volumetry of MRI serves as a biomarker in neuromuscular patients. Muscle Nerve. 2020;61(5):600-607. doi:10.1002/mus.26827
6. Verdύ-Díaz J, Alonso-Pérez J, Nuῆez-Peralta C, et al. Accuracy of a machine learning muscle MRI–based tool for the diagnosis of muscular dystrophies. Neurology. 2020;94(10):e1094-e1102. doi:10.1212/WNL.0000000000009068
7. Morrow JM, Pia Sormani M. Machine learning outperforms human experts in MRI pattern analysis of muscular dystrophies. Neurology. 2020;94(10):421-422. doi:10.1212/WNL.0000000000009053
download issueDownload Issue : April 2021