Machine learning muscle magnetic resonance imaging (MRI)-based tool may help physicians in the diagnostic process of muscular dystrophies, according to study results published in Neurology.

Although next-generation sequencing significantly improved the diagnostic efficiency for muscular dystrophies, it has various limitations. Machine learning uses algorithms to analyze data, learns from them, and then makes a determination or predictions. The researchers applied a machine learning strategy to a large dataset of muscle MRI studies performed in patients with genetically confirmed diagnoses of muscular dystrophies.

The primary research question was whether machine learning is able to correctly suggest diagnosis of muscular dystrophies based on muscle MRI of the pelvis, thighs, and leg muscles.

The study included MRI studies from the pelvic, thigh, and leg muscles of patients with genetically confirmed diagnosis of muscular dystrophies: dystrophinopathies (both Duchenne and Becker muscular dystrophies, n=46 patients), limb-girdle muscular dystrophy [LGMDR]-1 (CAPN3 gene, n=73), LGMDR2 (DYSF gene, n=181), LGMDR3 to -6 (sarcoglycanopathies, n=69), LGMDR9 (FKRP gene, n = 40), LGMDR12 (ANO5 gene, n=28), muscular dystrophies caused by mutations in the LMNA gene (n=41), facioscapulohumeral muscular dystrophy (FSHD, n=269), oculopharyngeal muscular dystrophy (OPMD, n = 171), and Pompe disease (n=68).

A total of 10 MR images were excluded because they were normal and did not show any fat replacement. Therefore, the study included a total of 976 MRIs.

Of 2000 different models, the best model selected had a diagnostic accuracy of 95.7%, with an overall sensitivity of 92.1%, specificity of 99.4%, positive predictive value of 98.06%, and negative predictive value of 99.53%.

Related Articles

The researchers compared the accuracy of the model to predict the diagnosis of muscle dystrophies with 4 experts in the field using a new set of 20 MRIs and found that the model generated was able to distinguish among disorders with higher accuracy than the experts. The software obtained a final score of 55 of 60 points, whereas the experts obtained 42, 41, 38, and 31 of 60 points, respectively

One of the limitations of the study, according to the researchers, is the inclusion of 10 different muscular dystrophies, but not including other potential diagnoses as a result of the limited numbers of MRIs from these patient groups.

“This study can be considered as a proof of concept that demonstrates that artificial intelligence can be applied to the field of muscle MRI. We need to implement the tool by adding more diseases,” concluded the researchers.

Reference

Verdú-Díaz J, Alonso-Pérez J, Nuñez-Peralta C, et al. Accuracy of a machine learning muscle MRI-based tool for the diagnosis of muscular dystrophies [published online ahead of print, February 6, 2020]. Neurology. doi: 10.1212/WNL.0000000000009068