Multi-muscle MRI Texture Analysis for Therapy Evaluation in Duchenne Muscular Dystrophy

摘要

The study presents a strategy for indicating the textural features that are the most appropriate for therapy evaluation in Duchenne Muscular Dystrophy (DMD). The strategy is based on 'multi-muscle' texture analysis (simultaneously processing several distinct muscles) and involves applying statistical tests to pre-eliminate features that may possibly evolve along with the individual's growth. The remaining features, considered as age-independent, are ranked using the Monte Carlo selection procedure, from the most to the least useful in identifying dystrophy phase. In total 124 features obtained with six texture analysis methods are investigated. Various subsets of the top-ranked age-independent features are assessed by six classifiers. Three binary differentiation problems are posed: the first vs. the second, the second vs. the third, and the first vs. the third dystrophy phase. The best vectors of age-independent features provide a classification accuracy of 100.0%, 86.9%, and 100.0%, respectively, and comprise 16, 12, and 9 features, respectively.