Texture-based identification of dystrophy phase. Indicating the most suitable features for therapy testing

In this study, texture analysis (TA) is applied for characterization of dystrophic muscles visualized on T2-weighted Magnetic Resonance (MR) images. The study proposes a strategy for indicating the textural features that are the most appropriate for testing the therapies of Duchenne muscular dystrophy (DMD). The strategy considers that muscle texture evolves not only along with the disease progression but also with the individual’s development. First, a Monte Carlo (MC) procedure is used to assess the relative importance of each feature in identifying the phases of growth in healthy controls. The features considered as age-dependent at a given acceptance threshold are excluded from further analyses. It is assumed that their application in therapies’ evaluation may entail an incorrect assessment of dystrophy response to treatment. Next, the remaining features are used in differentiation among dystrophy phases. At this step, an MC-based feature selection is applied to find an optimal subset of features. Experiments are repeated at several acceptance thresholds for age-dependent features. Different solutions are finally compared with two classifiers: Neural Network (NN) and Support Vector Machines (SVM). The study is based on the Golden Retriever Muscular Dystrophy (GRMD) model. In total, 39 features provided by 8 TA methods (statistical, filter- and model-based) are tested.