Multi-muscle Texture Analysis for Dystrophy Development Identification in Golden Retriever Muscular Dystrophy Dogs

摘要

The study assesses the suitability of multi-muscle texture analysis (TA) for the dystrophy development characterization in Golden Retriever Muscular Dystrophy (GRMD) dogs. Textural features, statistical and model-based, are derived from T2-weighted Magnetic Resonance Images (MRI) of canine hindlimb muscles. Features obtained from different types of muscles (EDL, GasLat, GasMed, and TC) are analyzed simultaneously. Four phases of dystrophy progression, including the 'zero phase' - the absence of the disease, are differentiated. Two classifiers are applied: Support Vector Machines (SVM) and Adaptive Boosting (AdaBoost). A Monte Carlo-based feature selection enables to find features (and the corresponding muscle types) that are the most useful in identifying the phase of dystrophy. The simultaneous consideration of several muscles improves the classification accuracy by maximum 12.5% in comparison to the best corresponding result achieved with single-muscle TA. A combination of 17 textural features derived from different types of muscles provides a classification accuracy of approximately 82%.