Duchenne muscular dystrophy detection using EMG analysis and machine learning

Muscular disease such as Duchenne Muscular Dystrophy (DMD) are myogenic disorders inherited in a genetic manner are characterized by progressive muscle weakness and wasting that typically begins in early childhood. Their identification has hitherto relied mostly on electrophysiological measures of EMG signals and seldom DMD data. Our approach consists of using healthy and DMD patients' data collected using EMG electrodes to extract 22 temporal and spectral features. Then we feed those features to 10 machine learning models after going through two preprocessing steps, namely the outlier's removal and feature selection using sequential feature selection method (SFS). The results are highly encouraging since the classification rates of all the classifiers are above 80% except Support Vector, Extreme Gradient Boost and LDA classifiers that have classification rate between 70% and 80%. These classification performances were yielded using the following features that was selected by the SFS method: SM2, MAV1, RMS, AAC, DASDV, MYOP, MAVSLPk2, MAVSLPk3.