Abstract
The objective of this paper is to propose a new method for classifying the Electromyography (EMG) signal in discriminating among neurogenic and myopathic for patients that have Spinal Cord Injuries (SCI). The method can have possible connections with physical processes and the main assumption in this approach is that the different forms of neurologic conditions have different information encoded in EMG, possibly quantifiable by EMG complexity measures. This approach is the main contribution in this paper and the results can open new possibilities to investigate the presumable connection of SCI with chaotic components in EMG signals as descriptors for neurologic type of diseases. In comparison with other methods (machine learning and deep learning), the main objective of this paper is to extract knowledge with less effort, that is, to use selection of features and neural network classifiers in a simpler manner with acceptable results. In a prospective study, the three types of EMG were used as time series in order to calculate the complexity of signals (normal, myopathic and Amyotrophic lateral sclerosis (ALS) as neuropathic type, because used database has no neuropathic EMG records as result of patients’ SCI). The combinations of the most frequent measures of complexity (for biomedical signals) were used for the classification (Approximate Entropy, Sample Entropy, Reyni Entropy, Lempel Ziv Entropy and T-complexity). The classification reached 100% for both training case and discrimination of myopathy and 0.8663 accuracy for the three classes. These results suggest a possibility that the complexity can be a good indicator of type of disease using EMG with perspective study in EMG. The obtained results were in accordance with results from literature obtained by Machine Learning and Deep Learning. The prediction of SCI’s evolution in time can be also investigated using coefficients of complexity, but the preliminary results showed that, due to large variability of the individual, the mathematical model is nonlinear and its analytical formula is difficult to be guessed in this stage of research. However, the classification using a simple multilayer perceptron (MLP) and Extreme Learning Machine (ELM) gives satisfactory results, comparable with the ones published in literature using Deep Learning but in a much simple manner.