Automatic detection of oral and pharyngeal phases in swallowing using classification algorithms and multichannel EMG.

Swallowing is a complex process that involves sequential voluntary and involuntary muscle contractions. Malfunctioning of swallowing related muscles could lead to dysphagia. However, there is a lack of standardized and non-invasive methods that support and improve the diagnosis and ambulatory care. This paper presents a classification scheme of two swallowing phases (oral and pharyngeal) based on signals of surface electromyography (sEMG). Eight acquisition channels recorded the EMG activity of 47 healthy subjects while they swallowed water, yogurt and saliva. Every signal was processed, segmented and labeled with background activity, oral or pharyngeal classes. Nine time domain and four frequency domain features were extracted from the segments, assessed individually and then compared in groups according to a correlation analysis. A support vector machine (SVM) with radial basis function kernel and a feedforward artificial neural network (ANN) with one hidden layer were used as classifiers. Different hyperparameters of the SVM and number of hidden neurons of the ANN were assessed for the proposed scheme. The recognition accuracy of SVM (92,03%) was higher than ANN's (90,26%). Time domain features were found to have better capability of representation than their frequency domain counterpart. Nevertheless, expanding the feature space improved the performance of the classifiers. Experimental results show that proposed sEMG-based method can correctly distinguish between oral and pharyngeal swallowing phases and can be used for assessment of continuous swallowing tasks. This paper extends previous reported findings to small muscles with low signal-to-noise ratio and high crosstalk acquired in multichannel systems.