Swallowing function is a complicated process involving several highly coordinated events. In recent years, acoustical analysis of the swallowing mechanism has received considerable attention due to its diagnostic potentials. This thesis presents novel approaches for analysis and modeling of swallowing sounds in individuals with and without swallowing disorder (dysphagia).
Different techniques based on nonlinear dynamics, recuffence quantification analysis, hidden Markov model (HMM), and multiresolution wavelet analysis were used to detect characteristic features of swallowing sounds for automatic swallowing sound detection, swallowing sound segmentation, as well as classification of normal and dysphagic swallowing sounds. Data from 27 healthy subjects and 11 dysphagic patients were used, in which swallowing and breath sounds were recorded with an accelerometer placed over the trachea (suprasternal notch), while the subjects were fed three different textures with a bolus size of 5 ml. Submental electromyogram signals were also recorded in 12 normal subjects simultaneously with their swallowing sounds. In terms of swallowing sound detection in the swallow and breath sound recordings, the performance of the HMM based method using recurrence plot features was superior to that of the other methods. As for the segmentation of the swallowing sounds, multiscale products of wavelet coefficients along with an HMM yielded the least error in detecting the boundaries between the segments. Lastly, in terms of classification of normal and dysphagic
swallowing sounds, nonlinear metric tools, i.e., correlation dimension and time delay, resulted in a high accuracy of 837o. The results of HMM based methods for classification of swallowing sounds between the two groups of healthy and dysphagic subjects using RMS achieved 85.5Vo accuracy. The results of the study on the timing of submental muscle contraction in relation to swallowing sounds showed that the movement of the larynx precedes the fust audible segment of swallowing sounds. It also showed that the lag between the onset of submental muscle contraction and the beginning of initial discrete sounds is shorter for thin liquid texture.
Overall, the outcomes of this study have paved the way for a better understanding of the swallowing mechanism and improving clinical assessment techniques of swallowing disorders using its acoustic signatures.
