Automated Algorithm for Swallowing Sound Detection

Abstract

This study examines the automated detection of swallowing sounds for normal subjects. A normal swallowing sound is characterized by three phases including oral, pharyngeal and esophageal where complications lead to swallowing disorder or dysphagia. Current gold standard testing for this abnormality is videofluorography, an x-ray based procedure with detrimental radiation side effects. New non-invasive techniques are necessarily explored to help assess the performance of the swallowing mechanism. Recent developed studies in acoustical airflow estimation indicate the need to detect and extract swallowing segments from sound signals. Extraction is currently a manual process, both subjective and time-consuming. Thus, an automated, objective and quick method is developed in the form of a “smart” algorithm with the ability to make decisions like trained technicians and physicians. Three sound signal features were explored to assist in the classification process (AR- coefficients, RMS values and average power). Utilizing the features, classification sequences were produced for six healthy subjects swallowing sounds. The results were compared with known values (acquired through visual and auditory means). RMS features in combination with the “smart” code yield the lowest error, on average 20.7 ± 4.6%. Future studies include testing variations of smart algorithm code in order to create a robust algorithm. Also, future work includes varying test subject ages, test media (bolus textures) and creating a program-user interface for decision-making assistance.

Keywords: swallowing, dysphagia, spectrogram, RMS, automated detection.