Non-Invasive Monitoring of Swallowing Function Using Multi-Channel Auscultation

• Published in: Proceedings of IEEE Sensors ... pp. 1 - 4
• Main Authors: Yin, Ziyuan, Wong, Marc T. L., Lan, Chuanlin, Wong, Sunny H., Chan, Rosa H. M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 29.10.2023
• Subjects: Acoustics; Computational modeling; Convolutional neural networks; ensemble modeling; Robustness; Sensors; sound; Sternum; Stethoscope; swallowing
• ISSN: 2168-9229
• DOI: 10.1109/SENSORS56945.2023.10325175

This study investigates a non-invasive and cost-effective solution for monitoring swallowing function by employing off-the-shelf microphones or digital stethoscopes, which could greatly benefit patients with dysphagia or related disorders and improve their care management. Our results demonstrated that superficial sounds recorded at the levels of thyroid cartilage, cricoid cartilage, sternum, and gastric cardia effectively reflected swallowing of three ingestibles with varying consistencies. Our results showed that one acoustic input was insufficient for identifying different ingestible consistencies. By employing an ensemble convolutional neural network with the spectra of all four sound channels as inputs, we successfully classified swallowing sounds associated with three distinct ingestible consistencies, achieving an average accuracy of 0.68 for cross-subject generalized model and 0.87 for cross-session personalized model. However, using superficial sounds recorded around thyroid cartilage level and sternum already offered comparable leave-one-out classification accuracy. This approach showcases the potential of using microphones for the clinical development of long-term bedside monitor for swallowing.