Dysphagia Discrimination Model Development based on Ultrasonic Doppler Sensor

Abstract

The development of a dysphagia discrimination model using an ultrasonic Doppler sensor aims to address the limitations of traditional diagnostic methods for swallowing disorders. Dysphagia, which encompasses difficulties in moving food from the mouth to the stomach, can arise from neurological disorders, aging, and other health conditions. Prompt and accurate diagnosis is critical, as dysphagia can lead to severe complications such as aspiration pneumonia and malnutrition. This study presents the Swallowing Monitoring and Assessment System (SMAS), a novel, non-invasive device that utilizes ultrasonic Doppler sensors to monitor and assess swallowing function. Traditional diagnostic methods like Videofluoroscopic Swallowing Study (VFSS) and Fiberoptic Endoscopic Evaluation of Swallowing (FEES) pose several challenges, including exposure to radiation, invasiveness, and the difficulty of continuous monitoring in daily life. In contrast, the SMAS device, designed to be worn comfortably over the shoulder, allows for real-time monitoring and assessment of swallowing activity in various settings. The primary objectives of this research are threefold: (1) to perform a quantitative analysis of ultrasonic signals generated during swallowing, (2) to establish the correlation between hyoid bone movement and ultrasonic signals, and (3) to develop a machine learning-based model for dysphagia discrimination. The study involves a series of experiments with both healthy individuals and dysphagic patients, who are asked to swallow liquids and semi-solids of different viscosities and volumes. The ultrasonic signals captured during these swallowing events are processed to extract key features such as peak amplitude, duration, number of peaks, peak-to-peak interval, and energy. Simultaneous measurements using VFSS and the ultrasonic Doppler sensor-based SMAS are conducted to validate the system. The correlation between the movement of the hyoid bone, as observed in VFSS images, and the ultrasonic signals is analyzed to interpret the clinical significance of the signals. The data collected is used to train various machine learning models, including Convolutional Neural Networks (CNNs) and Random Forest classifiers, to distinguish between normal and dysphagic swallowing patterns. The results indicate that the SMAS can reliably capture and analyze swallowing signals, providing a robust basis for the detection and assessment of dysphagia. The machine learning models developed in this study demonstrate high accuracy in classifying swallowing disorders, highlighting the potential of this technology for clinical application. The non-invasive nature of the SMAS, combined with its ability to provide continuous and real-time monitoring, represents a significant advancement over traditional methods. The significance of this study lies in its contribution to the field of dysphagia diagnosis and management. By leveraging the capabilities of ultrasonic Doppler sensors and machine learning, the SMAS offers a safer, more comfortable, and effective solution for both patients and healthcare providers. This research paves the way for the integration of advanced sensor technology in routine clinical practice, potentially improving the quality of care for individuals with swallowing disorders. Further research and development are encouraged to refine the system and expand its applications in diverse clinical settings.