Automatic Estimation of Laryngeal Vestibule Closure Duration Using High- Resolution Cervical Auscultation Signals

Area of Science:

• Swallowing physiology
• Biomedical engineering
• Medical diagnostics

Background:

• Safe swallowing relies on laryngeal vestibule closure (LVC) to prevent aspiration.
• Current assessment methods like videofluoroscopic swallowing studies involve radiation and are not universally accessible.
• There is a need for noninvasive techniques to evaluate swallow function.

Purpose of the Study:

• To investigate if a noninvasive sensor-based system can predict laryngeal vestibule status in real-time.
• To determine if the system can accurately estimate laryngeal vestibule closure duration.
• To compare the system's accuracy to trained human raters.

Main Methods:

• Utilized high-resolution cervical auscultation (HRCA), a noninvasive sensor-based system.
• Employed advanced machine learning and deep learning models for HRCA signal analysis.
• Trained the model on 588 swallows from patients and tested on 45 from healthy participants.

Main Results:

• The HRCA system achieved 74.90% and 75.48% accuracy in distinguishing LVC status across datasets.
• Automated LVC duration estimates showed high correlation with human judgments (ratios of 1.13 and 0.93).

Conclusions:

• HRCA signal analysis with machine learning can effectively predict laryngeal vestibule status and estimate closure duration.
• HRCA offers a promising noninvasive tool for diagnosing swallowing disorders and for biofeedback, avoiding X-ray exposure.