Automatic discrimination between safe and unsafe swallowing using a reputation-based classifier.

Area of Science:

• Biomedical Engineering
• Clinical Informatics
• Gerontology

Background:

• Swallowing accelerometry offers a non-invasive method for bedside dysphagia screening.
• Previous research primarily used single-axis vibration, overlooking additional data from a second axis.
• Limited studies have explored automatic classification of swallowing accelerometry in adult populations.

Purpose of the Study:

• To investigate the efficacy of dual-axis accelerometry for automatic classification of safe versus unsafe swallows.
• To develop and evaluate a reputation-based classifier for swallowing accelerometry data.
• To assess the potential of this method for point-of-care dysphagia assessment.

Main Methods:

• Collected dual-axis accelerometric signals from 30 older adults suspected of dysphagia.
• Utilized videofluoroscopic examination to label 224 swallowing samples as safe (60) or unsafe (164).
• Employed a reputation-based classifier combining three Support Vector Machine (SVM) models and eight distinct features.

Main Results:

• The reputation-based algorithm achieved 80.48% accuracy, 97.1% sensitivity, and 64% specificity in distinguishing safe from unsafe swallows.
• Unsafe swallows exhibited lower mean vibration amplitude and faster autocorrelation decay, indicating reduced hyoid excursion and coordination.
• The reputation-based algorithm outperformed a democratic majority voting algorithm.

Conclusions:

• Dual-axis accelerometry, classified using a reputation-based algorithm, shows promise for dysphagia screening.
• The method's computational efficiency and high sensitivity support its consideration for point-of-care swallow assessment.
• This approach can enhance clinical informatics for swallowing disorders.