Novel Approach Combining Shallow Learning and Ensemble Learning for the Automated Detection of Swallowing Sounds in a...

Area of Science:

• Biomedical Engineering
• Speech and Hearing Science
• Medical Acoustics

Background:

• Cervical auscultation is a noninvasive dysphagia diagnostic tool, but its reliability is limited by evaluator subjectivity.
• Existing automatic swallowing sound detection methods lack specialization for clinical conditions.
• Objective and reliable automated dysphagia diagnosis remains an unmet clinical need.

Purpose of the Study:

• To develop and evaluate a novel, objective method for automatically detecting swallowing sounds.
• To enhance the diagnostic accuracy of cervical auscultation for dysphagia.
• To address the limitations of current automated swallowing sound analysis.

Main Methods:

• Feature extraction using Mel Frequency Cepstral Coefficients (MFCCs) and Mel Frequency Magnitude Coefficients (MFMCs).
• Application of an ensemble learning model combining Support Vector Machine (SVM) and Multi-Layer Perceptron (MLP).
• Validation using a synchronized database of swallowing sounds and videofluorographic swallowing studies from 74 patients.

Main Results:

• The proposed method achieved an F1-micro average of approximately 0.92.
• The system demonstrated a high diagnostic accuracy of 95.20%.
• The results indicate superior performance in detecting swallowing sounds compared to existing methods.

Conclusions:

• The novel approach offers a significant advancement in the objectivity of cervical auscultation for dysphagia diagnosis.
• The method shows high efficacy within the tested clinical recording database.
• Further validation across diverse databases is essential to confirm broad applicability and clinical impact.