Classifying Dysphagic Swallowing Sounds with Support Vector Machines

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning Applications in Healthcare

Background:

• Cervical auscultation of swallowing sounds provides insights into swallowing function.
• Existing research highlights differences in swallowing sound characteristics between healthy and dysphagic individuals.
• The precise classification of dysphagia severity using swallowing sounds requires further investigation.

Purpose of the Study:

• To investigate the application of machine learning for classifying swallowing sounds into distinct dysphagia severity levels.
• To evaluate the effectiveness of Support Vector Machines (SVMs) in categorizing normal, mild, moderate, and severe dysphagia based on acoustic features.

Main Methods:

• Swallowing sounds were recorded from patients diagnosed with dysphagia.
• Feature extraction was performed on the recorded swallowing sounds.
• Support Vector Machines (SVMs) were trained and evaluated using cross-validation techniques for classification.

Main Results:

• In a two-class scenario (normal vs. dysphagic), the maximum F-measure achieved was 78.9%.
• In a four-class scenario (normal, mild, moderate, severe dysphagia), F-measure values were 65.6%, 53.1%, 51.1%, and 37.1%, respectively.
• The study demonstrates the potential of machine learning in differentiating dysphagia severity levels.

Conclusions:

• Machine learning, specifically SVMs, can be utilized to classify swallowing sounds for dysphagia assessment.
• The classification accuracy varies with the number of dysphagia severity classes considered.
• Further research is warranted to enhance the accuracy and clinical applicability of this non-invasive screening method.