Related Experiment Video
Updated: Oct 28, 2025
08:32
Adapting Human Videofluoroscopic Swallow Study Methods to Detect and Characterize Dysphagia in Murine Disease Models
Published on: March 1, 2015
21.6K
Machine learning based analysis of speech dimensions in functional oropharyngeal dysphagia
Sebastian Roldan-Vasco1, Andres Orozco-Duque2, Juan Camilo Suarez-Escudero3
1Faculty of Engineering, Instituto Tecnológico Metropolitano, Medellín, Colombia; Faculty of Engineering, Universidad de Antioquia, Medellín, Colombia.
Computer Methods and Programs in Biomedicine
|July 14, 2021
Summary
This study shows that analyzing speech patterns can effectively screen for functional oropharyngeal dysphagia. Machine learning models accurately distinguish between patients with dysphagia and healthy individuals, offering a non-invasive screening method.
Area of Science:
- Neurology
- Speech Science
- Machine Learning
Background:
- Swallowing (deglutition) involves complex neurological control and anatomical coordination.
- Dysphagia, or swallowing dysfunction, arises from impaired coordination and current screening methods are subjective.
- Speech and swallowing share neural pathways and anatomical structures, suggesting potential for speech analysis in dysphagia assessment.
Purpose of the Study:
- To evaluate the efficacy of automatic speech processing and machine learning for screening functional oropharyngeal dysphagia.
- To determine if speech characteristics can serve as reliable biomarkers for dysphagia.
- To explore a non-invasive and objective method for dysphagia detection.
Main Methods:
- Collected speech recordings from 46 patients with neurological dysphagia and 46 healthy controls.
- Analyzed speech dimensions: phonation, articulation, and prosody, extracting specific acoustic features.
- Applied machine learning models (e.g., Random Forest) with nested cross-validation, optimizing with AUC-ROC.
Main Results:
- The Random Forest model, focusing on articulation features, achieved high performance (AUC=0.86±0.10, sensitivity=0.91±0.12).
- Combining multiple speech dimensions in a voting ensemble further improved discrimination accuracy.
- Results indicate that speech signal features provide valuable information for identifying dysphagia.
Conclusions:
- Speech-based machine learning models demonstrate suitability for automatically differentiating individuals with and without dysphagia.
- This approach offers a promising, non-invasive, and cost-effective method for screening functional oropharyngeal dysphagia.
- Findings support the use of speech analysis as a valuable tool in neurological dysphagia assessment.
