Related Experiment Video
Updated: Jul 7, 2025
08:32
Adapting Human Videofluoroscopic Swallow Study Methods to Detect and Characterize Dysphagia in Murine Disease Models
Published on: March 1, 2015
21.3K
Machine-learning assisted swallowing assessment: a deep learning-based quality improvement tool to screen for
Rami Saab1, Arjun Balachandar1, Hamza Mahdi1
1Hurvitz Brain Sciences Program, Division of Neurology, Department of Medicine, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, ON, Canada.
Frontiers in Neuroscience
|December 22, 2023
Summary
Deep learning models show promise for detecting post-stroke dysphagia using voice biomarkers. This automated screening method could improve early detection and patient outcomes for swallowing difficulties after a stroke.
Area of Science:
- Neurology
- Biomedical Engineering
- Speech Pathology
Background:
- Post-stroke dysphagia is a common complication, leading to increased morbidity and mortality.
- Current bedside screening methods for dysphagia can be subjective and may limit patient access.
- Voice changes are recognized as a potential indicator of dysphagia, offering a non-invasive biomarker.
Purpose of the Study:
- To develop and evaluate a proof-of-concept deep learning model for automated dysphagia screening in post-stroke patients.
- To assess the feasibility of using voice recordings as a biomarker for detecting dysphagia.
Main Methods:
- A single-center study involving 68 post-stroke patients (40 training, 28 testing).
- Voice data (vowels, words, sentences) were collected and processed into Mel-spectrogram images.
- Deep learning models (DenseNet, ConvNext) were trained and tested on audio clip data for dysphagia classification.
Main Results:
- Clip-level analysis showed a sensitivity of 71% and specificity of 77% (AUC = 0.80).
- Participant-level analysis achieved higher accuracy with 89% sensitivity and 79% specificity (AUC = 0.91).
Conclusions:
- This study demonstrates the feasibility of using deep learning on vocalizations for post-stroke dysphagia detection.
- The findings suggest a potential for enhancing dysphagia screening through objective, voice-based biomarkers.
