Related Experiment Video
Updated: Jun 21, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
Swallow segmentation with artificial neural networks and multi-sensor fusion
Joon Lee1, Catriona M Steele, Tom Chau
1Bloorview Research Institute, Toronto, Canada.
Medical Engineering & Physics
|August 4, 2009
Summary
This study shows that using multiple sensors like accelerometry, MMG, and nasal airflow with artificial neural networks (ANN) improves swallow segmentation accuracy. Combining all four signals yielded the best results for analyzing swallowing signals.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Swallowing Disorders Research
Background:
- Accurate swallow segmentation is crucial for analyzing swallowing function.
- Current methods often rely on manual segmentation, which is time-consuming and subjective.
- Automated segmentation using physiological signals can enhance diagnostic capabilities.
Purpose of the Study:
- To investigate the impact of incorporating multiple physiological signal sources on automated swallow segmentation performance using artificial neural networks (ANN).
- To identify the most effective signal or combination of signals for accurate swallow segmentation.
- To evaluate the relationship between the number of signal sources and segmentation accuracy.
Main Methods:
- Acquired cervical dual-axis accelerometry, submental mechanomyography (MMG), and nasal airflow signals from 17 healthy adults during various swallowing tasks.
- Constructed feature vectors using signal variances within moving windows for single and multiple signal combinations.
- Trained and tested a 3-layer ANN for each participant-signal combination to classify swallow events.
Main Results:
- Segmentation performance, including sensitivity, specificity, and accuracy, improved as more signal sources were integrated.
- The combination of all four signal sources (accelerometry, MMG, nasal airflow) achieved the highest mean accuracy (88.5%) and adjusted accuracy (89.6%).
- Anterior-posterior accelerometry was the most discriminatory signal source; MMG and nasal airflow provided the least performance improvement.
Conclusions:
- Multi-sensor fusion approaches utilizing ANNs significantly enhance automated swallow segmentation.
- Integrating diverse physiological signals offers a promising avenue for improving the objective analysis of swallowing function.
- Further research into ANN-based, multi-sensor fusion techniques is warranted for clinical applications in swallowing studies.