Related Experiment Video
Updated: Jul 11, 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
Use of deep learning to segment bolus during videofluoroscopic swallow studies
Nadeem Shaheen1, Ryan Burdick2,3, Rodolfo Peña-Chávez2,4
1Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, United States of America.
Biomedical Physics & Engineering Express
|November 10, 2023
Summary
Artificial intelligence (AI) bolus segmentation shows promise for improving video fluoroscopic swallow study (VFS) analysis. However, AI performance degrades in the oral cavity due to anatomical interference, requiring further network optimization.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in healthcare
- Swallowing disorders diagnostics
Background:
- Artificial intelligence (AI) segmentation offers potential to enhance video fluoroscopic swallow study (VFS) analysis by enabling objective metric determination.
- Understanding AI limitations is crucial for reliable clinical application in VFS.
Purpose of the Study:
- To evaluate the performance of an AI bolus segmentation network for VFS analysis.
- To identify challenges and limitations affecting AI segmentation accuracy in VFS.
Main Methods:
- Manual contouring of thin/liquid bolus from oral cavity to swallow end in 80 patients.
- Training and validation using a 75/25 data split with 4-fold cross-validation.
- U-Net architecture tested with Dice coefficient as loss and performance metric.
Main Results:
- Average Dice coefficient of 0.67 on the validation set.
- AI performance degraded in the oral cavity due to misclassification of teeth and residue.
- High variability observed in network training processes and performance across different swallow phases.
Conclusions:
- AI bolus segmentation is effective but faces challenges, particularly within the oral cavity.
- Network performance is influenced by anatomical structures and bolus size, necessitating targeted improvements.
- Further research is needed to optimize AI models for robust VFS analysis.
