Related Experiment Video
Updated: Mar 31, 2026
Adapting Human Videofluoroscopic Swallow Study Methods to Detect and Characterize Dysphagia in Murine Disease Models
Published on: March 1, 2015
Assessing the Performance and Reliability of Deep Learning Autosegmentation in Videofluoroscopic Swallowing Studies:
Wei-Kai Chuang1, Bing-Fong Lin2, Yu-Hao Lee3
1Department of Radiation Oncology, Shuang Ho Hospital, Taipei Medical University, New Taipei City, Taiwan; Department of Biomedical Imaging and Radiological Sciences, National Yang Ming Chiao Tung University, Taipei, Taiwan; Department of Radiation Oncology, Saint Paul's Hospital, Taoyuan, Taiwan.
Deep learning auto-segmentation in videofluoroscopic swallowing studies (VFSS) shows high accuracy for anatomical structures like the bolus and cervical spine. Further standardization is needed to improve clinical use.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Swallowing Disorders Research
Background:
- Videofluoroscopic swallowing studies (VFSS) are crucial for diagnosing dysphagia.
- Manual segmentation of anatomical structures in VFSS is time-consuming and subjective.
- Deep learning (DL) offers potential for automated segmentation, improving efficiency and consistency.
Purpose of the Study:
- To systematically evaluate the accuracy and reliability of DL-based auto-segmentation methods in VFSS.
- To perform a meta-analysis of existing studies on DL auto-segmentation in VFSS.
- To identify performance variations across different anatomical targets and methodologies.
Main Methods:
- A comprehensive literature search was performed across major scientific databases (PubMed, IEEE Xplore, Embase, Web of Science, Cochrane Library) from 2013 to 2024.
- Studies utilizing DL for auto-segmentation of bolus, cervical spine, hyoid bone, or thyroid cartilage-vocal fold complex (TVC) in VFSS were included.
- Data extraction and quality assessment were conducted by two independent reviewers using CLAIM and QUADAS-2 tools.
Main Results:
- Ten studies met the inclusion criteria for the meta-analysis.
- The overall pooled Dice similarity coefficient (DSC) for DL auto-segmentation was 0.83 (95% CI: 0.76-0.88), indicating high accuracy.
- Subgroup analyses showed comparable performance for bolus (DSC=0.84) and cervical spine (DSC=0.83) segmentation, despite substantial heterogeneity (I² > 74%).
Conclusions:
- DL-based auto-segmentation in VFSS demonstrates significant promise for accurately segmenting various anatomical structures.
- Observed methodological variability highlights the need for standardized protocols and multi-center datasets.
- Further research comparing DL model architectures is essential to enhance generalizability and clinical applicability of automated VFSS analysis.
