Autonomous Swallow Detection in VFSS Using Manometric Data in Dysphagia Patients

Overview

This study presents a novel autonomous framework combining video-fluoroscopic swallow study (VFSS) and high-resolution impedance manometry (HRIM) data to accurately detect swallow events in head and neck cancer (HNC) patients with oropharyngeal dysphagia (OD). The approach reduces clinical workload by automating swallow detection and classification, improving diagnostic accuracy and consistency.

Background

Swallowing is a complex physiological process involving coordinated muscle contractions across oral, pharyngeal, and esophageal phases. Dysphagia, common in HNC patients, leads to serious complications such as malnutrition and aspiration pneumonia. VFSS is a standard diagnostic tool but requires expert interpretation and is prone to variability. HRIM provides quantitative pressure data but is complex to analyze, especially in patients with altered anatomy. Combining VFSS and HRIM can enhance assessment but increases clinical workload and complexity.

Data Highlights

The framework utilizes an optimized double-sweep optical flow algorithm to detect swallow candidates in VFSS videos. A pressure-based swallow template derived from pre-annotated HRIM data classifies these candidates, enabling two-way verification. This method recovers missed manometric data and isolates swallows present in VFSS streams. Validation was performed on data from HNC patients, a population with high swallow abnormality incidence.

Key Findings

• Swallow candidates in VFSS were robustly detected using an optimized double-sweep optical flow algorithm.
• A pressure-based swallow template from HRIM data enabled accurate classification and cross-verification of swallow events.
• The framework recovered relevant manometric data missed during live HRIM annotations, enhancing data completeness.
• Automation significantly reduced clinician workload by minimizing manual annotation and supervision.
• The approach improved consistency and objectivity in swallow assessment in a complex HNC patient population.

Clinical Implications

This autonomous framework facilitates more accurate and efficient swallow event detection in patients with oropharyngeal dysphagia, particularly those with head and neck cancer. By reducing manual input and inter-rater variability, it supports personalized treatment planning and may improve patient outcomes. Integration of this technology could streamline clinical workflows and enhance diagnostic reliability.

Conclusion

The proposed autonomous swallow detection framework effectively integrates VFSS and HRIM data to improve diagnostic accuracy and reduce clinical workload in dysphagia assessment. This represents a significant advancement toward fully automated, objective evaluation of swallowing function in complex patient populations.

References

1. International Dysphagia Diet Standardisation Initiative (IDDSI) 2016 -- Framework and Terminology
2. Farnebäck 2003 -- Two-Frame Motion Estimation Based on Polynomial Expansion
3. Medtronic® ManoScan™ and Laborie® Solar GI™ -- HRIM Analysis Software