Uncertainty of Spatial Segmentation of High-Speed Videoendoscopy and Its Temporal and Spatial Dependency

Area of Science:

• Laryngology and Voice Science
• Medical Imaging Analysis
• Biomechanical Engineering

Background:

• Spatial segmentation of vocal fold edges in high-speed videoendoscopy (HSV) is crucial for accurate voice analysis.
• Segmentation uncertainty directly influences the reliability of derived vocal fold measures.
• Understanding factors affecting segmentation uncertainty is key to improving diagnostic accuracy.

Purpose of the Study:

• To quantify spatial segmentation uncertainty in HSV recordings.
• To investigate the dependency of this uncertainty on glottal cycle phase and anterior-posterior location.
• To provide insights for improving automated HSV analysis pipelines.

Main Methods:

• Three experts performed manual segmentation of vocal fold edges on 12 HSV recordings.
• An iterative process with blinded reconciliation was used to refine segmentations.
• Segmentation uncertainty was calculated as the pixel distance between expert markings.

Main Results:

• Segmentation uncertainty was significantly higher in the anterior and posterior glottis compared to the middle section.
• Anterior glottis uncertainty was 40% greater than the middle section.
• Higher uncertainty was observed during more open glottal configurations (31% increase) and correlated with normalized glottal area.

Conclusions:

• Vocal fold edge segmentation uncertainty is influenced by both glottal cycle phase and anterior-posterior location.
• This variability necessitates careful consideration of uncertainty levels for different HSV measures.
• Findings can inform the development of more robust automated segmentation algorithms for HSV data.