Segmentation of Glottal Images from High-Speed Videoendoscopy Optimized by Synchronous Acoustic Recordings

Area of Science:

• Laryngology
• Biomedical Engineering
• Acoustics

Background:

• Laryngeal high-speed videoendoscopy (LHSV) provides detailed visualization of vocal fold vibratory activity.
• Current image analysis methods often require manual annotation for accurate vocal fold edge detection.
• Automated analysis is needed to improve efficiency and objectivity in voice disorder assessment.

Purpose of the Study:

• To develop and validate a fully automatic method for vocal fold edge detection and glottal area analysis using LHSV and acoustic data.
• To optimize image segmentation by synchronizing video and acoustic signals.
• To assess the potential of computed geometric indices for differentiating normal and pathological voices.

Main Methods:

• Synchronous recording of laryngeal video and acoustic data during sustained vowel phonation.
• Image segmentation algorithm optimized by matching Fourier spectra of video and acoustic signals.
• Analysis of geometric indices, including Open Quotient and Minimal Relative Glottal Area, on normophonic and dysphonic voice recordings.

Main Results:

• The proposed method successfully segments the glottal area by combining video and acoustic data.
• Computed geometric indices effectively discriminate between normal and pathological voices.
• Median Open Quotient was 0.69 (normal) vs. 1 (dysphonic); Minimal Relative Glottal Area was 0.06 (normal) vs. 0.35 (dysphonic).

Conclusions:

• The developed automatic method offers an objective approach to voice disorder assessment using LHSV.
• Geometric indices derived from glottal area analysis are reliable indicators of vocal fold function.
• This technique holds promise for clinical application in phoniatrics and voice research.