Automatic workflow for narrow-band laryngeal video stitching

Area of Science:

• Medical imaging
• Otolaryngology
• Computational pathology

Background:

• Narrow band (NB) laryngeal endoscopy is crucial for detecting laryngeal malignancies by examining vascular patterns.
• Manual review of endoscopic videos is time-consuming due to non-informative frames (blurred, noisy, redundant).
• Accurate diagnosis relies on clear visualization of laryngeal structures and vascularity.

Purpose of the Study:

• To develop an automatic frame selection strategy for NB laryngeal endoscopy.
• To generate a unique larynx panorama by discarding non-informative frames.
• To enhance diagnostic efficiency and accuracy in laryngeal cancer detection.

Main Methods:

• Anisotropic diffusion filtering was used to reduce noise and enhance image features.
• A feature-based stitching approach was employed to create a comprehensive larynx panorama.
• An automatic frame selection strategy was developed and tested on pathological NB endoscopic videos.

Main Results:

• The proposed automatic frame selection strategy outperformed standard methods (uniform, random sampling).
• Visual assessment showed superior preservation of vascular structures in the generated panoramas.
• Numerical evaluation using a blur estimation metric confirmed the effectiveness of the method.

Conclusions:

• The automatic frame selection strategy significantly enhances the quality of NB laryngeal endoscopy.
• This method provides clinicians with a unique larynx panorama, improving diagnostic capabilities.
• The approach offers a more efficient and visually informative alternative to standard video review for laryngeal malignancy diagnosis.