Feasibility of Real-Time Automated Vocal Fold Motion Tracking for In-Office Laryngoscopy.

ABSTRACT:

Objectives

Major advancements have been made in applying artificial intelligence and computer vision to analyze videolaryngoscopy data. These models are limited to post hoc analysis and are aimed at research settings. In this work, we assess the feasibility of a real-time solution for automated vocal fold tracking during in-office laryngoscopy.

Methods

We trained a keypoint detection model that tracks 39 individual keypoints of the larynx, optimized for real-time deployment, on 1254 frames extracted from 57 laryngoscopy videos. Images were annotated with keypoints along the true and false vocal folds, and supraglottic structures. After training, the model was evaluated on a validation dataset of 140 test images, and second independent test dataset of 50 images. Performance w

SUBMITTER: Koivu A 

PROVIDER: S-EPMC12718639 | biostudies-literature | 2026 Feb

REPOSITORIES: biostudies-literature