Computer Vision Techniques for Vocal Fold Surgery Assistance
Research Internship (MITACS GRI Award) at the Medical Computer Vision and Robotics Lab, University of Toronto, Onsite: May’23 - Jul’23, Guide: Prof. Lueder Kahrs University of Toronto
Summary: Image quality is important when training neural networks and noisy images can hurt performance. This project involved tracking the motion of vocal folds and processing laryngoscopy images prior to their 3D visualization.
Poster presented at the 145th Annual meet of the American Laryngological Association, COSM 2024, Chicago
- Developed a preprocessing pipeline including specularity removal, colour and illumination correction and image denoising using histogram and GAN-based methods to create a fully annotated laryngoscopic dataset with improved PIQE scores
- Performed automated vocal fold motion tracking using image processing techniques and the AGATI software