Advancements in Cochlear Imaging: Employing GANs for Generating Hair Cell Images and YOLOv12 for Detection of Inner and Outer Hair Cells
Loading...
Authors
Krudwig, Cole
Issue Date
2025-12
Volume
10
Issue
Type
Journal Article
Language
Keywords
Alternative Title
Abstract
Hearing loss is a significant global health challenge impacting over 1.5 billion people. Among these, 430 million individuals currently suffer from disabling hearing loss, with projections estimating this number will rise to 700 million by 2050 (Pan American Health Organization, n.d.). The repercussions of hearing loss are particularly profound in education and employment. According to the World Health Organization (WHO, 2024), hearing loss results in an annual global economic cost of approximately $980 billion. These costs include hearing devices, educational support, lost productivity, and broader societal impacts. Hearing health research has traditionally relied on manually counting auditory hair cells (HCs) in the cochlea—a process that is laborious, time-consuming, and expensive. Recognizing the need for a more efficient approach, we have developed a deeplearning solution that employs generative adversarial networks (GANs) and a YOLOv12 object detection model to automate cochlear sample collection and hair cell counting. This innovative technique not only expedites data collection and analysis but also significantly reduces time, effort, and labor costs. Our approach uses advanced deep-learning algorithms to detect inner and outer HCs from mouse inner ear images and to segment hair cells for further analysis. Our findings indicate that synthetically generated cochlear samples are effective in training an object detection model for automated hair cell counting. Furthermore, once the deep-learning model is trained, testing on new cochlear images can be completed in seconds.
Description
Citation
Publisher
Creighton University
License
This material is copyrighted