Team Members: Dr. Corey Clark, Myque Ouellette
This research explores an innovative approach to retinal image analysis using deep neural networks (DNNs) enhanced with human computation game (HCG) inputs. The study focuses on improving the accuracy of retinal layer segmentation, particularly for the retinal pigment epithelium (RPE) and outer segment/photoreceptor layers, in optical coherence tomography (OCT) scans.
The key innovation lies in leveraging human pattern recognition skills through gamification to guide machine learning algorithms, allowing for accurate segmentation with significantly smaller datasets than traditional methods. The research demonstrates that HCG-enhanced DNNs can achieve superior performance compared to standard DNNs, even with datasets of fewer than 350 images. This approach shows promise in adapting to both low and high-resolution OCT images and in handling complex, mixed pathologies such as drusen, double layer sign, and geographic atrophy. By combining machine learning with human-guided inputs, the study aims to create more efficient and adaptable AI tools for early detection and monitoring of retinal diseases, potentially accelerating the adoption of advanced imaging technologies in clinical settings.
