Detection and Mosaicing Techniques for Low-Quality Retinal Videos

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Specialized equipment for retinal fundus imaging is expensive and lacks portability.
• Smartphone-based devices like D-Eye offer a more accessible and cost-effective solution for eye disease screening.

Purpose of the Study:

• To evaluate the effectiveness of smartphone-based devices for pre-screening eye diseases.
• To develop and compare methods for extracting relevant retinal information from lower-quality video captured by these devices.

Main Methods:

• Proposed two methods for retinal zone extraction: classical image processing (thresholds, Hough Circle transform) and a YOLO v4 neural network.
• Implemented a mosaicing technique using GLAMpoints and homography transformations to create a single, higher field-of-view image from multiple retinal regions.

Main Results:

• The YOLO v4 neural network demonstrated superior performance for retinal region extraction compared to classical methods.
• The mosaicing technique successfully combined extracted regions into a more informative image with an expanded field of view.

Conclusions:

• Smartphone-based D-Eye devices, despite lower image quality than specialized equipment, are suitable for medical pre-screening of eye diseases.
• The developed YOLO v4 and mosaicing methods enhance the utility of these portable devices for wider accessibility to eye care.