Real-time OCT image denoising using a self-fusion neural network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is crucial for ophthalmic diagnostics but suffers from variable image quality and motion artifacts.
• Current methods like frame-averaging improve image quality but can degrade resolution and increase acquisition time, especially with patient movement.

Purpose of the Study:

• To develop a real-time OCT denoising method using a convolutional neural network (CNN) to overcome the computational cost of self-fusion.
• To enhance OCT image quality, signal-to-noise ratio (SNR), and robustness to motion artifacts.

Main Methods:

• Implemented a convolutional neural network (CNN) for real-time OCT denoising, pretrained for fusing adjacent frames using a self-fusion technique.
• The network was optimized to achieve near video-rate frame rates for efficient image processing.

Main Results:

• The self-fusion network significantly improved peak SNR compared to raw and frame-averaged OCT B-scans.
• The CNN-based self-fusion demonstrated robustness to motion artifacts, outperforming traditional frame-averaging.
• Achieved near video-rate processing, enabling real-time image enhancement.

Conclusions:

• Real-time self-fusion using a CNN provides a fast and robust alternative for OCT image denoising, surpassing frame-averaging.
• This advancement improves the localization of OCT field-of-view and enhances sensitivity for detecting anatomical disease features.
• Enables higher quality OCT imaging for improved ophthalmic diagnostic accuracy.