LEARNING TO CORRECT AXIAL MOTION IN OCT FOR 3D RETINAL IMAGING

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) provides high-resolution 3D imaging of biological tissues, significantly advancing retinal imaging.
• Involuntary eye movements during OCT scans introduce motion artifacts, degrading image quality and hindering accurate diagnosis.
• Existing methods for motion correction in OCT are often limited in their effectiveness, especially with significant eye movements.

Purpose of the Study:

• To develop and evaluate a novel convolutional neural network (CNN) for correcting axial motion artifacts in OCT scans.
• To assess the proposed CNN's ability to correct substantial motion while preserving retinal structure.
• To compare the performance of the CNN-based method against conventional techniques in both healthy and diseased retinas.

Main Methods:

• A single volumetric OCT scan is used as input for the proposed convolutional neural network.
• The CNN is trained to learn and correct axial motion artifacts inherent in OCT data.
• The method's performance is evaluated based on visual quality and quantitative error metrics.

Main Results:

• The proposed CNN effectively corrects significant axial motion artifacts in OCT images.
• The method successfully preserves the overall retinal curvature, maintaining anatomical integrity.
• Experimental results demonstrate substantial improvements in visual quality and reduced overall error compared to conventional methods.
• The CNN shows efficacy in correcting motion artifacts in both normal and disease-affected retinal scans.

Conclusions:

• The developed convolutional neural network offers a robust solution for correcting motion artifacts in OCT imaging.
• This AI-driven approach enhances the reliability and diagnostic value of OCT scans.
• The method holds promise for improving clinical OCT analysis and patient outcomes.