Reducing speckle in anterior segment optical coherence tomography images based on a convolutional neural network

Area of Science:

• Medical Imaging
• Biomedical Optics
• Artificial Intelligence in Healthcare

Background:

• Speckle noise significantly degrades the quality of anterior segment optical coherence tomography (OCT) images.
• This noise obscures crucial structural details, hindering accurate diagnostic interpretation.
• Existing denoising methods often struggle to balance noise reduction with preservation of image fidelity.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based algorithm for speckle noise reduction in anterior segment OCT images.
• To improve image quality by effectively removing speckle noise while preserving essential anatomical information.
• To assess the algorithm's performance against state-of-the-art methods in terms of denoising efficacy, detail retention, and computational efficiency.

Main Methods:

• A convolutional neural network (CNN) model was designed to learn speckle noise distribution from a custom OCT dataset.
• The algorithm indirectly generates denoised images by learning noise characteristics, rather than direct image filtering.
• Performance was evaluated through qualitative (visual) and quantitative (parameter-based) assessments, alongside running time analysis.

Main Results:

• The proposed CNN algorithm demonstrated significant speckle noise reduction across various anterior segment OCT images.
• The method exhibited strong generalization capabilities and effectively preserved fine details and texture information.
• Superior edge preservation was observed compared to other tested denoising algorithms.
• Denoised images were generated within 0.4 seconds, meeting clinical application time constraints.

Conclusions:

• The developed CNN-based denoising algorithm offers a robust solution for speckle noise in anterior segment OCT imaging.
• The approach successfully enhances image quality by removing noise while retaining critical structural and textural details.
• The algorithm's efficiency and effectiveness make it suitable for real-time clinical applications, improving diagnostic accuracy.