Probabilistic volumetric speckle suppression in OCT using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Speckle noise in Optical Coherence Tomography (OCT) degrades image quality and hinders accurate diagnosis.
• Developing effective speckle reduction techniques is crucial for enhancing OCT's clinical utility.
• Existing methods often struggle with preserving fine details or are computationally intensive.

Approach:

• A conditional generative adversarial network (cGAN) framework was designed to leverage volumetric OCT data for speckle reduction.
• The cGAN takes partial OCT volumes as input, producing artifact-free, despeckled volumes with preserved resolution in all three dimensions.
• Volumetric non-local means despeckling (TNode) was utilized to generate high-quality training data, serving as a gold standard despite its computational demands.

Key Points:

• The proposed cGAN achieves efficient speckle suppression and preserves tissue structures comparable to TNode.
• The deep learning approach is two orders of magnitude faster than TNode, enabling real-time applications.
• The framework demonstrates high-quality despeckling across different tissue types and OCT systems.
• An open-source, all-software implementation facilitates retraining and deployment on various OCT systems.

Conclusions:

• The developed deep learning framework offers a fast and effective solution for volumetric speckle reduction in OCT.
• This method significantly improves image quality by removing speckle noise while maintaining structural integrity.
• The accessibility and speed of the proposed technique make it a valuable tool for advancing OCT imaging applications.