Retinal OCT Denoising with Pseudo-Multimodal Fusion Network

Area of Science:

• Ophthalmology
• Biomedical Imaging
• Medical Image Processing

Background:

• Optical coherence tomography (OCT) is crucial for retinal imaging but suffers from multiplicative speckle noise.
• Speckle noise obscures vital anatomical details like blood vessels and tissue layers.
• Current noise reduction methods, like frame averaging, increase acquisition time and introduce artifacts.

Purpose of the Study:

• To develop an efficient, learning-based method for speckle noise suppression in OCT images.
• To enhance the visibility of retinal anatomical structures.
• To preserve fine features such as small blood vessels.

Main Methods:

• A learning-based approach utilizing single-frame noisy B-scans and a self-fusion-generated pseudo-modality.
• A fusion network to combine features from both modalities with adjustable weights.
• Evaluation using intensity-based and structural metrics.

Main Results:

• Effective suppression of speckle noise in OCT retinal images.
• Enhanced contrast between retinal layers while preserving overall structure.
• Preservation of fine features, including small blood vessels.
• Improved structural similarity to low-noise B-scans (0.559 to 0.576).

Conclusions:

• The proposed method successfully reduces speckle noise in OCT.
• It enhances retinal layer contrast and preserves critical anatomical details.
• This approach offers a promising alternative to traditional noise reduction techniques.