Sub2Full: split spectrum to boost optical coherence tomography despeckling without clean data.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Image Processing

Background:

• Speckle noise significantly degrades image quality in Optical Coherence Tomography (OCT), particularly in high-resolution visible light OCT (vis-OCT).
• Existing noise reduction methods often require clean, artifact-free data, which is challenging to obtain for OCT.
• The deterioration of image quality impacts the visualization of fine structures, such as sublaminar details in the retina.

Purpose of the Study:

• To introduce an innovative self-supervised learning strategy for OCT image despeckling.
• To address the challenge of speckle noise in high-resolution vis-OCT imaging.
• To develop a method that does not require paired clean data for training.

Main Methods:

• A novel self-supervised strategy named Sub2Full (S2F) was proposed for OCT despeckling.
• The method utilizes two repeated B-scans: a low-resolution input is generated by splitting the spectrum of the first scan.
• The full spectrum of the second scan serves as the high-resolution target for training.

Main Results:

• The Sub2Full (S2F) method was successfully validated on visible light OCT (vis-OCT) retinal images.
• The technique demonstrated effective despeckling, improving the visualization of sublaminar structures in the outer retina.
• S2F outperformed established self-supervised methods like Noise2Noise (N2N) and Noise2Void (N2V) in performance.

Conclusions:

• The proposed Sub2Full (S2F) strategy offers an effective solution for OCT image despeckling without clean data.
• This self-supervised approach significantly enhances image quality in vis-OCT, aiding in the visualization of delicate retinal structures.
• S2F represents a promising advancement in medical image processing for OCT applications.