Deep-learning-based optical coherence tomography reconstruction for high-speed and contrast morphology and...

Area of Science:

• Ophthalmic imaging
• Medical image analysis
• Deep learning applications

Background:

• Spectral-domain optical coherence tomography (SD-OCT) offers high resolution but faces challenges with imaging speed and sensitivity.
• Improving both speed and sensitivity simultaneously in SD-OCT is a significant technical hurdle.

Purpose of the Study:

• To develop a deep learning approach to enhance both imaging speed and sensitivity in SD-OCT systems.
• To overcome hardware limitations in SD-OCT through advanced computational methods.

Main Methods:

• A modified U-Net deep neural network (DNN) architecture incorporating a visual state space model was employed.
• The model synthesizes high signal-to-noise ratio (SNR) OCT and OCT angiography (OCTA) from high-speed acquisitions.
• Performance was evaluated using quantitative metrics like multiscale structural similarity index measure (MS-SSIM) and contrast-to-noise ratio (CNR).

Main Results:

• The deep learning model effectively reconstructed high-SNR OCT and OCTA images.
• Improved contrast between retinal layers and enhanced delineation of layer boundaries were observed.
• Fine retinal structures, including microcapillaries and choroid, were successfully restored.

Conclusions:

• A novel DNN-based architecture enables simultaneous improvements in SD-OCT imaging speed and sensitivity.
• This approach enhances OCT image quality without compromising acquisition speed.
• The method offers a viable solution for overcoming inherent hardware limitations in SD-OCT systems.