Research on Dispersion Compensation of FD-OCT System via Pix2Pix GAN Technique

Abstract

Dispersion in optical coherence tomography (OCT) poses a challenge that is exacerbated by the increased spectral bandwidth, which leads to image blur and feature loss. In this paper, we present a straightforward and cost-effective approach for dispersion compensation in OCT. To achieve this, we employed a pixel-to-pixel (Pix2Pix) generative adversarial network (GAN) architecture customized for image-to-image translation. Two data groups with varying amounts of training image data and epochs were used. The Pix2Pix GAN was trained to generate clear OCT images from the corresponding dispersion-affected OCT images in paired datasets. According to the experimental results, the Pix2Pix GAN technique demonstrated a substantial improvement over the basic GAN. Specifically, it increases the peak signal-to-noise ratio (PSNR) by 159%, structural similarity index (SSIM) by 370%, and Fréchet inception distance (FID) by 274%. These outcomes indicate that the proposed model can generate images with resilience and effectiveness, particularly when dealing with dispersion-affected OCT data.

Keywords

• Generative adversarial network
• optical coherence tomography
• Pix2Pix