A Novel Diffusion-Model-Based OCT Image Inpainting Algorithm for Wide Saturation Artifacts

Saturation artifacts in Optical Coherence Tomography (OCT) images will affect the image quality and reduce the accuracy of clinical diagnosis. Recently, the researcher proposed various OCT image inpainting algorithms for saturation artifacts, and these algorithms were limited to .oct format files only (spectral data) or simple interpolation algorithms, which led to the failure of the best performance on wide saturation artifacts. In this paper, a novel image inpainting model based on a generative model (diffusion model) is proposed, which can recover degraded regions in OCT images. Experimental results show that the average PSNR and SSIM values outperformed existing approaches. Besides, the classification models, vision transformer (ViT), for OCT images were implemented to compare the accuracy difference before and after the proposed image inpainting algorithm. The proposed algorithm presents a promising solution for better OCT image inpainting methods.