Histology Image Artifact Restoration with Lightweight Transformer Based Diffusion Model

Chong Wang, Zhenqi He, Junjun He, Jin Ye, Yiqing Shen

Abstract: Histology whole slide images (WSIs) are extensively used in tumor diagnosis and treatment planning. However, the presence of artifacts resulting from improper operations during WSI collection can impede both manual and deep learning-based analysis. To bridge this gap, we propose an innovative denoising diffusion model tailored for inpainting artifact-laden regions in histology WSIs. Our method focuses on preserving the intricate morphological structures, which are essential for accurate diagnosis. To ensure the preservation of morphological structures during regional artifact inpainting, we have developed a novel lightweight Transformer-based denoising network, that can capture the correlations between the regional artifact with the global morphological structures. In comparison to existing generative adversarial network (GAN) based solutions, our method minimizes changes in morphology while maximizing preservation of the stain style during the restoration of the artifact. By providing a more reliable and accurate restoration of artifact-affected areas, our model facilitates better analysis and interpretation of histological images, thereby potentially improving the accuracy of tumor diagnosis and treatment decisions. The code is available at https://github.com/zhenqi-he/artifact-restoration.