A Self-Supervised Denoising Method for pCLE Images

When using probe based confocal laser endomicroscopy (pCLE) to diagnose diseases, the fiber bundle makes a honeycomb pattern on the result which causing significant impact on the clinical diagnosis. Therefore, one of the main challenges of this technique is the quality of images. Recently, deep learning technique have made significant progress on image denoising. However, acquiring noise-clean training image pairs for each scene is costly in practice. There have been a few works that allow training a denoising network on signal noisy image. Taking one step further, this paper propose a novel self-supervised denoising method by designing a image quality assessment network to cooperate with the denoising network. Compared with previous self-supervised denoising methods, our method shows better results on pCLE images and fluorescence microscope images.