Self-supervised learning for line artifact removal in light sheet fluorescence microscopy

Abstract

Light sheet fluorescence microscopy (LSFM) is a microscopic technique to obtain biological images rapidly, allowing thin light sheets to pass through the plane of a sample to obtain a single two-dimensional (2D) slice image. However, during the process of illuminating the entire slice in one plane and detecting light from a direction perpendicular to it, shadow-like line artifacts can be generated when the light is blocked by structures such as blood vessels. This paper proposes a deep learning-based solution to remove the line artifacts in LSFM. The characteristics of line artifacts were utilized to separate the LSFM image into a line artifact mask and an artifact-removed image and the network was fully trained in self-supervised learning. By comparing images before and after removing line artifacts, it is shown that the proposed method effectively removed the line artifacts. If the proposed method is applied to remove line artifacts after obtaining LSFM images, it will be able to obtain high-quality biological images at a high speed.