Imaging Enhancement of Light-Sheet Fluorescence Microscopy via Deep Learning

Abstract

The complementary beam subtraction (CBS) method can reduce the out-of-focus background and improve the axial resolution in light-sheet fluorescence microscopy (LSFM) via double scanning a Bessel and the complementary beams. With the assistance of a compressed blind deconvolution and denoising (CBDD) algorithm, the noise and blurring incurred during CBS imaging can be further removed. However, this approach requires double scanning and large computational cost. Here, we propose a deep learning-based method for LSFM, which can reconstruct high-quality images directly from the conventional Bessel beam (BB) light-sheet via a single scan. The image quality achievable with this CBS-Deep method is competitive with or better than the CBS-CBDD method, while the speed of image reconstruction is about 100 times faster. Accordingly, the proposed method can significantly improve the practicality of the CBS-CBDD system by reducing both scanning behavior and reconstruction time. The results show that this cost-effective and convenient method enables high-quality LSFM techniques to be developed and applied.