3D Imaging Restoration of Spinning-Disk Confocal Microscopy Via Deep Learning

Abstract

Due to the multipoint excitation and simultaneous detection strategy applied, spinning-disk confocal microscopy (SDCM) results in an increased imaging speed compared to conventional confocal microscopy. Additionally, the super-resolution radial fluctuations (SRRF) approach can further improve the imaging resolution of SDCM in 3D imaging at the cost of imaging time due to the large amounts of data acquisition and the increased risk of photo-bleaching and photo-toxicity due to the multiple excitations. Here, we propose a deep learning-based method for 3D SDCM, where the neighboring pixels in $z$ -scanning slices are taken into account for 3D reconstruction. Consequently, high-quality imaging slices can be reconstructed directly from the SDCM stacks with a single scan. The image quality achievable with this SRRF-Deep method is comparable with the SRRF method, whereas it achieves image reconstruction about 30 times faster using 100 times fewer images. Thus, practicality of the SDCM system can be significantly improved in 3D imaging.