Deep learning virtual Zernike phase contrast imaging for singlet microscopy

Abstract

Singlet microscopy is very attractive for the development of cost-effective and portable microscopes. In contrast to conventional microscope objectives, which consist of multiple lenses, the manufacturing process for singlet lenses is done without extensive assembling and aligning. In this manuscript, we report a novel singlet virtual Zernike phase contrast microscopy setup for unstained pathological tumor tissue slides. In this setup, the objective consists of only one lens. There is no need for the inset Zernike phase plate, which is even more expensive than a whole brightfield microscopy setup. The Zernike phase contrast is virtually achieved by the deep learning computational imaging method. For the practical virtual Zernike phase contrast microscopy setup, the computational time is less than 100 ms, which is far less than that of other computational quantitative phase imaging algorithms. With a conceptual demo experimental setup, we proved our proposed method to be competitive with a research-level conventional Zernike phase contrast microscope and effective for the unstained transparent pathological tumor tissue slides. It is believed that our deep learning singlet virtual phase contrast microscopy is potential for the development of low-cost and portable microscopes and benefits resource-limited areas.