3D high resolution generative deep-learning network for fluorescence microscopy imaging
Hang Zhou, Ruiyao Cai, Tingwei Quan, Shijie Liu, Shiwei Li, Qing Huang, Ali Ertürk, Shaoqun Zeng
Abstract
Microscopic fluorescence imaging serves as a basic tool in many research areas including biology, medicine, and chemistry. With the help of optical clearing, large volume imaging of a mouse brain and even a whole body has been enabled. However, constrained by the physical principles of optical imaging, volume imaging has to balance imaging resolution and speed. Here, we develop a new, to the best of our knowledge, 3D deep learning network based on a dual generative adversarial network (dual-GAN) framework for recovering high-resolution (HR) volume images from high speed acquired low-resolution (LR) volume images. The proposed method does not require a precise image registration process and meanwhile guarantees the predicted HR volume image faithful to its corresponding LR volume image. The results demonstrated that our method can recover ${20} {\times} /1.0\text-{\rm NA}$20×/1.0-NA volume images from coarsely registered ${5} {\times} /0.16\text-{\rm NA}$5×/0.16-NA volume images collected by light-sheet microscopy. This method would provide great potential in applications which require high resolution volume imaging.