Three-Dimensional Virtual Optical Clearing With Cycle-Consistent Generative Adversarial Network

• DOI: 10.3389/fphy.2022.965095
• Publication Date: 2022-07-19T04:50:36Z
• Authors: Jiajia Chen; Zhenhong Du; Ke Si
• Keywords: Clearance; Clearing

High-throughput deep tissue imaging and chemical clearing protocols have brought out great promotion in biological research. However, due to uneven transparency introduced by anisotropy imperfectly cleared tissues, fluorescence based on direct still encounters challenges, such as image blurring, low contrast, artifacts so on. Here we reported a three-dimensional virtual optical method unsupervised cycle-consistent generative adversarial network, termed 3D-VoCycleGAN, digitally improve quality of samples. We demonstrated the good deblurring denoising capability our mouse brain kidney tissues. With 3D-VoCycleGAN prediction, signal-to-background ratio (SBR) images areas also showed above 40% improvement. Compared other deconvolution methods, could evidently eliminate opaqueness restore larger 3D inside imperfect tissues with higher efficiency. And after virtually cleared, depth were increased up 30%. To knowledge, it is first interdisciplinary application CycleGAN learning model fields, promoting development high-throughput volumetric techniques.