X-Microscopy: Multicolor Super Resolution Image Reconstruction from Conventional Microscopy with Deep Learning

Deep learning shows an outstanding potential to transform low-resolved images into super-resolved images. However, multicolor super resolution microscopy (SRM) imaging from wide-fields (WFs) and enabling super-resolved images across different microscopic modalities by deep learning are poorly defined. Here, we devise X-Microscopy, a deep learning network, that leverages binary branches symmetrically and reciprocally to extract distinct fine details of two inputs, and then the extracted features are hierarchically embedded into a reconstructed high-fidelity SRM image by channel attention and residual. X-Microscopy can reconstruct various super-resolved images from different focus of WFs in a few seconds while providing robust input-size flexibility and considerable generalization to different structures and experimental conditions. X-Microscopy can also fast and accurately display multicolor super-resolution images. Furthermore, X-Microscopy achieves cross-modality imaging between different microscopic systems without any iterations or parameter search. Thus, X-Microscopy has great potential applications in biomedical research fields.