Datasets, Code, and Trained Models for X-Microscopy

We present X-Microscopy, a computational tool comprising two deep learning subnets: UR-Net-8 and X-Net. This tool enables STORM-like superresolution microscopy image reconstruction from wide-field images with input-size flexibility.To train X-Microscopy, datasets were acquired consisting of wide-field images (WFs) and paired superresolution microscopy images (SRMs) of immunostained various subcellular structures in different cell lines. The WFs and SRMs were obtained using a PlanApo TIRF 100×/1.49 NA oil immersion objective (CFI SR HP Apochromat TIRF 100x/1.49NA, Nikon). All WF-SRM training pairs were well matched, and there was no need for pixel registration.The folder named "example data" provided example training datasets of indicated structures. The training details were described as follows: UR-Net-8 was constructed to generate mimic undersampled SRM images (MU-SRMs) from WFs. For the MT model instance, WFs were defined as network inputs, and undersampled SRM images (k=10000) served as the ground truth during training. The network outputs were mimic undersampled SRM images (MU-SRMs). X-Net was equipped with two computational paths. The input of the upper path was the WF, and the input of the lower path was the MU-SRM generated by UR-Net-8. The W-SRMs were defined as the ground truth images during training, and the F-SRMs were used to calculate SSIM values after training.The folder named “the trained models” provided the trained models of these structures to test low resolution WFs. The folder named "code-X-Microscopy " provided the code of X-Microscopy. This code required a standard computer with enough RAM to support the in-memory operations and the GeForce GTX 1080 GPU (The NVIDIA Inc.) to support GPU computing.