This code is for generating the high-density multi-color sSMLM super-resolution images from the low density ones (upto 8-fold fewer frames) from our following Biomedical Optics Express Paper.
Sunil Kumar Gaire, Yang Zhang, Hongyu Li, Ray Yu, Hao F. Zhang, and Leslie Ying, "Accelerating multicolor spectroscopic single-molecule localization microscopy using deep learning," Biomed. Opt. Express 11, 2705-2721 (2020)
Three separate codes to test the reconstruction of high-density data from the low-density one obtained from sSMLM/STORM imaging. For sSMLM, each code will reconstruct the images of tubulin, mitochondria, and peroxisome channels.
-
Input: The test image must be in the "Test/*" directory. Tub is for tubulin, Mito is for mitochondria and Pero is for peroxisome. Image must be .tif file format (For example, .tif image from ThunderSTORM) Training is configured for 10x magnification. So 10x magnified (average shifted histogram) from sSMLM/STORM localization data will give the best results.
-
Output: Results will be saved in the "result" directory Output will be saved in a .tif file format. An intensity thresholding is necessary after prediction. Use separate postprocessing Matlab code: MSSSIM_calculation for proper final image visualization. After running MSSSIM_calculation code in MATLAB the image can also be exported in ImageJ using Show_Image_ImageJ Matlab code.
Size is the patch size: 32 or 64 is available (only 64 for tubulin)
Note: Make sure that the test image size is multiple of 32 or 64. If not, the output image will be slightly reduced in size.
Required Python dependencies to run the code are in the requirements.txt file.
Reference:
Sunil Kumar Gaire, Yang Zhang, Hongyu Li, Ray Yu, Hao F. Zhang, and Leslie Ying, "Accelerating multicolor spectroscopic single-molecule localization microscopy using deep learning," Biomed. Opt. Express 11, 2705-2721 (2020)