Super Resolution Method for Fluorescent Image Deconvolution

Abstract

A new algorithm for microscope fluorescent images deconvolution is presented that retrieves the information with super resolution. The method is particularly fit for sparse data, such as encountered in typical intracellular experiments, where only a minor fraction of the explored volume has fluorescent markers. The technique retrieves by construction only positive values for the spatial density, avoiding the need for nonlinear constrains found in prior deconvolution techniques such as Tikhonov-Miller and Richardson-Lucy. A genetic algorithm is performed to optimize the solution. The method automatically subtracts the background from the image. Based on the measured point spread function with the information of the quality of the fit and the information on the noise figure of the camera as a function of the read signal, the method provides a predictor of the uncertainty in the reconstruction both in lateral resolution and amplitude. Artificially synthesized images including noise were used to test the method, showing the reconstruction below the 50nm spatial resolution when analyzing arrays of molecules or fluorescent lines. Images of 100nm and 200nm fluorescent beads and bead clusters are also efficiently and precisely deconvolved. For the images obtained with an Andor Zyla5.5 camera with a measured noise given by Noise= 23+sqrt(N) (where N is the count of the pixel) the theoretical predictor mentioned before gave a lateral resolution of 44nm consistent with the experimental and simulated results. Examples of intracellular fluorescent image reconstruction are presented using bovine pulmonary artery endothelial cells with fluorescent labels for the F-actin, microtubules and mitochondria. The extension of the technique to 3D deconvolution will also be discussed.