Microscope deconvolution neural network model construction method based on Richardson-Lucy algorithm

Abstract

The invention discloses a microscope deconvolution neural network model construction method based on a Richardson-Lucy algorithm. A large number of fluorescence microscope images in a training set and corresponding deconvolution sample estimation are input into a constructed neural network; a deconvolution relationship between acquired images blurred by a fluorescence microscope and sample estimates as labels is learned. It is worth mentioning that the network is universal, shows high deblurring ability in the testing process for data collected by the same fluorescence microscope, and also has good generalization ability for different types of data sets. In general, according to the method, a universal framework is established by using a neural network based on a Richardson-deconvolution structure, and a deconvolution mapping relation is established through a strong feature extraction capability, so that the non-iterative method deblurring of the fluorescence microscope image is completed.

Description

Technical field [0001] The present invention belongs to the field of fluorescence microscopic imaging, and in particular, a method of constructing a microscope deconvolum neural network model based on Richardson-Lucy algorithm. Background technique [0002] The optical microscope has a long history. It is one of the most powerful methods in modern biology research. The development of fluorescent labeling technology has promoted micro technology toward higher resolution and higher contrast. However, due to the diffraction of light, the micro image is inherently blurred; for three-dimensional fluorescent image, since only part of the light is collected from one direction from one direction, there is also a severe resolution anisotropy, the restricted resolution is usually not satisfied. Biology research is difficult for sample visualization, and will cause difficulty in post-processing of fluorescence microscopic images. Point expansion function (PSF) describes the response of the ...

AI Technical Summary

Problems solved by technology

Ideally, researchers hope to automatically find a suitable model from sample data to solve a certain type of problem through a data-driven method. The Chinese patent with the publication number CN111524078A proposes a dense network-based microscope image deblurring method. From the perspective of neural network, the deconvolution problem of the same fluorescence microscopy data is solved. However, this method is similar to a black box, lacks reasonable interpretation, sometimes introduces false artifacts, and also limits the generalization of the network. ability

Images