Cell counting method based on depth deconvolution neural network

Abstract

The invention discloses a cell counting method based on a depth deconvolution neural network. The method comprises the steps: 1, constructing the depth deconvolution neural network; 2, carrying out the preprocessing of a cell image; 3, training a network model; 5, firstly setting a threshold and removing miscellaneous points, and secondly calculating the number of remaining communication blocks in the image, wherein the number of number of remaining communication blocks in the image serves as the number of cells; 5, firstly inputting a non-trained cell image after preprocessing to the depth deconvolution neural network with the optimized network connection weight, secondly obtaining a Gaussian kernel heat map of an original cell image, and finally obtaining a final count value after post-processing. The method carries out the mining and extraction of the feature and spatial information of the cell image through a deconvolution deep learning network, enables an output layer image to be recovered to a size equal to the size of an input layer image, trains one end-to-end network, and can guarantee the higher counting accuracy.

Description

technical field [0001] The invention relates to the fields of pattern recognition and machine learning, in particular to a method for counting cell images under a microscope based on a deconvolution neural network. Background technique [0002] In the study of functional diversity and clinical pathology of biological samples, cell counting is extremely important. Cell images acquired under a microscope have many different shapes, including isolated cells and adherent cells. The isolated cell count is relatively Easy, but for the case of high overlap of cohesive cells, the traditional method is generally based on the segmentation method, and counting after segmentation, but for cell images with high overlap, it is difficult to get a better segmentation effect, Further, it is difficult to obtain accurate counting results. [0003] In recent years, deep learning has been applied to various fields of image processing. As a field most closely related to human life, deep learning...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method for counting cell images under a microscope based on a deep deconvolution neural network. The present invention regards the cell counting problem as a regression problem from the original cell image to the output Gaussian kernel thermal image, and solves the traditional method for Situations where high cell overlap makes segmentation difficult

Images