Cell nucleus center point detection method based on multi-task convolutional neural network

Abstract

The invention discloses a cell nucleus center point detection method based on a multi-task convolutional neural network. The method comprises the following steps: dotting and marking central points ofcell nucleuses, generating two types of masks during training directly according to coordinates of dotting, with a single type of Gaussian kernel being a Gaussian distribution two-dimensional array with the central point coinciding with the marked central point, and two types of equivalent kernels being an equivalent circular structure element two-dimensional array with the central point coinciding with marking data; a full convolutional network structure is adopted, two types of equivalent structure masks are firstly used for training two types of segmentation models, parameters of the segmentation models are used for initializing a multi-task model, and then the whole model is finely adjusted by combining losses of two task branches; and during prediction, inputting an RGB image, outputting two types of segmentation images and a single-type probability image by the model, and determining the position and the type of a final cell nucleus by combining the two types of segmentation images. According to the method, the problems of false negative, false positive and low convergence rate during direct regression of Gaussian nucleuses or equivalent structural element segmentation during pathological image cell nucleus detection are solved.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a method for detecting the central point of a cell nucleus based on a multi-task convolutional neural network. Background technique [0002] The expression of ki67 is closely related to the occurrence of many tumor diseases, such as breast cancer, ovarian cancer, lymphoma and so on. Clinically, the ki67 index is often detected by immunohistochemical methods to reflect the proliferation activity of normal and diseased tissues or cells, to differentiate benign and malignant tumors, and to help the early diagnosis of malignant tumors, the selection of treatment methods and the evaluation of curative effects. The current method for determining the positive expression rate of ki67 is to check by pathologists under a microscope, but this is very time-consuming and accompanied by the risk of low consistency. At present, more and more institutions are trying to build a computer-a...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims to provide a method for detecting the nucleus center point based on a multi-task convolutional neural network, which solves the problems of false negatives, false positives, and convergence speed when directly regressing to Gaussian kernels or dividing equivalent structural elements in pathological image nucleus detection. slow problem

Images