Neural network-based manually annotated white blood cell error calibration method and device

Abstract

The present invention discloses a neural network-based manually annotated white blood cell error calibration method and device. The calibration method comprises the steps of S1 obtaining an original image; S2 pre-processing the image; S3 segmenting the white blood cells and extracting the edge pixel point coordinates of the cells; S4 selecting the cells having the obvious characteristics as the training data and using the other cells as the test data to train a network; S5 using a softmax classifier to score the test cells, and dividing the test cells into the concrete categories and the sub-categories according to the scores; S6 carrying out the polar coordinate data enhancement on the training cells and the sub-category cells; S7 retaining the enhanced training images, using the sub-category images to test and scoring; S8 according the scores, rejecting the cells that are not within the appointed categories and dividing the other cells into the concrete categories. According to thepresent invention, during the classification process, the interference cells that are meaningless to the classification can be rejected, and the peripheral blood white blood cells can be classified into a plurality of small categories, thereby increasing the cell classification accuracy substantially.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to an error calibration method and device for artificially marking white blood cells based on a neural network. Background technique [0002] In recent years, with the rapid development of artificial intelligence in various fields, how to use artificial intelligence to replace doctors' manual operations to detect and classify various cells in medical pathology detection has become a hot issue of widespread concern. The process of artificial intelligence detecting and classifying cells is as follows: First, obtain the labeled training samples of the cells to be detected, learn the characteristics of each classification after training the neural network, and then input the test cells into the network, and the network can , to intelligently classify cells. [0003] At present, in the medical laboratory department, the existing blood testing process is roughly divided into two steps: ...

AI Technical Summary

Problems solved by technology

[0006] The above-mentioned blood detection process has the following disadvantages: the steps of detecting blood cells are cumbersome, it is necessary to roughly detect 5 types of cells first, and then re-test for a specific cell, and the detection rate of misdiagnosis is high

And the number of cell classifications is too small, only 5 categories can be distinguished, and some small classification cells that affect disease detection cannot be recognized; if you want to increase the recognition of a small classification, you need to add a series of measures such as dyes, which are expensive. cumbersome procedures

Images