A structure-invariant color normalization method for pathological images based on deep learning

Abstract

The invention discloses a pathological image color standardization method based on deep learning with invariant structure. The method includes: constructing a model training set; selecting an image with excellent dyeing quality in the training set as a template image, and the remaining images as original images; Build a deep learning model; input the grayscale image of the original image and the Lab color space image of the template image into the network, train the deep learning model, and obtain standardized a and b channel images; after normalizing the L channel of the original image and a and b The channels are combined to obtain a normalized RGB channel image. The present invention uses a deep learning model combined with the principle of automatic coloring, utilizes the structural information of the grayscale image and the color information of the template image, converts the color style of the original image into the color style of the template image, and realizes color standardization of pathological images.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a structure-invariant pathological image color standardization method based on deep learning. Background technique [0002] With the innovation of medical technology, the cell-level imaging of tumors can be presented in pathological images through a series of processing, and pathological images have become an important reference for tumor diagnosis and prognosis. However, in the process of pathological image imaging, the preparation and digitization of tissue samples will lead to color changes, which not only cause interference when pathologists read the images, but also bring inestimable errors to computer pathological image analysis. Therefore, standardizing the pathological images and standardizing the obtained pathological images to a certain color distribution has very important clinical significance. [0003] There are mainly two existing standardizati...

AI Technical Summary

Problems solved by technology

Non-negative matrix decomposition is an iterative process, which has two disadvantages: 1. The separation result is easy to fall into the local minimum, 2. In order to calculate a more accurate value, it takes a long iteration time

But in fact, it is difficult for us to guarantee that all dyeing styles can be covered during training, and there is no way to predict what kind of dyeing styles will be encountered in practical applications.

Images