Digital pathology whole slice image retrieval method

Abstract

The present invention discloses a digital pathology whole slice image retrieval method, and the method is used in a digital pathology whole slice image database. The method comprises: extracting positions of dispersed SIFT feature points and SIFT feature vectors on a digital pathology whole slice image in the database; using an LDA model to obtain a high-level semantic feature value for each SIFT feature point; using an overlapping sliding window method to select alternative regions, and collecting statistics of semantic feature values of all the SIFT feature points in each alternative region, so as to obtain semantic representation vectors of the alternative region; and taking a query image as a region, using the same method to obtain the semantic representation vector of the query image, calculating cosine distances between the semantic representation vector of the query image and semantic representation vectors of all the alternative regions, sorting the distances, and returning to multiple regions with smallest distance. The method disclosed by the present invention can provide diagnosis reference information for the pathologist, and can be used for a digital pathology whole slice image database management and query system and computer-aided diagnosis.

Description

technical field [0001] A digital pathology full slice image retrieval method, belonging to the field of digital image processing and machine learning, especially digital image processing technologies such as scale-invariant feature transform (SIFT), content-based image retrieval, latent Dirichlet Allocation (LatentDirichletAllocation, LDA) and other machine learning techniques. Background technique [0002] Digital pathological full slide image (hereinafter referred to as full slide) is a large-size, high-resolution digital image obtained by scanning and collecting traditional glass pathological slides through a fully automatic microscope or an optical magnification system. It is an important basis for pathologists in diagnosis. In recent years, with the development of pathology and computer technology, the number of digital pathological full-slice images has grown rapidly. Finding a full-slice region similar to an undiagnosed small-size pathological image from the full-slic...

AI Technical Summary

Problems solved by technology

[0004] There are currently two types of methods for image representation: 1. Directly use the underlying features to represent images, but the underlying features are very different from human understanding of images, which cannot accurately summarize image content and affect image retrieval accuracy; 2. Semantic model representation Image, semantic model is generally a summary of the underlying features, which is more in line with human thinking and improves the accuracy of image retrieval

Images