A Pathological Full Slice Retrieval Method Based on Deep Hash

Abstract

The invention discloses a pathological full-slice retrieval method based on deep hashing, which includes: establishing a hash database: inputting a query image, obtaining labeled image blocks, inputting the labeled image blocks into a convolutional neural network, The CNN network framework is trained through the slice-level loss function, and all training images are converted into corresponding binary codes to obtain a hash database; the determination of the retrieval results: use the CNN network framework to extract the binary code of the test image, and pass the hash The code similar to the binary code of the test image is retrieved from the database, and then the corresponding image is returned, and the returned image is fused by non-maximum value suppression to obtain the retrieval result. The present invention utilizes the slice label information of the image block, and adds a slice-level suppression item in the loss function, so that images from the same slice have a greater degree of discrimination in the feature space.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, and more specifically relates to a pathological full slice retrieval method based on deep hashing. Background technique [0002] Digital pathological image analysis is an important auxiliary diagnostic method in current cancer diagnosis. Digital pathological image refers to the digital image obtained after the stained pathological section is imaged under a microscope. Content-based image retrieval can find pathological images and cases similar to the target patient in the database, and provide abundant auxiliary information for doctors to diagnose. In recent years, pathological image retrieval has become a research hotspot. In the previous research, the traditional retrieval method is mainly aimed at small-scale data sets, extracting image feature descriptors, such as SIFT, HOG, GIST features, etc., and then using the semantic analysis model to analyze and process the features, b...

AI Technical Summary

Problems solved by technology

[0004] The PDRH algorithm in the prior art is based on the retrieval of the image block level, and adopts a strategy of ignoring the slice information of the image block, and a large number of redundant results from the same slice and the same region may appear in the returned results (such as image 3 (a)); Although the PDRH algorithm can achieve good retrieval accuracy, the diagnostic information provided is very limited, and the existing diagnostic information data cannot be fully utilized

Images