A Medical Image Retrieval Method Based on Deep Learning and Radon Transform

Abstract

A medical image retrieval method based on deep learning and Radon transform involves the fields of computer vision and image retrieval. In the "coarse" retrieval stage, the BING target suggestion algorithm is used to detect regions with significant objects. By introducing partial mean pooling into the deep convolutional network architecture, significant distinguishing features based on regions can be extracted and feature dimensions can be reduced, and then aggregated to form A global feature representation. In the process of eigenvector quantization, the product quantization algorithm is used to solve the problem of high computational complexity of the similarity measure between eigenvectors. In the "fine" retrieval stage, with the help of Radon transformation, the image can be integrally projected from multiple angles to obtain the characteristics of more detailed information of the image. The Top50 images obtained in the "coarse" retrieval are generated through Radon transformation to generate a Radon barcode, and the similarity measurement is achieved. Search more precisely. The invention improves the accuracy rate of medical image retrieval, and overcomes the problem of low feature distinction and high feature dimension of medical image retrieval caused by direct use of convolutional neural network.

Description

Technical field [0001] The invention relates to the field of computer vision and image retrieval, in particular to a medical image retrieval method based on deep learning and Radon transformation. Background technique [0002] With the increasing application of image processing technology in the medical field, a large number of medical images are generated every day, such as CT images, B-ultrasound images, MRI images, etc., which are an important basis for clinical diagnosis and treatment and medical image research. How to effectively manage these medical images and retrieve the information that doctors need from them is an important topic in medical image research today. The traditional content-based medical image retrieval (CBMIR) method is to sequentially compare the query image with the image in the database one by one. Its linear complexity leads to its shortcomings such as low efficiency and low scalability in the real environment, while extracting features It is a low-lev...

AI Technical Summary

Problems solved by technology

At the same time, in the process of eigenvector quantization, the product quantization algorithm is used to solve the problem of high computational complexity of the similarity measure between eigenvectors

Images