Research And Implementation Of Medical Image Retrieval System Based On Attention Clustering Feature

Medical images have the characteristics of high gray resolution,great similarity and many categories.It is still a challenge to retrieve the required images efficiently and accurately in a large number of medical images.The current content-based medical image retrieval(CBMIR)system still has the problem of low retrieval accuracy and unsatisfactory retrieval effect.In order to better improve the performance of medical image retrieval,this paper proposes a medical image retrieval model combining attention network with clustering network.The main idea is to assign clustering characteristics to features extracted by the designed multi-scale attention convolutional neural network through clustering layer,and then apply the features to image retrieval.This paper is based on the IRMA medical image data set for image feature extraction and retrieval research.In this project,the author mainly completed the following tasks:(1)Related algorithm research.This paper briefly introduces the background of medical image retrieval,analyzes the principle and application of convolutional neural network,and studies several key technologies of medical image retrieval in detail.(2)Proposal.By studying the mechanism of neural network,an attention-based network structure is proposed.The proposed scheme uses the designed attention network algorithm to extract features,and then performs feature clustering through the proposed CenterVLAD layer to obtain the final robust medical image features for retrieval.The experimental results show that the scheme is effective for image retrieval,and the mAP on the experimental data set reaches 0.79.(3)Design and implementation of retrieval system.This project developed a medical image retrieval system based on PyQT5+MySQL.The system integrates the proposed algorithm and other main retrieval algorithms,and mainly realizes the functions of image preprocessing,image retrieval and image viewing.This system is convenient to assist doctors to retrieve images of similar cases.