The Research And Implementation On Medical Image Automatic Classification Algorithm Applied For Manage System Of Medical Image

With the continuous development of medical informatization in China and the continuous breakthroughs in the field of computer vision,these advances provide opportunities for data mining research based on medical image diagnosis,teaching and biomedical research.Before the medical images in the cloud center of image is reused,it needs to be classified first.Due to the huge increase of the number of medical images,manual classification can no longer meet the demand,and the need of automatic classification and management of medical images by computer is imminent.After summarizing the relevant research progress of medical image classification at home and abroad,this paper introduces the common classification methods.According to the previous research results and requirements of automatic classification of medical images,we start at feature extraction and feature fusion first,and worked at the following three aspects:(1)A medical image classification method based on the fusion of global feature and local feature is proposed.After introducing the common global features of medical images and the extraction methods of local features,this paper analyzes the advantages and disadvantages of local features and global features.In order to solve the problem of low classification accuracy based on single features,a feature extraction method combining texture features and SURF features is proposed.Firstly,the texture features of medical images are extracted by using Gabor filter and gray level co-occurrence matrix.Secondly,the local features of the images are extracted by SURF algorithm and transformed into a histogram by bag of visual word.Finally,the above features are normalized and fused.The method includes features of the medical images in its entirety and detail,enabling the medical image to be described and identified from a variety of perspectives.(2)An improved algorithm of recursive feature elimination is proposed.This paper deeply studies the algorithm of feature optimization.The advantages and disadvantages of the principal component analysis method,stability selection method and recursive feature elimination algorithm are analyzed in detail.In order to solve the problem of reduced classification accuracy caused by high feature dimension and feature redundancy after feature fusion,this paper proposes an improved RFE algorithm combining stability selection method and recursive feature elimination algorithm.Firstly,the original data set is randomly divided into several mutually exclusive subsets.The RFE algorithm is used to sort the features of each subset to obtain a sorting results.Then the above process is repeated by several times,and finally each round of sorting results are combined with,the final feature sorting is obtained.Compared with the original RFE algorithm,the algorithm can effectively improve the accuracy of classification results.(3)An improved KNN-SVM classification algorithm is proposed.This paper deeply studies the classification algorithm of medical image.The K-nearest neighbor algorithm and SVM classification algorithm are analyzed in detail,the principle as well as advantages and disadvantages of KNN-SVM classification algorithm and KD tree are studied.According to the needs of medical image management system,the shortcomings of traditional KNN-SVM algorithm and the problems of the KD tree,an improved KNNSVM algorithm is proposed based on incremental learning.Firstly,the feature vectors are optimized before constructing the KD tree.Secondly,for the sample to be classified,several nearest neighbors are found through the KD tree,and different nearest neighbors search algorithms are selected according to the feature optimization result.Finally,the nearest neighbor points are weighted by similarity and the SVM classification is trained by using these points,and we can get the classification result of the sample to be classified.The algorithm improves classification efficiency as well as accuracy and is suitable for medical image management systems.In this paper,the algorithm is verified by experiments.The feature fusion,improved RFE algorithm and improved KNN-SVM algorithm are used in the open data set respectively,and comparative experiments and analysis are carried out to prove the improved RFE algorithm and KNN-SVM classification algorithm can effectively improve the classification accuracy and efficiency of medical images in this paper..Finally,the algorithm is applied to medical image management system,which improves the efficiency of medical image classification and facilitates the further study of subsequent medical images.