AD Medical Image Mining Based On Class-imbalanced Learning Technology

With the development of information technology,medical imaging equipment and technology have become the main means for hospitals to diagnose diseases.The use of medical images such as CT and MRI is increasing,resulting in a large number of multiangle and high-resolution medical image data.It can provide doctors with more detailed image details,reduce the burden on doctors in diagnosing diseases,and help they improve the accuracy of diagnosis to analyze medical image data by computer-aided diagnostics.The mining analysis of medical images is an important part of computer-aided diagnosis.In practical applications,the acquired medical images are often classimbalanced,which leads to a significant decline in the performance of traditional classifiers.So this paper proposes AD medical image mining based on class-imbalanced learning technology,the purpose is to study the classification and identification of medical images of Alzheimer's disease with class-imbalanced distribution.The research idea firstly preprocesses the brain MRI medical image,determines the region of interest and extracts morphological features and texture features.Then it proposes feature selection algorithm based on random forest for imbalanced data to solve the classimbalanced problem.Finally,the medical images are classified.The experimental results verified the validity of the proposed method.Integrate these methods to study brain medical images of Alzheimer's disease with class-imbalanced distribution to explore new ways to diagnose Alzheimer's disease.The main study contents are as follows:(1)Summarize and analyze the related technologies of classification and recognition of brain medical images with classimbalanced distribution,including imaging technology,feature extraction,feature selection,data characteristics,classification algorithms,and analysis of the principle of each stage and its pros and cons.(2)Through the research and analysis of current brain medical image feature extraction technology,it is found that the feature fusion method can retain more details of the image,so this paper uses feature fusion method to fusion image morphological features and texture features.(3)Considering that redundant and uncorrelated features can directly affect classification performance,and the data in this paper are imbalanced.This paper studies the current feature selection method of classimbalanced data and proposes a feature selection algorithm RF-AUC-Cor based on random forest.It selects those features that are useful for identifying minority samples so that this paper optimizes the feature subsets in the feature selection stage and solves the class-imbalanced problem to improve the final classification effect.