Based On Multi-core Learning Mode Of Medical Image Classification

In last decades, with the development of computer technology and medicalimaging technology, medical images have become more diversified and increasedrapidly. How to effectively manage these image data becomes an urgent issue. Theclassification of medical images is an important means to achieve effective managementof medical images, but the traditional manual classification can not meet the rapidgrowing of medical images. Automatic medical image classification is to automaticallyextract medical image features and train classifiers to classify images, thereby increasesthe efficiency of management. In recent years, automatic medical images classificationhas attracted more and more attention from researchers.This thesis mainly studies the medical image classification methods which fusinginformation of different modalities. It first reviews the research works of medical imageclassification, then studies content-based and context-based medical imageclassification, and focuses on fusing different features by Multiple Kernel Learning(MKL), which gets promising results. The main contributions of this thesis are:(1) Different visual features are compared for content-based medical imageclassification. Specifically, Gabor, SIFT, ModSIFT, Tamura, LBP and GLCM featuresare explored in our study, and medical images are classified with Support VectorMachine (SVM). The best classification accuracies of various features are achievedrespectively and compared.(2) Context-based medical image classification is studied. The contexts used hereare captions of images from medical journals. Vector Space Model (VSM) and TF-IDFweighting method are used to represent the text information, and images are alsoclassified with SVM.(3) MKL is used to fuse various features for medical image classification. Medicalimage classification algorithms which fusing visual features with MKL and fusing bothvisual and text features are implemented respectively. The experiments show that fusingvisual and text features with MKL can significantly improve the accuracy of medicalimages classification.