Medical Image Retrieval Based On Adaboost Classifier Learning

Medicine is one of the most important computer applications. With the development of computer hardware and software technology, digital images of the medical field have also increased rapidly. The medical image has become a fundamental tool, and occupies an irreplaceable position in the clinical diagnosis and treatment. Therefore, effective management and retrieval of medical image data is very necessary. The method of image retrieval is unable to describe the rich information contained in medical images, which confines the application of traditional text-based image retrieval technique in medical diagnosis and research. Therefore Content-Based Image Retrieval (CBIR) technology is proposed.Computer-aided diagnosis system is designed to help doctors reading medical images objectively,and providing some diagnostic opinions about suspicious lesions.We extracted many lung nodules from the LIDC database with pathological diagnostic classifications as the samples for CBIR. The experiment is designed aiming at the retrieval of the region of interesting (ROI) of Suspicious lung nodules. First, we extract the nodule image ROI region and nine lesions characteristic according to the XML annotation of LIDC database, calculate gray scale, texture, shape, size and other characteristics of the ROI regions,and constitute the nodule image feature database.,the nodule images can be classified according to the degree of malignancy. We proposed an Adaboost CART algorithm for image retrieval, combined classification and retrieval, The main idea of the algorithm is to use the CART decision tree as the Adaboost algorithm weak classifier training nodule image features and then use the Adaboost method samples from multiple sampling conducted multiple iterations of the expression and the classification accuracy is used as the basis to build the strong classifier.in the retrieval step, we retrieve the most similar images by calculate similarity in a sub class space. The proposed method can achieve a higher retrieval accuracy, and help doctors diagnose the malignancy of nodules.