Reseach Of Medical Imaging Storage And Retrieval System Based On Hadoop And Multi-Level Indexing Technology

Along with the continuous development of digital imaging, image compression and computer network technologies, the IT applications in the field of medical imaging has become increasingly widespread,more and more medical imaging equipments have been introduced to hospitals. As a basic medical diagnostic method, medical imaging has become more and more important to doctors. However, due to the high price of medical imaging equipment, professional nature of medical imaging diagnosis and the difficulties in medical images’ disaster recovery and backup, the primary medical instutions’ imaging diagnostic power is very weak. The high cost and uneven distribution of medical resources has exacerbated the social medical problem.Therefore, the public have an urgent need for the medical image sharing platform.We need to have a storage and retrieval mechanism for massive medical images.Althongh the traditional medical image storage architectures(FC SAN, NAS, etc.) can basically meet the needs of a single-point hospital, they have many drawbacks such as high costs, bad compatibility and bad suitability for data-intensive computing, so it hardly cope with the massive medical images which is measured in PB. Hadoop is a popular open source distributed computing framework for large data, it has many advantages in efficiency, reliability and scalability. We can try to solve the problem of storing and retrieving massive medical images with Hadoop.This thesis is an advance research project of Sichuan Province Regional Medical Imaging Center. According to the actual needs, we designed a hadoop-based system achitecture to store and retrieve massive medical images. And then, we implement the interfaces of Hadoop I/O to design the specific classes for Dicom files.We have also designed a MSFile model and MapReduce program,which is based on Hadoop Mapfile and Hadoop Sequencefile, to solve he problem of low efficiency of retrieving small dicom files. As HDFS is not suitable for randomly reading a single record, we designed a multi-level indexing system, which is based on HBase database and MSFile model, to meet the needs of quick retrieval. Finally, we have done a series of experiments to verify the correctness of the system.