Research And Implementation Of Medical Imaging Storage And Retrieval System Based On Hadoop

Medical imaging examination is more and more reliable and efficient as the continuous improvement of medical informatization, constant development of image processing technology and frequent upgrade of image acquisition device. As a result it has become a primary reference in modern diagnosing process. To store and let doctors access to patients' diagnostic imaging, the modern hospitals use the joint work of Radiology Information Syestems(RIS) and Picture Archiving and Communication Systems(PACS). RIS handles tasks like managing patients' reservation and medical records. PACS is responsible for storing medical imaging and enable doctors to have access to medical images. However, with the establishment of regional imaging sharing and cooperat ion platform in recent years, traditional online-LAN-offline storage strategy which PACS is using can't satisfy the storage need of gigantic diagnostic image data. Meanwhile, PACS is using relational database, and this kind of database has poor searching and retrieving performance when the imaging data is huge. What is more, PACS doesn't have backup scheme and fault tolerance.In this thesis we analyzed the feasibility and the advantages of HDFS storing gigantic medical imaging data, and the problems to be solved. We did research on how to effectively perform precise retrieving and range retrieving in Hadoop, just like what PACS can do in its comfortable data size range. The main research tasks include: research on how to effectively store massive medical imaging with HDFS and how to improve its performance. We put forward a medical imaging data storage optimization algorithm which combined the image files that had the same Series UID into one single Sequencefile. And our optimization algorithm solved the problem that when HDFS stored a large number of medical imaging, the memory usage of the namenode was too much. At the same time, The optimization algorithm significantly improved the image reading and writing efficiency; To achieve the precise retrieval of PACS, we put forward the Hbase multilevel indexing method based on bloom filter to improve the image retrieval efficiency; In order to solve the problem that the range search couldn't be directly performed in Hbase(which was an easy task in traditional PACS), we established a mapping model which could transform a structured query on the structured data to a Map Reduce task on the unstructured data. Finally, we fulfilled the storing and retrieval system with Hadoop clusters. The test results were positive, the medical image storage optimization method could effectively reduce the memory consumption of namenode. In conclusion, the experiments show that our system respond faster than PACS in both precise retrieval and range retrieval, which proves that our system have a better retrieval efficiency and better performance.