A non-relational approach for distributed medical imaging databases

摘要

Health institutions, particularly medical imaging laboratories, are producing an ever-increasing amount of data with each passing day. The easy access to new modalities, the propagation of diagnosis clinics, and the need of having the acquired data readily accessible imply the usage of very large storage archives and powerful processing systems to handle the querying of data. Additionally, healthcare institutions usually need to have the data stored across multiple repositories for redundancy in case of failure, or load balancing to improve performance. The non-relational technology tackles the scalability problem, allowing horizontal scaling. Particularly, the document-based databases as MongoDB or CouchDB, for instance, match the DICOM Information Model. In fact, DICOM objects may be directly converted to JSON objects, allowing the storing and indexing of the metadata in databases as the document-based ones. In this paper, we present an overview of the current state of the non-relational databases, discussing the strengths and weaknesses of this type of database in our use cases. Moreover, we focus on the implementation of such databases in the medical imaging use case. We present an implementation that was integrated into the Dicoogle open-source PACS archive. The results of such implementation are then revealed and discussed.