4. RELATED WORK
The management of medical data is an important domain of research. The Picture Archiving Systems (PACS) systems are currently used in most of medical centers. These systems are very expensive and propose a low expressiveness (per-defined queries over certain attributes). Additionally they do not cope with the heterogeneity problem, since they mostly use a relational database that stores all heterogeneous attributes in a blob-like datatype without any ability to interrogate them.
Oracle 11g has developed DICOM support [20], which adds a new Java-Class-like data type ORDicom; thereby any column of a table can hold DICOM content. Oracle provides indexing and compression techniques. Their approach stores each DICOM file in a separate object, so there will be a lot of data redundancy. This model increases the storage space and reduces the performance especially when using certain DICOM-specific methods.
Another interesting example is eDiaMoND [21], a grid-enabled medical imaging database that employs an object-relational approach to the storage of DICOM files. EDiaMoNd supports only three modalities and restricts users to a set of pre-determined queries. This system is designed over the grid (limited number of dedicated severs); therefore it is not suitable for a huge infrastructure of unreliable machines (such the cloud).
Exploiting the power of the cloud for medical domain has been proposed by Xbase [18], a XML-based appliance that uses RDBMS and Hadoop for storing healthcare records. Xbase is a record-oriented system that manages medical information with record IDs. This system does not support DICOM files which have a complex structure and evolutive/heterogeneous schema.
Other commercial cloud-enabled medical systems are DicomGrid (www.dicomgrid.com) and IronMountain (www.ironmountain.fr). There are no documentations or research papers about them.
5. CONCLUSIONS
In this position paper, we propose cloud-enabled hybrid database for the management of medical data. The challenges in this context are due to the high heterogeneity and huge volumes of DICOM files. For that we propose a new architecture that could: (1) provide ease of use, extensibility, high performance and ad-hoc queries over DICOM files and (2) get benefit of the elasticity, billing by use and scalability of the cloud.
The next objective is to implement our complete prototype in the cloud and validate it for real medical applications. We plan to achieve a high level of QoS that allows querying large amounts of data via different types of computing devices. The security of medical data over the cloud could be an interesting future work. Additionally, some optimization (e.g. materialized views, cache manager) should be rethought for our particular structure.
6. ACKNOWLEDGMENT
This work is in collaboration with the Yansys Company /Syseo® project. It is sponsored by CR d’Auvergne, SGAR Auvergne and the ANR under grant SYSEO ANR-10-TECSAN-005-01.