1 Hector DUQUE, Creatis/LIRIS Johan MONTAGNAT, Creatis Jean-Marc PIERSON, LIRIS Lionel BRUNIE, LIRIS Isabelle E. MAGNIN, Creatis DM 2 : A Distributed Medical Data Manager for Grids. CREATIS, CNRS UMR 5515 (Centre de Recherche et d'Application en Traitement de l'Image et du Signal, Lyon, France) LIRIS (LISI), CNRS/INSA FRE 2672 (Laboratoire d'InfoRmatique en Image et Syst ₩ mes d'information, Lyon, France) BIOGRID'03, May 14th, 2003, Tokyo, Japan

2 Goals – To supply a Distributed System for the medical community able to deal with medical images and sensitive medical metadata – To provide hybrid content based queries and metadata queries over large databases of medical images (image processing) – To transparently use GRIDS in order to get advantage of its computing power and its massive storage capacity – To propose an arquitecture for developing this kind of systems (DSE: Distributed System Engines)

3 challenges – content-based queries over images – Image processing – Interfacing Grids and Image Archiving Systems – distributed queries – massive storage – high performance computing

4 Raid 5 DICOM client DICOM server DICOM hospital architecture DICOM data acquisition device (IRM) DICOM file = metadata + image DICOM

5 " Very large databases of images with a very long term storage " Data distributed over acquisition sites " Sensitive medical data (strong security constraints) " Read-only access to raw scanner data (replicas needed) " Keep track of data processing " Different kinds of users (patients, physicians and researchers) medical data constraints

6 applications " Hybrid queries to images's databases of Mammography and IRM cardiological sequences - metadata - parametric analysis (indexing) - content processing " Similarity queries Source image (e.g. thorax, MRI) Target images (thorax, MRI) Most similarLess similar Similarity measures

7 DSE hospital a distributed system = set of engines GRID messages DSE: Distributed Systems Engines

8 DSE0 DSE4 DSE3 DSE2 DSE1 0 message passing 0 distribution 0 Application 0 DM 2 0 transaction 0 user middleware applications semantic level DSE DSE is the definition of the multilevel architecture plust its implementation Distributed Systems Engines (multilayer architecture)

9 DSE0 network_in / ipc_out processes (NIIO) ipc_in / network_out processes (IINO) ipc_in processes only (IIO) DSE0 = set of processes and a MPK IINO NIIO IIIO IIO tcp/ip ipc ipc_in / ipc_out processes (IIIO) Message Passing Kernel (MPK)

DSE1 Tool Drivers (TOD) Request Drivers (RQD) server machine client machine Task Drivers (TKD) DSE1 = a set of drivers Transaction Drivers (TRD) a transaction is a set of tasks a task is a set of requests a request is a set of messages message passing kernel (MPK)

network IPC client machines server machines messages TRD RQD TKD TOD

METADATA TKD DM2 TRD DM2 image = {DICOM file} METADATA DM2 DM2 RQD CLIENT (DM2 API) Grid heart mammo Security TOD request comparable images to an imageID Usecase: a- A physician queries DM 2 to find all images comparable to an image of interest 1 METADATA RQD

METADATA TKD Grid TKD DM2 TRD DM2 image = {DICOM file} METADATA Grid DM2 CLIENT (DM2 API) Grid heart mammo Security TOD request comparable images to an imageID Usecase: a- A physician queries DM 2 to find all images comparable to an image of interest 1 b- he queries the DM 2 to get a list of images similar to a DM2image request similar images to an imageID 2 METADATA RQD DM2 RQD Grid RQD

Cache TOD DICOM TKD Grid TKD DM2 TRD DM2 image = {Dicom file} DICOM METADATA RQD Grid DICOM RQD Grid RQD DICOM Grid heart mammo Images TOD Security TOD GET (imageID) Usecase c- the Grid requests an image METADATA METADATA TKD

Results (Intel Pentium 4 CPU 1.70GHz, memory 512 MB) number of transactions * / minute: without cache cache (file) cache (image) (totally concurrent) (sequence) * transfer 10 files x.5 MB = 5 MB built 1 image of 5 Mb until 10*3 + 8 messages = 38 messages restricted to 10 channels with a DICOM server

in summary " The DM 2 architecture is designed for high performance and extensibility. " The proposed architecture (DSE) is appropriate for developing distributed medical systems. " The concept of multilayered architecture has allow us to deal with the design and development of a complex system as a distributed medical application is. " Computing GRIDS are very useful for medical image storage and analysis (distributed data, high processing power). " The DM 2 system allows the physicians to get secure access to their data and to issue hybrid requests over huge databases.

This work is partly supported by: / the IST European DataGrid project / the French ministry for research ACI-GRID project www-sop.inria.fr/aci/grid/public / ECOS-Nord Committee (action C03S02)