12-16 April 2010
Uppsala University
Europe/Stockholm timezone

Grid-enabled Medical Image Analysis: From Jobs to Workflow-based Service

Apr 13, 2010, 4:40 PM
20m
Room X (Uppsala University)

Room X

Uppsala University

Oral Experiences from application porting and deployment Medical Imaging

Speaker

Dr Silvia Olabarriaga (Academic Medical Center of the University of Amsterdam)

Description

Medical image analysis requires large computational effort. The construction of “atlases” specific for each study of a specific brain disease is a good illustration: typically all images need to be aligned (registered) to each other. In this work we explore the Dutch Grid (part of EGEE) to implement a service that automates the construction of such atlases for Diffusion Tensor Imaging (DTI), facilitating usage by neuroimaging researchers. Initial experiments show that the service is robust and valuable for medical imaging research.

Impact

In this project we started running individual jobs with command-line tools and evolved into a workflow-based service that automatically handles grid computation and experiment recovery. This service can be used directly by end-users that are not experienced with this methodology for DTI atlas construction, neither have knowledge nor access to such computing resources. It facilitates the dissemination of a sophisticated data analysis process into the medical imaging community. After the initial results, we already observe an increasing interest for grid-enabled applications in our closest user community, which starts to consider more ambitious experiments (for example large parameter sweeps) that would not be performed in their usual infrastructure. Additionally, the knowledge and tools developed for this specific service can be now reused to more rapidly port other (life science) applications and services to the Dutch Grid. For example, large multi-hospital studies to brain diseases could also benefit immensely from distributed management and processing of the data by multiple researchers.

Detailed analysis

Initially we adopted conventional gLite command line utilities to run medical image analysis jobs on the grid, however these were considered too difficult by the researchers. Later we adopted the e-BioInfra platform developed in the VL-e Medical project, which is based on the MOTEUR workflow management system and the Virtual Resource Browser (VBrowser) front-end. Although this platform provided a high-level abstraction to describe and run large experiments on the grid, current limitations on the Scufl language force the separation of analysis into various workflows, imposing off-line actions and hampering usage by non-expert researchers and methodology dissemination. The developed service encapsulates all steps (workflows) needed to perform the complete image analysis, and also automates experiment monitoring and recovery. A web service interface was chosen for its implementation to enable usage in both from programmatic and interactive interfaces. A robot-certificate is available for users who do not own a grid certificate, facilitating usage from a web interface.

Conclusions and Future Work

The system is used in two on-going studies (schizophrenia and neuroscience) involving typically a total of 30K jobs, 2500 hours of CPU, and 30 GB of data. Other services are planned in the near future, as well as training activities to the PhD students at the AMC. We also see possibilities to improve the system for example by incorporating into workflow management the general functionality that is currently implemented by the service. Moreover, the security model needs to be revised to address privacy and usability requirements in medical applications.

Justification for delivering demo and/or technical requirements (for demos)

no demo

URL for further information http://www.bioinformaticslaboratory.nl
Keywords Medical Imaging, Neuroimaging, e-Science, Grid workflows, grid applications, MOTEUR, VBrowser

Primary authors

Dr Carsten Byrman (Academic Medical Center of the University of Amsterdam) Mr Matthan Caan (Academic Medical Center of the University of Amsterdam) Dr Silvia Olabarriaga (Academic Medical Center of the University of Amsterdam)

Co-authors

Mrs Angela Luyf (Academic Medical Center of the University of Amsterdam) Prof. Antoine van Kampen (Academic Medical Center of the University of Amsterdam) Dr Frans Vos (Academic Medical Center of the University of Amsterdam) Mr Jan Just Keijser (NIKHEF and BiG Grid) Prof. Lucas van Vliet (Technical University Delft, NL) Dr Tristan Glatard (CNRS Creatis Lyon, FR)

Presentation materials