Research position open at I3S laboratory, Sophia Antipolis, France

Research engineer in medical image analysis in the context of high performance computing (HIPCAL project).

Context

The medical imaging community faces new challenges coming from (i) the amount of data to process (the yearly production of images in a radiological department sums up to tens of TB of data), (ii) the distribution of data (widespreading of data sources over the territory) and (iii) its heterogeneity (lack of well established standards), confidentiality and complex structure. Grid have been identified as a tool able to process large medical databases, enabling large scale applications such as epidemiology, statistical study over populations, medical simulation or research on rare diseases. This project aims at developing grid infrastructures tackling the needs of the medical imaging community and deploying applications demonstrating the relevance of grids for this challenging application area. The program of work will benefit from our former experience acguired in the context of the MEDIGRID, AGIR and EGEE (biomedical applications) projects in particular. Development will particularly target the Grid5000 grid infrastructure.

References.

HIPCAL project

HIPCAL is a research project funded by the "High Performance Computing and Simulation" program (CIS) of the French National Research Agency (ANR). HIPCAL studies a new paradigm (grid substrate) based on confined virtual cluster concept for resource control in grids. In particular, we propose to study and implement new approaches for bandwidth sharing and end to end network quality of service guarantees. The goal of this project is to explore an approach in a break with current services-oriented principles developed in grids to jointly enhance the application portability, the communication's performance control and their security. The proposal will be validated and evaluated at different scales on the Grid5000 testbed with biomedical applications, demanding in security, performance and reliability.

Description of work

In this project, our role is to develop medicl image databases analysis application use cases. The federation of full databases, as needed in many challenging medical studies require the federation of full databases over geographically distributed data sources (data virtualization). The setting up of such data and compute intensive applications involves (i) massive intensive (ii) secured, and (iii) efficient data transports between data centers and computing resources. Although grid computing tackles large image databases processing needs, the effective control of the data flow involved when applying multiple and concurent processings on data distributed over different sites remains an opened research area. Indeed, medical image databases represent gigabytes of data to process (a single image represents tens to hundred of megabytes). Moreover, data confidentiality needs to be preserved when data is transported from acquisition sources to processing sites and for storage of intermediate computing results to fulfill patients privacy requirements. Finally, The acceptation of computerized medical analysis procedures and their practical deployment in clinical environments requires limited response times (often in the order of minutes at most).

In particular, we envisage applications to anatomical atlases computations, medical image analysis algorithms validation, or image databases indexing. The funded engineer will work on defining precise use cases, especially taking into account network requirements. He/She will then deploy medical image analysis applications procedures on the experimental infrastruture set, based on previous works realized by the I3S laboratory and its partners. The envisaged data-parallel applications are:

  1. The indexing of medical image databases. Indexing medical images is needed to enable images queries from medical data stores for further processing and analysis, or to provide diagnostic assistance. Image indices enable data retrieval considering image content. Both the image indexing and the content-based data retrieval parts may require intensive computations and apply to very large data sets. These operation may benefit from a fully data-parallel implementation as different images are indexed independently from each others.
  2. The co-registration of medical data to produce anatomical atlases. The registration of medical images is a key basic algorithm needed for many image processing algorithms. Several medical images acquired with different imaging devices or at different instant in time are alined in a same spatial frame. This processing is compensating for patient position or acquisition geometry differences in order to compare different images. Co-registration is needed for building anatomical atlases which require the alignment of hundreds of patient image databases in a common frame. On-the-fly atlas generation is a challenging application to personalized medicine that require federation of large data sets and limited response time to match clinical needs. The data-parallel co-registration implementation will be based on standard rigid registration algorithms available in various toolkits such as ITK.
  3. The validation of medical image analysis algorithms. In the case of registration algorithm for instance, statistical procedures such as the so called bronze standards technique may be used to statisticaly quantify the maximal error resulting from a given registration algorithm. The larger the sample image database is, and the larger the number of registration algorithms to compare with is, the most accurate is the method. This procedure is therefore very scalale and it requires to compose a complex application workflow describing data transfer dependencies between processes. This application will be deployed through MOTEUR, a workflow manager developed at I3S and designed to optimize the execution of workflows of application codes by exploiting the data parallelism of such applications as well as the code parallelism intrinsic to the applications workflows.

Profile

Information

The hired engineer will work in an active research environment. He/She will have to adapt to the broad spectrum of technologies used. His/Her work will involve collaboration with a large national consortium involved. The work will require some travelling, to participate to meetings related to the project (in France) and scientific conferences (international).

Application

To apply, send your CV and recommandations to:
Johan Montagnat mail:johan@i3s.unice.fr
EPU tél:(+33) 492 96 51 03
930 route des Colles web:http://www.i3s.unice.fr/~johan/
BP 145
06903 Sophia Antipolis Cedex