Services for Interaction
| will provide services to move Grids from exclusive batch-oriented to general purpose, including interactive jobs. This is essential for any activity that involves humans, such as on-demand data product generation, exploration of datasets and visualization. For example, in high-degree-of-freedom optimisation problems such as non-linear image registration or segmentation of noisy datasets, computational steering can be used to help algorithms avoid local minima. These interactions may trigger asynchronous computation or data retrieval operations.
The general issue is to integrate the response-time constraints into the grid scheduling protocol stack.
Two classes of services will be considered.
Application-level schedulers. Many interactive applications with parallel components require fine grain dynamic scheduling. Application level schedulers will avoid the traversal of the Grid information systems for executing each and every small parallel task. A related question is the appropriateness of the service concept and implementations (and more specifically Web Services) to the applications-level schedulers.
Soft real-time brokers. Advance reservation can be considered as a basic option for true real-time situations (e.g. intra-operative imaging). For less critical situations (e.g. algorithm research, or surgical planning), the appropriate goal is soft real-time schedulers, defined by a high probability of QoS with admission control. The algorithms developed in this field, either for processor scheduling or network control, can be transposed at the level of grid meta-schedulers.
|
|
Medical Data Management
| is not limited to medical images but encompasses a rich amount of metadata which semantics is meaningful. The medical data manager needs to easily identify data sets according to clinically relevant research criteria that involve both metadata and image contents. Manipulating real medical data implies strong security constraints in order to preserve patients privacy. The goal for this task is to provide a service for data manipulation (data storage and data query) dedicated to medical data distributed over the grid resources.
In the MEDIGRID project, a medical images and associated metadata manager is being developed.
In this project, we intend to study the use of medical data stored in such a medical server for the image processing algorithms developed. It is addressing the complexity of the DICOM standard and the lack of standardisation for Radiological Information Systems deployed in hospitals. It will take into account the requirements of medical applications, particularly concerning the semantics of queries and preloading of data sets for computations on the grid. The problems arising when accessing concurrently data on a server and transferring huge amounts of data (3D images or even sequences of 3D images) will be addressed |
|
Medical Data Protocols
|
Storage and transmission of the tremendous volume involved in medical applications can easily overflow existing systems. This task will address compression services and streaming protocols focused on medical images
Compression algorithms
Lossless algorithms are preferred for diagnosis purposes. However, in some cases a non-optimal quality of information is sufficient, and lossy algorithms can be thus an efficient response to storage or transmission problems. For example, an optimal quality is required only within a region of interest (ROI).
Protocols
The key idea is to focus the resource usage on the useful data, while the Grid system can exploit the delays related to human interaction or remote processing in the background to refine the information finally delivered.
The general goal of this subtask is to define an analogue of the streaming protocols used for remote access to multimedia data, which can be termed as medical streaming protocols.
Two aspects will be considered.
Coupling compression and transfer.
Fast encoding and decoding should be privileged for data transmission while longer but more efficient compression algorithms are suited for long term storage. We will therefore study a compression scheme allowing an overlap between compression and communication and the dynamic adaptation of the compression ratio depending on the machine speed and the application usage.
A progressivity factory.
This task considers mainly the fine grain access to individual medical images in the context of interactivity. Algorithms targeting the retrieval and navigation inside selected views (e.g. medical windowing) are required for remote interactive access.
|
|
 |
