Organiser: Dr Enrico H Gerding
Time: 27/10/2008 13:00-14:00
Brain tumour diagnosis usually requires the invasive procedure of taking of a biopsy. New advances in medical imaging have allowed Magnetic Resonance Imaging (MRI) to be combined with Magnetic Resonance Spectroscopy (MRS) to produce spectra for single or multiple voxels that contain tissue within a patient's body. The spectra allow surgeons to determine the chemical makeup of the tissue within that voxel. Research is underway that will help doctors to match the chemical makeup of the tissue with named tumour types (to make a diagnosis) and therefore predict the best form of treatment and formulate a prognosis. However, lack of ground-truth examples for certain tumour types or for certain patient groups means that classification of chemical descriptors is largely inaccurate.
The HealthAgents project (Jan 2006 - Dec 2008) developed a system to provide distributed data-marts of brain tumour cases such that the aggregation of cases from many hospitals allowed better classifiers to be built. The system provides decision support by allowing doctors to examine, anonymise and organise their patient data, as well as allowing them to find publications in the literature that are relevant to a particular case. It also allows experts to build classifiers using the distributed data-mart and then allows doctors to run those classifiers on new data. They can then examine the resulting classification and see how it compares to other cases and tumour-types. This functionality is all achieved over an agent network.
In our seminar we will describe the project and the software that was built, including the distributed framework and architecture. We will show some of the compromises we made such that the system might be used in a real hospital environment. We will also describe the security developments that allow hospitals to determine who on the network can access their data.
Any member of ECS with their name preceded by a sign has not given their permission for their information to appear on the public website.