Ghita Berrada (2014-02-19 14:00 in ZI-2042)
When diagnosing patients, clinicians gather huge amounts of very diverse patient data, for example textual data such as symptoms' description, time series' data such as EEG or image data such as MRI. The data collected is not only extremely large but also scattered and hard to trace. This not only potentially compromises the security of the data but also presents risks for the patient in the form of delayed or overlooked diagnoses and unnecessary batteries of tests since the patient history is not made easily accessible to clinicians involved in the diagnosis process. Furthermore, the test data is highly uncertain and difficult (and expensive) to interpret. So there have been attempts to automate at least part of the test data interpretation process, with varying success. One of the biggest hurdles the automated interpretation methods face is that they are not easily comparable as they are usually developped on small, distinct datasets. Therefore, whether it be for clinical purposes or research purposes, medical data would benefit from being shared (at national level at least) between concerned parties (i.e clinicians, researchers and patients) for instance by building a medical data repository accessible to concerned parties. In the present talk, we will explain the challenges involved in building a medical data repository and outline some possible solutions to those challenges. We will in particular focus on the reduction of diagnosis uncertainty through user feedback.