Speaker: Jonathan Young
UCL Contact: Dominique Drai (Visitors from outside UCL please email in advance).
Date/Time: 11 Jul 12, 12:00 - 13:00
Conventional studies comparing patients of Alzheimer's disease with other forms of dementia and healthy controls are focused on finding statistically significant differences between groups, as measured by a variety of biomarkers. While these are of great help in understanding disease processes and directing further research, they are of limited applicability in a clinical setting as they provide little information about patients at the level of the individual. By learning a function to discriminate between instances of two classes such as patients and controls, modern machine learning techniques can address this issue and can be applied directly for diagnosis, or go beyond this and make predictions of the onset of serious dementia in those with relatively minor symptoms and possibly even in those with no symptoms at all. These latter problems are more difficult but are especially important to tackle as the new generation of drugs for AD promise to actually modify the disease process rather than mitigate the symptoms, but will only be effective if treatment begins at a stage early enough that a diagnosis of AD would not be made by conventional means.