Departmental seminar - Ji Hyun KoWhen:
Title: How can we improve early diagnosis of Alzheimer’s disease using machine learning and brain imaging?
Alzheimer’s disease (AD) is the most prevalent cause of dementia that affects >747,000 Canadians as of 2011. Number of disease modifying therapies and preventive interventions are under development. Early detection of AD may benefit patients and caregivers by providing a better chance of benefiting from treatment, relief from anxiety about unknown problems, more time to plan for the future and early access to psychosocial education for both patients and caregivers.
The current consensus is that patients with mild cognitive impairment (MCI), a state referring with cognitive deficits but not serious enough to interfere with patient’s independence, are at higher risk of developing AD or other dementia than age-matched normal controls. Therefore, routine tests of mental/neuropsychological performance and interviews by physicians may be used for early detection of AD. However, only 20-40% of amnestic MCI patients eventually convert to AD. The use of [18F]fluorodeoxyglucose (FDG) positron emission tomography (PET), a brain imaging technique that estimate regional glucose metabolism, has been widely accepted to aid physicians to complement clinical diagnosis, but the subjective impressions of FDG-PET readings are often found to be equivocal. Therefore more objective quantification of risk estimate using FDG-PET is highly demanded.
Here, we are investigating the use of scaled subprofile modeling (SSM), a form of principal component analysis (PCA), to automate the FDG-PET readings. Early preliminary data shows promising results. We are currently trying to employ machine learning algorithm to further improve the performance of automated early diagnosis of AD.Complete Seminar List