Combining multichannel ERP data for early diagnosis of Alzheimer's disease

As the average age of our population increases, the prevalence of Alzheimer's Disease (AD), the most common form of dementia, has grown sharply. Current diagnosis of AD primarily uses longitudinal clinical evaluations and/or invasive lumbar punctures for CSF analysis, available only at specialized hospitals, which are generally outside of financial and geographical reach of most patients. We expand on our previous work and describe an ensemble of classifiers based approach that combines decision and data fusion techniques for the early diagnosis of AD using event related potentials (ERP) obtained in response to different audio stimuli. In this contribution, we specifically examine various feature set combinations, obtained from different EEG electrode locations and in response to different stimulus tones to illustrate the accuracy of such a system for AD diagnosis at the earliest stage on a clinically significant cohort size of 71 patients.