Towards an Automated Portable Electroencephalography-based System for Alzheimer?s Disease Diagnosis

Alzheimer?s disease (AD) is a neurodegenerative terminal disorder that accounts for nearly 70% of dementia cases worldwide. Global dementia incidence is projected to 75 million cases by 2030, with the majority of the affected individuals coming from low- and medium- income countries. Although there is no cure for AD, early diagnosis can improve the quality of life of AD patients and their caregivers. Currently, AD diagnosis is carried out using mental status examinations, expensive neuroimaging scans, and invasive laboratory tests, all of which render the diagnosis time-consuming and costly. Notwithstanding, over the last decade electroencephalography (EEG specifically resting-state EEG (rsEEG), has emerged as an alternative technique for AD diagnosis with accuracies inline with those obtained with more expensive neuroimaging tools, such as magnetic resonance imaging (MRI), computed tomography (CT) and positron emission tomography (PET). However the use of rsEEG for AD diagnosis presents two major disadvantages: (i) the reliance on “artifact-free” segments, which are manually and meticulously selected by expert clinicians, and (ii) the need for research- and medical-grade EEG devices often with 16+ electrodes, thus making them hard to transport and expensive to fund, specially for low- and middle-income countries. In this doctoral thesis, we present the steps towards the development of an automated portable and low-cost rsEEG-based system for AD diagnosis. To achieve this goal, three main innovations have been developed. First, we explored the effects of several fully automated artifact handling (AAH) methods on rsEEG signals and reported their advantages and disadvantages for automated AD diagnosis. The outcome of this exploration showed the AD diagnosis based on AAH-enhanced rsEEG signals is inline with the performance obtained with “artifact-free” EEG data, suggesting that expert human intervention and the discard of EEG data (due to artifacts) can be avoided. Secondly, we evaluated and compared the use of lower-density (7-channel) EEG devices for rsEEG eyes-open and eyes-closed experimental protocols for AD diagnosis. By comparing the diagnosis performance obtained with the low-density EEG devices to the one obtained with a typical medium-density (20-channel) system, we found that the reduction of channels diminishes the classification performance. Lastly, we showed that conventional features proposed in the literature are not only sensitive to artifacts, but rely on information extracted from EEG frequency subbands that are suboptimal for AD diagnosis. As such, we proposed a new set of features based on the modulation spectrogram approach. Experiments showed the proposed features being robust to artifacts, thus bypassing the need for AAH algorithms, as well as providing more discriminatory information not only for AD diagnosis, but also for disease progression monitoring. These new benefits are invaluable for the development of low-cost portable devices and it is hoped that the insights presented herein can be used to improve the quality-of-life of individuals living with AD, their families, and caregivers.