Classifying Neurodegenerative Diseases from Selected Temporal EEG Electrodes: Towards Ear-EEG Applications.

Dementia is the seventh leading cause of death globally, with over 55 million people affected. As there is no cure to dementia, early detection is essential. Machine learning applied to EEG has been proposed for enabling early dementia detection, with multiple studies demonstrating successful binary classification of dementias. As a step towards the use of a more accessible tool, ear-EEG, this study investigates the extent to which two common forms of dementia, Alzheimer's Disease (AD) and Fronto-temporal Dementia (FTD), can be classified from temporal electrode EEG using a multiclass classification approach. EEG recordings from 88 participants (29 control, 36 AD and 23 FTD) at rest with eyes closed were used for classification. Features extracted from temporal electrodes (T3, T4, T5, T6, F7, F8) included sub-band power characteristics and band power ratios. Binary classification of healthy vs. AD achieved 81.8% accuracy, while multiclass classification of healthy, AD and FTD achieved 63.8% accuracy. The low accuracy in multiclass classification suggests challenges in differentiating between types of dementia using the proposed features.Clinical relevance- Our results demonstrate the potential of temporal EEG data in developing dementia classification models, highlighting the promise of more accessible tools such as ear-EEG for brain health monitoring. Further advancement of such tools and improvement of multiclass classification models could facilitate early detection of dementia, enabling timely interventions and improved patient outcomes.