Towards detecting Alzheimer’s disease with central auditory and physiological measures

Abstract

Alzheimer’s disease (AD) is the leading cause of dementia, with cases expected to rise substantially as the aging global population. Cognitive decline in AD often begins with subtle symptoms, such as difficulty understanding speech in noisy environments, before advancing to memory loss and severe impairments. Early diagnosis is critical, as it enables timely clinical care and creates opportunities for lifestyle changes that may delay or prevent dementia. However, current diagnostic methods, including neuropsychological assessments and neuroimaging, are expensive and unsuitable for large-scale screening, emphasizing the need for alternative accurate, accessible, and non-invasive approaches that could complement the traditional diagnostic methods. Recent research highlights central auditory, speech, and physiological functions as promising biomarkers for early detection of AD. Central auditory processing deficits, such as reduced ability to understand speech in noise, are linked to cortical regions affected by AD and can be evaluated using accessible and non-invasive tests. Similarly, speech and lexical changes along with biosignals, such as heart rate and pupil diameter, may provide potential additional diagnostic insights. Hearable devices offer a practical platform for capturing these multimodal signals using in-ear microphones due to the occlusion effect that amplifies low-frequency signals when the ear is occluded. However, there is a lack of multimodal datasets collected from individuals with AD and mild cognitive impairment (MCI) using a hearable. This thesis addresses this gap by introducing the Gaze and Auditory Response Database for Evaluating Neurocognitive Impairment and Alzheimer’s disease (GARDENIA), a multimodal dataset collected from 20 participants, including individuals with AD, MCI, and cognitively unimpaired controls. The dataset contains central auditory processing tests, a picture description task, in-ear biosignals, and eye-tracking data. Results indicate that cognitively impaired participants performed worse than controls across all central auditory tests, with dichotic tasks showing the strongest predictive value. Also, they demonstrate the capability of hearables to successfully administer central auditory tests. However, the analysis of heartbeats extracted from in-ear signals using Tempbeat highlighted the need for robust bio-signal extraction algorithms that are not affected by jaw movements. Overall, this work presents GARDENIA as a valuable resource for researching and developing tools for bio-signal processing that can help detecting AD.

Date	8 Dec 2025
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Rachel Bouserhal (Supervisor) & Christopher Niemczak (Co-supervisor)