Screening of Alzheimer’s Dementia up to 12 Years ahead from Conversational Speech of ILSE Study

Abstract:
Alzheimer’s disease (AD) is an incurable neurodegenerative disorder and successful symptomatic therapy requires early diagnosis. However, when the diagnosis is made by clinical screening, AD has already impaired the patient’s cognitive abilities and the optimal time point for early therapy has passed. Therefore, early diagnosis of AD is crucial. Spoken language skills are strong biomarkers for detecting dementia, as they are affected in the early stages of cognitive impairment. In this work, we aim to conduct predictive screening, i. e., predict future cognitive diagnosis, using the longitudinal conversational speech corpus ILSE. We extract acoustic and linguistic features from the speech of current time measurement. We apply non-parametric significance test for group differences between healthy and AD samples in predictive screening and analyze the distribution of features in AD screening. We train models for predictive screening of AD. Our classifier achieves an Unweighted Average Recall of 83.8% (in 5 years) and 82.5% (in 12 years).