Dementia Screening Using Automated Speech Analysis: A Pilot Study
Funded Grant
Overview
Affiliation
View All
Overview
description
Project Summary Early detection of Alzheimer's disease (AD) is vital if treatments are to be most effective, yet for myriad reasons, a staggering number of older Americans do not receive diagnostic services. Recent technological breakthroughs show considerable promise to identify biomarkers for AD, but nonetheless still require costly and often invasive in-person testing. What is needed is a new approach that can rival current technologies in terms of diagnostic accuracy, while also permitting wider accessibility and economy. The ubiquitous telephone will be the vehicle for this approach in the proposed study. Telephone-based diagnostic screening tests have emerged as a viable alternative to in-person testing, and appear well tolerated and cost-effective. These tests are effective at detecting frank dementia, yet lack sensitivity to detect mild cognitive impairment (MCI) a likely risk factor for AD. Deficits in language function, particularly semantic knowledge, are characteristic of AD, and these changes may occur decades prior to the emergence of overt symptoms. Research from our group has demonstrated that automated speech analysis can quantify aspects of semantic knowledge in psychiatric samples and can provide similar information to human raters in terms of diagnostic accuracy. Hence, our goal is to determine the acceptability, sensitivity and specificity of an automated speech analysis approach to the detection of AD and MCI from speech samples elicited via telephone. Such an approach has the potential to greatly outperform current screening methods, not only in terms of sensitivity to cognitive decline, but also accessibility, cost, and user satisfaction. If successful, this project could form an integral part of an adaptive, machine learning decision support system to accurately predict an individual's dementia conversion threshold and trajectory of decline.
