Premium
TD‐P‐020: TOWARD AN AUTOMATIC SPEECH‐BASED DIAGNOSTIC TEST FOR ALZHEIMER'S DISEASE
Author(s) -
Schaffer David,
Sadeghian Roozbeh,
Zahorian Stephen A.
Publication year - 2018
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2018.06.2036
Subject(s) - computer science , classifier (uml) , speech recognition , feature extraction , punctuation , feature (linguistics) , test set , artificial intelligence , set (abstract data type) , philosophy , linguistics , programming language
that present to primary care. Based on such data, it may be possible to determine the future development of Alzheimer’s Disease (AD) in a Pre-clinical Alzheimer’s diagnosis (PCD) with cognitively normal but high risk individuals, 18-24 months before any symptoms develop of cognitive impairment. Early treatment of these cognitively normal high-risk persons could prevent or delay the occurrence and severity of AD. However, such models still need biological samples and neuroimaging methods, whereas existing digital biomarkers could also provide individually tailored prognostic information with greater and feasible scalability. Methods: Such a novel digital biomarker, called Neuro-Motor-Index (NMI), was tracking micro-errors and reaction times for a period of 24 months, while the participants were interacting with a complex everyday activity task. Based on the tasks carried out by the user, a behavioral signature was computed based on a set of activity parameters (kx) with a high likelihood of corresponding to functional decline, e.g. omissions of the dual-task interactions between start and finish, perseverations of incorrect dual-task interactions, reaction time of ‘dual-task’ interactions, etc. The total sample included 260 (47%) males and (53%) females from the community, with a mean age of 76.1 years (SD1⁄47.3) and a mean self-reported education of 12.3 years (SD1⁄41.5). Results:A non-linear regression (Machine Learning) model was used to compute the functional impairment score. To derive cut-off values for the optimized operating point achieving desired sensitivity and specificity in classification of healthy and pre-dementia subjects, a discriminative analysis was carried out using receiver operator curves. Compared to the golden standard, NMI was then used to make a classification decision for MCI due to AD with a diagnostic accuracy >94%. Conclusions:The main goal of the present pilot is to evaluate NMI as a valid and meaningful digital prognostic biomarker at the primary care setting, which can detect subtle micro-errors and functional deficits in PCD patients at least 24 months before any symptoms develop of cognitive impairment.