Listening for Alzheimer's clues: machine learning analysis of multidomain speech features for cognitive impairment screening.
Обработка статьи · этапы, ошибки и управление
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Blazquez-Folch Josep, Calm Berta, Hinojosa-Calleja Adrián, García-Gutiérrez Fernando, Alegret Montserrat, Muñoz Nathalia, Cano Amanda, Fernández Maria Victoria, Miguel Andrea, Solivar Ariadna, De Rojas Itziar, Valenzuela-Seba Alejandro, García-González Pablo, Puerta Raquel, Olivé Clàudia, Capdevila-Bayo Maria, Muñoz-Morales Álvaro, Bayón-Buján Paula, Montrreal Laura, Orellana Adelina, Ortega Gemma, Sanz-Cartagena Pilar, Rosende-Roca Maitee, Cantero-Fortiz Yahveth, Gurruchaga Miren Jone, Tarraga Lluis, Butler Christopher, Montalban Xavier, Boada Mercè, Ruiz Agustín, Marquié Marta, Valero Sergi
Аннотация
METHODS: This study analyzed speech from 2,320 individuals spanning the cognitive spectrum-including those with subjective cognitive decline (SCD), MCI, and Alzheimer's disease dementia (ADD)-using three spoken tasks (∼3 min) and extracted multidomain features including acoustic, lexical, syntactic, and semantic features. Machine learning models were trained to classify cognitive status, predict amyloid, tau and neurodegeneration (ATN) biomarker positivity, and estimate scores across six neuropsychological domains.
RESULTS: Multidomain speech models achieved high performance in differentiating cognitive stages, with AUC values of up to 0.94 for SCD vs. ADD and 0.82 for SCD vs. MCI classifications. In biomarker prediction, the models yielded AUCs of 0.71, 0.74, and 0.73 for ATN classification, respectively. Speech-based models also showed strong correlations (up to 0.83) with cognitive function scores. Feature importance analysis revealed that verbal fluency measures were the most predictive. Explainability analyses indicated minimal dependency on age, sex, or education, supporting model fairness.
DISCUSSION: These findings show that multidomain speech features capture clinically and biologically relevant information across the AD continuum, enabling cognitive classification, biomarker prediction, and cognitive estimation. These results underscore the potential of speech analysis as a non-invasive, accessible tool for scalable cognitive screening and early detection of AD. These results underscore the potential of speech analysis as a non-invasive, accessible tool for scalable cognitive screening and early detection of AD.