Theoretical and Computational Neuroscience
Author: Ivan Caro | email: ivan.caro.strokes@gmail.com
Ivan Caro 1°, Enzo Tagliazucchi 2°, Adolfo Martín García 1°
1° Cognitive Neuroscience Center, Universidad de San Andrés, Buenos Aires, Argentina
2° Departamento de Física, Universidad de Buenos Aires and Instituto de Física de Buenos Aires (IFIBA-CONICET) Pabellón I Ciudad Universitaria, CABA, Buenos Aires, Argentina
Increased life expectancy in contemporary society implies a rise in aging-related neurocognitive alterations. These form a continuum ranging from subjective cognitive impairment (SCI) to mild cognitive impairment (MCI) and Alzheimers disease dementia (ADD). Automated speech metrics offer an objective, affordable, and scalable approach to predict those neurocognitive states. However, no study has used this approach to predict and anticipate neuropsychological and neural alterations along this continuum. We aim to generate a machine learning model that meets these goals. We have obtained data from 300 people over 70 years old, 60 of whom underwent a second evaluation two years later. All of them completed seven speech tasks (from which we will extract acoustic and linguistic features), neuropsychological tests (MMSE, IFS), and MRI/fMRI recordings (from which we will extract neuroimaging features). Using canonical correlation analyses, we will examine whether acoustic and linguistic features can predict (a) MMSE and IFS scores and (b) neuroimaging features. Then we will apply XGBoost regressors to obtain the models that optimally predict (a) and (b) along the SCI-MCI-ADD continuum. Finally, we will employ the same approach over the variables obtained two years later. The resulting model will favor the automatic detection of brain health profiles in older adults, offering an affordable and scalable alternative to the high costs and low availability of standard approaches.