175 | Automated speech analysis for the detection of mild cognitive impairment: A multidimensional neurocognitive approach

Disorders of the Nervous System

Author: Ivan Caro | Email: ivan.caro.strokes@gmail.com

Ivan Caro , Gonzalo Pérez , Joaquín Ponferrada , Franco Ferrante , Joaquín Valdés , Joaquín Migeot , Alejandro Sosa Welford , Agustín Ibañez , Andrea Slachevsky , Adolfo García

1° Cognitive Neuroscience Center (CNC), Universidad de San Andrés, Buenos Aires, Argentina
2° National Scientific and Technical Research Council (CONICET), Buenos Aires, Argentina
3° Department of Psychiatry, School of Medicine, Pontificia Universidad Católica de Chile, Santiago, Chile
4° Latin American Brain Health (BrainLat) Institute, Universidad Adolfo Ibáñez, Santiago, Chile

Detecting early markers of neurocognitive decline is vital in brain aging research. Recent works show that automated analysis of timing and word property patterns in verbal fluency tasks can reveal robust markers of Alzheimer’s disease. Here we examine whether these approaches can boost the detection of mild cognitive impairment (MCI). Fifty-two MCI patients and 54 healthy controls performed phonemic and semantic fluency tasks. Automated tools were used to extract timing (e.g., articulation rate) and word property (e.g., frequency, granularity) features from participants’ responses. These features were analyzed via a generalized linear model (GLM) and machine learning tools, compared with standard cognitive measures, and used for brain atrophy prediction. A GLM showed that word frequency, granularity, phonemic length, and imageability were significantly altered in MCI subjects, with no significant differences for timing measures. Machine learning analysis yielded robust classification (AUC = 0.77 ± 0.05), outperforming classification based on standard cognitive tasks. MCI participants showed atrophy of the left temporal pole, and their frequency and granularity patterns correlated with the volume of frontal and temporal regions, respectively. These results suggest that automated word property analysis in verbal fluency tasks can reveal robust markers of MCI, highlighting the utility of fine-grained language screenings to better characterize brain (dys)function in the elderly.