An Exploratory Multimodal Analysis of Digital Handwriting and Spontaneous Speech Features to Distinguish Early Cognitive Impairment from Normal Aging

Abstract

Early screening for mild cognitive impairment (MCI) is crucial for managing neurodegenerative progression. This study explores the discriminatory power of two non-invasive digital biomarkers for cognitive decline detection: linguistics metrics extracted from spontaneous speech (via OpenAI Whisper) and spatio-temporal features of digital handwriting, collected from a balanced sample of 78 participants (39 MCI, 39 healthy controls). Speech-to-text linguistic analyses revealed that lexical variety emerged as the most sensitive marker of MCI, with patients producing fewer unique words and showing a tendency toward fewer total words and less dense and semantically rich discourse. From the handwriting analyses, specific tasks elicited task‑dependent anomalies in execution time, stroke production, and spatial organization. Overall, the proposed multimodal protocol demonstrates promising sensitivity for early cognitive decline detection, and future work will focus on expanding the dataset and developing advanced data‑fusion models.