Linguistic biomarkers for the detection of mild cognitive impairment

A timely diagnosis of the prodromal stages of dementia remains a big challenge for healthcare systems: many assessment tools have been proposed over recent years, but the commonest screening instruments are largely unreliable for detecting subtle changes in cognition. The scientific literature contains a rising number of reports about language disturbances at the earliest stages of dementia, a clinical syndrome known as "Mild Cognitive Impairment" (MCI). Here we take advantage of these findings to develop a novel NLP method capable of identifying cognitive frailty at a very early stage by processing Italian spoken productions. This study constitutes a first step in the creation of an automatic tool for non-intrusive, low-cost dementia screening exploiting linguistic biomarkers. Our findings show that acoustic features (i.e., fluency indexes and spectral properties of the voice) are the most reliable parameters for MCI early identification. Moreover, lexical and syntactic features, grabbing the erosion of verbal abilities caused by the pathology, emerge as statistically significant and can support speech traits in the classification process.