Validation of linguistic markers in biologically defined prodromal AD and testing of their validity in differential diagnosis

The early identification of prodromal Alzheimer’s disease (AD) is a pressing clinical challenge, particularly as disease-modifying therapies emerge. This thesis investigates whether semantic and linguistic features derived from speech can contribute towards the diagnosis of mild cognitive impairment (MCI) due to AD. It also evaluates the feasibility of remote, automated cognitive assessment using a virtual clinical agent.
Two primary studies were conducted. The first involved the development and testing of CognoSpeak, a fully automated system that administers verbal fluency and interview-style tasks. CognoSpeak demonstrated high classification accuracy, distinguishing patients with neurodegenerative conditions from healthy controls and those with functional cognitive symptoms.
The second study explored advanced semantic features extracted from verbal fluency and free speech samples, comparing their diagnostic utility to that of traditional cognitive assessments and structural imaging biomarkers, including hippocampal volume and voxel-based morphometry.
Key findings indicate that specific semantic language features correlate significantly with hippocampal volume and contributed towards accurate prediction distinction of prodromal AD from controls. When combined with cognitive scores and imaging data, language features enhanced classification performance. The research supports the hypothesis that semantic markers derived from speech can serve as valid indicators of early cognitive decline and, when embedded in automated systems, offer scalable, accessible tools for screening and monitoring.
These findings underscore the diagnostic value of language features and their potential integration into clinical pathways. The thesis concludes that language-based tools are not only feasible but also clinically meaningful adjuncts in the early detection of Alzheimer’s disease.