Initial Proposal Abstract: This project utilizes AI-driven analysis of remotely collected speech data to develop a non-invasive approach for detecting early signs of Alzheimer’s disease (AD). Since AD-related pathological changes begin decades before cognitive symptoms appear, traditional diagnostic methods relying on cognitive assessments often result in delayed intervention. While PET imaging and cerebrospinal fluid (CSF) biomarkers can detect AD pathology early, their high cost and invasiveness limit widespread clinical use. This study explores an accessible alternative by analyzing speech motor control, which is affected by early Aβ and tau deposition in brain regions crucial for speech production.

To achieve this, we will develop and validate AI models trained on speech acoustic and kinematic features to detect AD-related changes. In Aim 1, machine learning (ML) algorithms will identify atypical speech acoustic features—such as alterations in temporal and spectral properties—associated with lower Aβ42/Aβ40 ratios and higher p-tau 217 levels. Aim 2 will focus on validating and extracting kinematic features of facial movements during speech using monocular computer vision software, ensuring accurate and reliable tracking of subtle speech motor impairments. Aim 3 will integrate the validated acoustic and kinematic features into an AI model designed to predict individuals at high risk for AD with high accuracy, offering a scalable tool for early detection.

This study will utilize remotely collected speech and video recordings from 150 participants (ages 50–90), including individuals who are cognitively intact and those with mild cognitive impairment (MCI). Participants will also have comprehensive clinical, neuropsychological, and blood biomarker data, allowing for robust validation of speech-based AI predictions. By establishing speech motor features as sensitive indicators of AD pathology, this project aims to develop an AI-powered, non-invasive screening tool that facilitates early identification and intervention, ultimately improving diagnostic precision and personalized treatment strategies.

Outcomes:

• 
  Preprint: Detecting Preclinical Alzheimer’s Disease Risk in Cognitively Normal Adults Using Speech Acoustics: Validation with Plasma p-Tau217 and APOE-ε4 StatusJanuary 16, 2026
  Authors: Mehrdad Dadgostar, Lindsay C. Hanford, Maryam Tavakoli, Steven E. Arnold, David H. Salat, Tatiana Sitnikova, Pia Kivisakk Webb, Jordan R. Green, Hengru Liu, Brian D. Richburg, Mariam Tkeshelashvili, Marziye Eshghi Abstract INTRODUCTION We tested whether spontaneous speech acoustics provide a scalable digital marker of biologically defined Alzheimer’s disease (AD)… Read more: Preprint: Detecting Preclinical Alzheimer’s Disease Risk in Cognitively Normal Adults Using Speech Acoustics: Validation with Plasma p-Tau217 and APOE-ε4 Status
• 
  Publication: Kinematic correlates of early speech motor changes in cognitively intact APOE-ε4 carriers: a preliminary study using a color-word interference taskDecember 17, 2025
  Authors: Mehrdad Dadgostar, Lindsay C Hanford, Jordan R Green, Brian D Richburg, Averi Taylor Cannon, Nelson V Barnett, David H Salat, Steven E Arnold, Marziye Eshghi Abstract Introduction: Alzheimer’s disease (AD) is the most prevalent form of dementia and a major public health challenge. In the absence of a cure,… Read more: Publication: Kinematic correlates of early speech motor changes in cognitively intact APOE-ε4 carriers: a preliminary study using a color-word interference task