Author ORCID Identifier
https://orcid.org/0000-0002-4601-8017
Date of Award
5-2026
Document Type
Thesis (Undergraduate)
Department
Psychological and Brain Sciences
First Advisor
Karen L. Fortuna
Abstract
Background
Alzheimer’s disease and related dementias affect over seven million Americans. Pre-clinical dementia symptoms include unsteady gait, diminished saccades, altered facial expression intensity, and impaired verbal fluency and prosody. Current diagnostics for dementia are expensive, invasive, and delay diagnoses by two years due to late patient presentation. This pilot study uses digital phenotyping to classify early-stage ADRD-spectrum status using subtle changes in early behavioral symptoms.
Methods
Participants (n=104) completed an enrollment survey, which included the Quick Dementia Rating System, and tasks in a smartphone app related to gait, eye tracking, speech, and smiling. Logistic regressions were constructed for each modality and all multimodal combinations after z-score normalization and unimodal feature selection. Area under the receiver operating characteristics curve (AUC) from 3-fold stratified cross-validation and 80% sensitivity and specificity values were calculated.
Results
Speech-related features had the highest unimodal AUC and provided the strongest marginal contribution to multimodal regressions. The best-performing multimodal model achieved a mean AUC of 0.840 (95% CI [0.727, 0.924]) and included speech, gait, eye-tracking, and demographic features. Features included speech motor and spectral variables, fixation stability, and the synchronization of inter-axis acceleration distributions.
Conclusion
Subtle changes in acoustic signals and fixational microsaccades, alongside the reorganization of cross-dimensional gait control, can be detected by smartphones, achieving promising classification of early ADRD-spectrum status. Resulting models present similar operating-point tradeoffs to the reported performance of established screening tools and have the potential to support efforts to reduce rural health disparities and improve early screening.
Recommended Citation
Tapiavala, Vedant N., "RealVision: Detecting Early-Stage Dementia via Digital Phenotyping" (2026). Psychological and Brain Sciences Senior Theses. 2.
https://digitalcommons.dartmouth.edu/pbs_senior_theses/2
