Non-Invasive Alzheimer's Disease Diagnosis using Attention-based Deep Learning Models assisted by IoMT-driven Data across English and Spanish Languages
Year: 2025
Type: preprint
Abstract: Alzheimer's disease (AD) is a progressive neurodegenerative disorder marked by cognitive and linguistic impairments, including subtle but measurable changes in speech. While current diagnostic methods rely heavily on neuroimaging and invasive biomarker analysis, these approaches can be expensive, resource-intensive, and inaccessible in many settings. This research explores a noninvasive, speech-based methodology for detecting Alzheimer's disease using machine learning applied to acoustic features extracted from patient speech. In this study, spontaneous speech recordings-centered around the standardized "Cookie Theft" picture description task-were collected from both Alzheimer's patients and healthy control subjects. Each audio file was segmented into 5-second overlapping windows (with a maximum 50% overlap) to ensure temporal resolution of local acoustic variations. The extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS) was employed to extract meaningful paralinguistic features from these speech segments, capturing critical indicators such as pitch variability, articulation, and voice quality. To reduce dimensionality and improve classifier performance, ANOVA (Analysis of Variance) was used for statistical feature selection. Several machine learning algorithms-including Support Vector Machines (SVM), Random Forests (RF), and Logistic Regression-were trained and evaluated on the feature vectors to determine their ability to accurately classify speech samples as AD or healthy control. Performance metrics such as accuracy, precision, recall, and F1-score were used to compare model effectiveness. The results demonstrate that speech-derived acoustic features, when appropriately segmented and processed, can serve as robust indicators for early-stage Alzheimer's detection. This work supports the feasibility of using voice as a non-invasive, cost-effective biomarker for AD, with potential applications in remote monitoring and large-scale screening programs.
Source: HAL
Authors: Javed, Rahim, Umar Khan, M, Azam, Maleeha
Institution: COMSATS University Islamabad
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W4415182955