Author Type

Graduate Student

Date of Award

Spring 4-27-2026

Document Type

Dissertation

Publication Status

Version of Record

Submission Date

May 2026

Department

Computer and Electrical Engineering and Computer Science

College Granting Degree

College of Engineering and Computer Science

Department Granting Degree

Electrical Engineering and Computer Science

Degree Name

Doctor of Philosophy (PhD)

Thesis/Dissertation Advisor [Chair]

Hanqi Zhuang

Abstract

Alzheimer’s disease (AD) is the most common form of dementia and is characterized by progressive cognitive decline and memory impairment. Frontotemporal dementia (FTD), the second most prevalent form, primarily affects the frontal and temporal lobes and often leads to changes in personality, behavior, and language. Due to overlapping clinical symptoms, FTD is frequently misdiagnosed as AD. Electroencephalography (EEG) offers a portable, non-invasive, and cost-effective method for studying brain activity; however, its diagnostic utility for differentiating dementia subtypes is limited by signal complexity and noise. In this dissertation, I propose an EEG-based feature extraction framework that leverages deep learning to identify AD and FTD and to estimate disease severity. The analysis reveals elevated delta-band activity in frontal and central regions as potential biomarkers associated with these neurodegenerative conditions. By combining temporal and spectral representations of EEG signals, the proposed framework integrates a Convolutional Neural Network (CNN) with an attention-based Long Short-Term Memory network (aLSTM). The model achieves over 90% accuracy in distinguishing AD and FTD from cognitively normal (CN) individuals and predicts disease severity with relative errors below 35% for AD and approximately 15.5% for FTD. Differentiating AD from FTD remains challenging due to shared neural patterns; however, incorporating a feature selection step improves specificity, increasing AD–FTD classification accuracy from 26% to 65%. Building on these findings, a two-stage classification strategy is developed for AD, CN, and FTD by first identifying CN individuals and then separating FTD from AD, resulting in an overall classification accuracy of 84%. These results demonstrate the potential of integrating temporal and spectral EEG features with deep learning to improve the identification and characterization of dementia subtypes, offering a promising non-invasive approach that may support clinical screening and monitoring of neurodegenerative disease progression.

Download

Share

COinS