MSc.Thesis Defense: Zobia Batool
Towards Reliable Alzheimer’s Diagnosis from 3D MRI Scans: A Generalized Approach
Zobia Batool
Computer Science and Engineering, MSc. Thesis, 2025
Thesis Jury
Prof. Erchan Aptoula (Thesis Advisor),
Assoc. Prof. Faik Boray Tek, Assoc. Prof. Öznur Taştan
Date & Time: 23rd June, 2025 – 11:00 AM
Place: FENS L047
Keywords : Domain Generalization, Alzheimer’s Disease, Contrastive Learning, Mor- phological Networks
Abstract
The thesis aims to address Alzheimer's disease detection from 3D MRI scans under a single-domain generalization setting, where a model is expected to generalize to unseen domains with potentially diverse imaging protocols, patient demographics, and class imbalance levels. Three distinct approaches are investigated. First, a pseudo-morphological augmentation strategy uses learnable modules to produce anatomically coherent, class-specific augmentations, integrated with supervised contrastive learning to extract robust and discriminative features. Second, the MixStyle framework is extended to incorporate higher-order statistical moments including skewness and kurtosis alongside traditional mean and variance, enabling enhanced feature perturbation and focus on disease-specific artifacts. Third, a Mixup-based augmentation method leverages distance transforms to spatially decompose MRI scans into layered components and recompose them from multiple samples, preserving structural integrity while promoting diversity. Extensive experiments across three benchmark datasets, namely NACC, ADNI and AIBL demonstrate that the proposed techniques substantially enhance the generalization capabilities of underlying models, thus providing a strong basis for creating reliable, domain-agnostic tools for early Alzheimer's disease diagnosis.diagnosis.