From Binary to Multiclass: Adapting the IT Transformer for Fine-Grained Alzheimer’s Disease Stage Classification

-, Joseph Jacob Anjilimoottil (2025) From Binary to Multiclass: Adapting the IT Transformer for Fine-Grained Alzheimer’s Disease Stage Classification. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (648kB) | Preview
Preview	PDF (Configuration Manual) Download (626kB) | Preview

Abstract

Classification of Alzheimer disease (AD) based on neuroimaging is a difficult task with limited data, handcrafted features, and a low interpretation level in deep learning models. This study explores the use of Interpretable Transformer (IT) architecture in four-class FMRI-based AD staging. The primitive binary model of PET-MRI fusion is adjusted to a single-modality MRI platform to enhance accessibility and decrease the reliance on the PET information. A set of image normalization and stratified sampling are used to construct a preprocessing pipeline that makes it possible to build strong model development. This IT model is benchmarked with four fixed CNNs, namely ResNet-18, VGG-16, EfficientNet-B0 and MobileNet-V2, to compare the differences in performance and computational needs. The IT model achieves an intermediate level of accuracy of 66 percent has compared to the base model, but more detailed global features are represented, which shows that this model requires additional tuning and bigger datasets. This paper illustrates that transformer-based architectures are viable to interpretable brain MRI-scan AD classification and offers the possibilities of further development.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Vikas UNSPECIFIED
Subjects:	R Medicine > Diseases > Brain - Diseases > Dementia > Senile dementia > Alzheimer's disease R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry > Neurology. Diseases of the Nervous System. > Psychoses > Dementia > Senile dementia > Alzheimer's disease Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence R Medicine > Healthcare Industry
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	07 Sep 2026 08:28
Last Modified:	07 Sep 2026 08:28
URI:	https://norma.ncirl.ie/id/eprint/9841

Actions (login required)

View Item