EARLY DIAGNOSIS OF ALZHEIMER’S DISEASE BY LEARNING FROM MULTI-MODAL DATA FUSION /Aya Gamal Ali Ali

Supervisor: Dr. Mustafa Elattar

Thesis (M.A.)—Nile University, Egypt, 2024 .

"Includes bibliographical references"

Contents:
Acknowledgments.................................................................................................I
Abstract ...............................................................................................................II
List of Figures......................................................................................................V
List of Tables....................................................................................................VII
List of Equations..............................................................................................VIII
List of Abbreviation........................................................................................... IX
1. Introduction ......................................................................................................... 1
1.1Motivation ..................................................................................................... 1
1.2 Problem Statement ........................................................................................ 2
1.3 Thesis Outline and Summary of Contributions .............................................. 4
2. Background ......................................................................................................... 6
2.1Clinical Background ...................................................................................... 6
2.1.1 Alzheimer’s Disease ........................................................................... 6
2.1.2 How Alzheimer’s Disease Affects the Brain........................................ 7
2.1.3 Alzheimer’s Stage Determination and Its Progression....................... 12
2.2 Clinical Diagnosis Assessments ................................................................... 13
2.3 Challenges in Early Detection of Alzheimer’s Disease ................................. 15
2.4 Different Modalities for Diagnosis............................................................... 16
2.4.1 Magnetic Resonance Imaging (MRI)................................................ 16
2.4.2 Positron Emission Tomography (PET).............................................. 18
3. Related Work................................................................................................... 21
3.1 Datasets ...................................................................................................... 21
3.2 Pre-processing Pipeline................................................................................ 22
3.3 Input Management Approaches.................................................................... 27
3.4 Multimodal Data Fusion............................................................................... 28
3.4.1 Fusion Strategies............................................................................. 29
3.4.2 Analysis of Various Modalities in Relevant Studies...................... 30
3.4.3 Challenges with Multi-Modal Approach........................................ 31
3.5 Alzheimer’s disease Diagnosis using Traditional Methods............................ 32
iv
3.6 Alzheimer’s disease Diagnosis using Deep Learning Methods.................. 34
4. Automatic Early Diagnosis of Alzheimer's disease using Structural
MRI Images ...................................................................................................... 40
4.1 Dataset........................................................................................................ 40
4.2 Data preprocessing .................................................................................... 42
4.3 Augmentation ............................................................................................ 43
4.4 3D classification models ............................................................................. 44
4.4.1 3D Convolution neural networks.................................................... 44
4.4.2 Modified 3D DenseNet201 based transfer learning ....................... 44
4.4.3 Vision Transformers........................................................................ 45
4.4.4 Channel and Spatial Attention-based DenseNet ........................... 47
4.5 Evaluation Metrics .................................................................................... 47
4.6 Experiments details and Results............................................................... 49
4.6.1 Dataset and preprocessing setup.................................................... 49
4.6.2 Ablation study ................................................................................ 50
4.6.3 Experimental setting....................................................................... 53
4.7 Discussion................................................................................................... 57
5. Enhancing Early Alzheimer’s Disease Diagnosis through Multimodal
Data Fusion ...................................................................................................... 62
5.1 A Novel Diagnostic Model for Early Detection of Alzheimer’s Disease
Based on Clinical and Neuroimaging Features ......................................... 64
5.1.1 Methodology .................................................................................... 64
5.2 Deep Feature Fusion Framework for Alzheimer’s Disease Staging
using Neuroimaging Modalities................................................................... 75
5.2.1 Methodology .................................................................................... 75
5.2.2 Experiments Setup and Results..................................................... 79
5.2.3 Discussion ........................................................................................ 83
6. Conclusions and Future work .......................................................................... 86
6.1 Conclusion.................................................................................................. 86
6.2 Future directions........................................................................................ 87
Bibliography...........................................................................................................111

Abstract:
Alzheimer’s Disease (AD) is a severe neurodegenerative disease. Early
identification of AD is crucial for enhancing the quality of life for both
patients and their families. Unfortunately, most existing diagnostic
techniques rely on subjective assessments of behavioral and cognitive
symptoms, leading to misdiagnosis. In recent years, advances in medical
imaging technology have led to the emergence of neuroimaging-based
methods for the diagnosis of early AD stages. However, these methods
often rely on the analysis of a single modality, which may not capture the
full complexity of the disease. Multimodal data fusion has been proposed
as a promising approach to address this limitation by combining
information from different modalities. Our study is divided into two
phases, each with a distinct focus and methodology. In the initial phase,
our focus is on utilizing structural MRI images as a sole modality for
detecting AD as it is considered one of the most used modalities in AD
diagnosis. A novel methodology is introduced to achieve this objective by
discussing each component necessary to structure the entire system. Our
proposed methodology surpasses previous studies in distinguishing
between individuals with AD and cognitive normal (CN), AD and mild
cognitive impairment (MCI), and MCI and CN, achieving AUC scores of
95.09, 91.28, and 88.42, respectively.
Moving to the second phase, our work is extended to integrate different
modalities. Specifically, we start by incorporating extracted neuroimaging
and clinical features. This method of combining clinical and neuroimaging
features extracted from MRI and PET images yielded superior results,
achieving an AUC score of 97% for classifying between stable and
progressive MCI. Following this, we develop an automated multimodal
system to integrate MRI and PET images at an intermediate level of fusion,
enabling the automatic diagnosis of AD. We achieve an AUC score of
97.67% with an accuracy (ACC) of 95.24% for the AD vs CN task.
Keywords
Alzheimer’s Disease (AD), Multimodal Data Fusion, Neuroimaging Features, 3D
Image Classification, Structural MRI, Positron Emission Tomography, and Automated
Diagnosis.

Text in English, abstracts in English and Arabic