Tutorials
The role of the tutorials is to provide a platform for a more intensive scientific exchange amongst researchers interested in a particular topic and as a meeting point for the community. Tutorials complement the depth-oriented technical sessions by providing participants with broad overviews of emerging fields. A tutorial can be scheduled for 1.5 or 3 hours.
Tutorial on
Computer-Aided Diagnosis Machine Learning Based for Alzheimer’s
Disease on
Positron Emission Tomography Brain Images
Instructor
|
Mouloud Adel
Aix-marseille University
France
|
|
|
|
Brief Bio
Not Available
|
Abstract
This tutorial addresses advanced image processing and connectivity analysis techniques for
fluorodeoxyglucose positron emission tomography (FDG-PET) brain images, with applications
to neurodegenerative disease analysis. FDG-PET provides voxel-wise measurements of
cerebral glucose metabolism, offering a powerful imaging modality for studying brain function
at rest. However, the high dimensionality of PET images and inter-subject variability pose
significant challenges for automated analysis.
The tutorial presents a complete image-based computer-aided diagnosis (CAD) pipeline, from
preprocessing and anatomical normalization to feature extraction, selection, and
classification. Emphasis is placed on volume-of-interest (VOI)–based dimensionality reduction
and metabolic connectivity modeling using graph theory. Several graph construction
strategies, including mean-based and probability density function (PDF)-based methods, are
introduced and compared in terms of structural similarity, hub topology preservation, and
individual variability.
Both classical machine learning approaches and recent deep learning architectures are
covered. In particular, the tutorial introduces multi-view neural network models that exploit
axial, coronal, and sagittal PET slices to jointly learn complementary image representations
while preserving spatial information. Practical case studies focus on Alzheimer’s disease and
mild cognitive impairment, illustrating how image-derived connectivity features can be used
for classification and disease progression prediction.
By bridging image processing, graph-based modeling, and learning-based methods, this
tutorial provides ICIP participants with a comprehensive and up-to-date perspective on FDGPET
image analysis for brain connectivity and clinical applications.
Keywords
computer-aided diagnosis, graph, image processing, feature selection, machine learning
Aims and Learning Objectives
Participants will:
Understand FDG-PET as an image-based modality for brain functional analysis.
Learn preprocessing and normalization techniques specific to PET images.
Apply VOI-based image representations for dimensionality reduction.
Model metabolic connectivity using graph-based representations derived from
images.
Compare graph construction strategies and their impact on downstream analysis.
Implement classical machine learning and deep learning models for PET image
classification.
The tutorial is innovative in its image-centric integration of graph theory and multi-view deep
learning for metabolic connectivity analysis.
Target Audience
The tutorial directly addresses core themes:
High-dimensional image representation and reduction
Graph-based image modeling
Learning-based image classification
Medical image analysis with real-world clinical constraints
It provides transferable methodologies applicable beyond PET imaging to other imaging
modalities.
Prerequisite Knowledge of Audience
Basic image processing concepts (filtering, normalization, feature extraction).
Introductory knowledge of machine learning.
No prior expertise in PET imaging or neuroscience is required.
Detailed Outline
1. FDG-PET as an Image Processing Problem
2. Preprocessing and Image Normalization
3. VOI-Based Image Representation
4. Feature Extraction from PET Images
5. Graph Construction from Images
6. Feature Selection and Classification
7. Deep Learning for PET Image Analysis
8. Applications and Case Studies
9. Open Challenges and Future Directions
Recent advances in medical image analysis increasingly rely on connectivity-aware and graphbased
representations derived from images. FDG-PET provides a unique testbed for such
methods due to its metabolic nature and clinical relevance. This tutorial introduces graph construction from images, VOI-based feature learning, and multi-view deep networks, all of which align with current trends in image processing research.