Related Experiment Video
Updated: Feb 16, 2026
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.9K
Deep Learning based Classification of FDG-PET Data for Alzheimers Disease Categories
Shibani Singh1, Anant Srivastava1, Liang Mi1
1School of Computing, Informatics and Decision Systems Engineering, Arizona State University, Tempe, AZ, USA.
Proceedings of Spie--The International Society for Optical Engineering
|December 22, 2017
Summary
Fluorodeoxyglucose (FDG) positron emission tomography (PET) shows promise for early Alzheimer's disease (AD) detection. Deep learning models effectively classify AD diagnostic categories using FDG-PET data, with max pooling outperforming mean pooling.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Biomarker Discovery
Background:
- Fluorodeoxyglucose (FDG) positron emission tomography (PET) is a functional imaging technique that measures cerebral glucose metabolism.
- FDG-PET provides unique metabolic information valuable for detecting Alzheimer's disease (AD) even in presymptomatic stages.
- The computational effectiveness of FDG-PET for classifying diverse AD diagnostic categories remains underexplored.
Purpose of the Study:
- To develop and evaluate deep learning models for accurate classification of Alzheimer's disease diagnostic categories using FDG-PET data.
- To investigate the efficacy of dimensionality reduction techniques, specifically probabilistic principal component analysis (PPCA) with max-pooling and mean-pooling, for FDG-PET data.
- To demonstrate the potential of FDG-PET as an effective imaging biomarker for AD through advanced computational analysis.
Main Methods:
- Utilized a dataset from the Alzheimer's Disease Neuroimaging Initiative (ADNI) including cognitively unimpaired (CU), mild cognitive impairment (MCI) (Early and Late), and AD patients.
- Applied probabilistic principal component analysis (PPCA) for dimensionality reduction on max-pooled and mean-pooled FDG-PET data.
- Developed and implemented multilayer feedforward neural networks for binary classification tasks, evaluated using F1-measure, precision, recall, and predictive values with 10-fold cross-validation.
Main Results:
- The proposed deep learning classifiers achieved competitive classification performance.
- Max-pooling of features demonstrated superior classification performance compared to mean-pooling.
- The study successfully discriminated between various AD diagnostic categories using FDG-PET data.
Conclusions:
- Deep learning models, combined with PPCA dimensionality reduction, show significant potential for classifying AD diagnostic categories using FDG-PET scans.
- FDG-PET analysis, enhanced by deep learning, can serve as a powerful imaging biomarker for Alzheimer's disease.
- Max-pooling is a more effective feature aggregation strategy than mean-pooling for this classification task.
