Effective deep convolutional neural network with attention mechanism for Alzheimer disease classification

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Neurocognitive disorders, particularly Alzheimer's disease (AD), are increasing in middle-aged and elderly populations.
• Early and accurate detection of AD is vital for timely intervention and preventing irreversible brain damage.
• Current computational approaches for AD detection face limitations in accuracy and clinical validation, especially for early stages.

Purpose of the Study:

• To review existing computational techniques for Alzheimer's disease detection.
• To propose a Deep Convolutional Neural Network (Deep-CNN) with an attention mechanism for enhanced early-stage AD detection.
• To improve the accuracy and interpretability of Alzheimer's disease diagnosis using machine learning.

Main Methods:

• A Deep Convolutional Neural Network (Deep-CNN) model incorporating an attention mechanism was developed.
• The model was designed to augment spatial attention and perform multi-class classification of Alzheimer's disease stages.
• The model was trained and evaluated on the OASIS dataset using subject-level data with standard preprocessing and statistical validation.

Main Results:

• The proposed Deep-CNN with attention model achieved a diagnostic accuracy of 97%.
• This accuracy surpasses existing methods, including Support Vector Machines (SVM) with kernels (90.5%, 85%) and traditional CNN (93.5%).
• Attention mechanism visualizations aligned with known Alzheimer's disease biomarkers, enhancing model interpretability.

Conclusions:

• Attention-guided deep learning models can significantly improve the accuracy of multi-class MRI classification for Alzheimer's disease.
• These models offer clinically useful regional explanations, aiding in the understanding of disease progression.
• The developed Deep-CNN with attention mechanism presents a promising tool for efficient and accurate early-stage Alzheimer's disease detection.