A Convolutional Mixer-Based Deep Learning Network for Alzheimer's Disease Classification from Structural Magnetic Resonance Imaging

Area of Science:

• Neuroscience
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder impacting cognitive function, necessitating accurate diagnostic tools for timely intervention.
• Current diagnostic methods require improvement for precise staging of AD progression.
• Structural magnetic resonance imaging (sMRI) offers valuable insights into brain changes associated with AD.

Purpose of the Study:

• To develop and validate a novel, efficient deep learning architecture for classifying Alzheimer's disease stages using sMRI.
• To address the challenge of class imbalance in medical imaging datasets for improved model generalization.
• To enhance the interpretability of the classification model using explainable AI techniques.

Main Methods:

• A hybrid convolutional neural network architecture integrating depthwise separable and traditional convolutional layers was proposed.
• A novel approach combining SMOTE (synthetic minority oversampling technique) and ENN (edited nearest neighbors) was employed to handle class imbalance.
• An explainable activation space occlusion sensitivity map (ASOP) method was utilized for visualizing critical image regions influencing classification.

Main Results:

• The proposed model demonstrated superior performance compared to several state-of-the-art transfer learning architectures.
• Achieved high classification metrics: 98.87% accuracy, 98.86% F1 score, 98.80% precision, and 98.69% recall.
• The hybrid sampling technique effectively mitigated the impact of class imbalance on model performance.

Conclusions:

• The developed deep learning model is highly effective for classifying Alzheimer's disease stages from sMRI scans.
• The integration of efficient architectural components and advanced data balancing techniques contributes to robust performance.
• Explainable AI methods provide insights into the model's decision-making process, increasing trust and clinical utility.