Dense convolution-based attention network for Alzheimer's disease classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Deep learning models for medical image classification show promise but often lack efficiency for clinical use.
• Existing Convolutional Neural Network (CNN), Transformer, and hybrid models face challenges balancing performance and complexity.
• Efficient Alzheimer's disease detection in 3D MRI requires models that are both accurate and computationally feasible.

Purpose of the Study:

• To propose DenseAttentionNetwork (DANet), a lightweight deep learning model for efficient Alzheimer's disease detection.
• To enhance feature extraction and capture long-range dependencies in 3D MRI data for improved diagnostic accuracy.
• To develop a model that balances high performance with a low parameter count for practical clinical application.

Main Methods:

• Proposed DenseAttentionNetwork (DANet), a lightweight architecture integrating dense connections and a linear attention mechanism.
• Utilized convolutional layers for localized feature extraction and linear attention for global context and efficient multi-scale feature reuse.
• Replaced traditional self-attention with a parameter-efficient linear attention mechanism to overcome limitations of standard self-attention.

Main Results:

• DANet achieved superior performance, indicated by the highest area under the receiver operating characteristic curve (AUC), across multi-institutional datasets.
• The model demonstrated robustness and effectiveness in capturing relevant features for Alzheimer's disease detection.
• DANet attained high accuracy with significantly fewer parameters compared to existing models.

Conclusions:

• DANet offers an efficient and effective solution for Alzheimer's disease detection using 3D MRI scans.
• The model's ability to highlight AD-relevant regions, verified by activation maps, provides clinically interpretable insights.
• DANet represents a practical advancement in applying deep learning for neurodegenerative disease diagnosis.