HDFT-MViT: a progressive core-enhanced mix framework for Alzheimer's disease classification using MRI images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate Alzheimer's disease (AD) diagnosis is crucial for timely intervention.
• Current MRI-based methods using CNNs and ViTs have limitations in capturing both local and global features efficiently.
• High computational complexity of ViTs hinders their use in resource-limited clinical settings.

Purpose of the Study:

• To develop a lightweight hybrid deep learning model for enhanced Alzheimer's disease diagnosis from MRI scans.
• To address the limitations of existing models in balancing local feature extraction and global dependency modeling.
• To create an efficient and accurate tool for clinical AD diagnosis.

Main Methods:

• Proposed HDFT-MViT, a hybrid architecture based on MobileViT, integrating hierarchical dynamic filters and lightweight Transformers.
• Employed a progressive Core-Enhanced Mix design with MobileNetV2 for shallow layers and a dual-branch module for deeper layers.
• Incorporated frequency-domain global modulation, spatial long-range dependency modeling, hierarchical fusion, and channel attention.

Main Results:

• Achieved state-of-the-art classification accuracies of 98.85% (3-class) and 98.07% (4-class) on ADNI MRI datasets.
• Maintained a lightweight profile with only 3.46 million parameters.
• Demonstrated superior effectiveness and efficiency compared to existing methods.

Conclusions:

• HDFT-MViT effectively balances local and global feature perception in a computationally efficient framework.
• The model presents a promising tool for accurate and efficient clinical diagnosis of Alzheimer's disease.
• The proposed architecture offers a viable solution for resource-constrained environments.