Progression-Aware and Explainable CNN-Transformer Framework for Multiclass Alzheimer's Disease Staging Using MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) presents progressive neurodegeneration, making accurate staging via MRI challenging.
• Current deep learning models often overlook disease progression, exhibit evaluation leakage, or lack interpretability.

Purpose of the Study:

• Introduce DeepAttentionADNet, a hybrid CNN-Transformer model for multiclass Alzheimer's disease staging using MRI.
• Address limitations of existing methods by incorporating progression awareness and interpretability.

Main Methods:

• Integrate convolutional neural networks (CNNs) for feature extraction with Transformers for global context modeling.
• Employ progression-aware ordinal learning and consistency regularization for robustness and capturing disease severity.
• Utilize a leakage-free cross-validation protocol and transformer-based importance maps for interpretability.

Main Results:

• Achieved high and consistent performance across cross-validation folds on an Alzheimer's disease MRI dataset.
• Reported a mean F1-score of 0.991 ± 0.003 and AUROC of 0.9998 ± 0.0002.
• Demonstrated transparency and progress awareness in classification decisions.

Conclusions:

• DeepAttentionADNet offers a robust and interpretable solution for classifying Alzheimer's disease severity from MRI.
• The framework effectively handles multiclass classification while maintaining transparency and awareness of disease progression.