Uncertainty-aware multi-path framework with dynamic arbitration for Alzheimer's disease MRI classification

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder requiring early diagnosis.
• Deep learning on structural MRI (sMRI) shows promise for computer-aided AD diagnosis but faces challenges with heterogeneity and uncertainty.
• Test-time perturbations can cause anatomical inconsistencies, impacting diagnostic stability.

Purpose of the Study:

• To develop an uncertainty-aware deep learning framework for robust Alzheimer's disease diagnosis using sMRI.
• To improve the reliability of computer-aided diagnosis by handling sample-specific heterogeneity and prediction uncertainty.

Main Methods:

• Proposed TriPathNet with Arbiter, an uncertainty-aware multi-path framework utilizing parallel encoding and metric-guided coordination.
• Implemented an augmentation-aware arbitration mechanism to correct predictions for high-uncertainty samples during inference.
• Evaluated performance on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset for AD vs. cognitively normal (CN) classification.

Main Results:

• Achieved high accuracy (98.85%), sensitivity (99.18%), specificity (97.06%), and AUC (99.61%) on the ADNI dataset.
• Demonstrated stable performance under test-time perturbations, indicating robustness.
• The framework effectively handles sample-specific heterogeneity and prediction uncertainty.

Conclusions:

• TriPathNet with Arbiter offers an accurate and robust solution for sMRI-based computer-aided Alzheimer's disease diagnosis.
• The proposed method shows significant potential for clinical application in early AD detection.
• Addressing uncertainty and perturbations is crucial for reliable deep learning in medical imaging.