Prediction of Alzheimer's Disease Based on Multi-Modal Domain Adaptation

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) diagnosis benefits from structural MRI and functional PET scans.
• Challenges include differing data structures and domain shift across datasets.
• Existing methods often overlook multi-modal complementarity and domain differences.

Purpose of the Study:

• To propose a Multi-modal Deep Domain Adaptation (MM-DDA) model for enhanced AD classification.
• To leverage complementary information from MRI and PET data.
• To mitigate domain distribution differences between datasets.

Main Methods:

• Developed an MM-DDA model integrating MRI and PET data.
• Employed CNNs for feature encoding, multi-head attention for feature fusion, and adversarial learning for domain transfer.
• Utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Main Results:

• Achieved high classification accuracies in transfer learning settings (e.g., ADNI1→ADNI2 and ADNI2→ADNI1).
• Accuracies for AD vs. CN reached up to 94.73%, and for pMCI vs. sMCI up to 81.81%.
• Demonstrated robust performance across various diagnostic comparisons.

Conclusions:

• The proposed MM-DDA model shows superior performance in AD prediction tasks.
• Effective integration of multi-modal data and domain adaptation significantly boosts classification accuracy.
• MM-DDA outperforms existing deep learning methods in transfer learning scenarios.