Brain Region-Centered MultiModal Hypergraph Fusion for MCI Conversion Prediction

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder.
• Mild cognitive impairment (MCI) is an early stage of AD, making MCI conversion prediction vital for timely intervention.
• Current AI-based diagnostic methods struggle with multimodal neuroimaging data heterogeneity and interpretability.

Purpose of the Study:

• To develop an advanced deep learning framework for accurate prediction of MCI conversion to AD.
• To address limitations in existing multimodal fusion techniques, including modality heterogeneity and poor modeling of inter-regional interactions.
• To enhance the interpretability of AI models in AD diagnosis.

Main Methods:

• Proposed the Brain Region-Centered MultiModal Hypergraph Fusion (BRC-MMHF) framework.
• Utilized parameter-free channel exchange and ROI-level feature extraction for MRI and PET data to reduce heterogeneity.
• Employed a multimodal hypergraph to model high-order cross-modal relationships and a lesion-aware module for interpretability.
• Integrated structured clinical data via a lightweight tabular encoder.

Main Results:

• BRC-MMHF achieved 80.79% accuracy and 89.78% AUC in MCI conversion prediction on the ADNI dataset.
• The framework outperformed existing state-of-the-art methods utilizing MRI and PET imaging.
• The lesion-aware module provided auxiliary interpretability, highlighting disease-relevant brain regions.

Conclusions:

• The BRC-MMHF framework offers a robust and interpretable solution for predicting MCI conversion to AD.
• This approach effectively integrates multimodal neuroimaging data, overcoming previous limitations.
• The findings support the potential of BRC-MMHF for early AD diagnosis and intervention strategies.