Alzheimer's disease classification based on multimodal consistent distribution and trusted fusion

Area of Science:

• Neuroimaging
• Biostatistics
• Machine Learning

Background:

• Alzheimer's disease (AD) classification benefits from multimodal data fusion, integrating sources like MRI, PET, and cognitive scores.
• Existing fusion methods struggle with data heterogeneity and varying reliability across different modalities, hindering diagnostic accuracy.
• Robust AD classification requires addressing these challenges to improve diagnostic performance.

Purpose of the Study:

• To develop a novel Alzheimer's disease classification framework using multimodal consistent distribution and trusted fusion.
• To effectively integrate structural magnetic resonance imaging (sMRI), positron emission tomography (PET), and mini-mental state examination (MMSE) data.
• To enhance the reliability and accuracy of AD classification by managing data heterogeneity and uncertainty.

Main Methods:

• Projecting sMRI, PET, and MMSE data into a unified latent feature space for consistent distribution alignment.
• Employing a Dirichlet distribution-based mechanism to estimate belief and uncertainty in each modality's predictions.
• Developing a novel fusion strategy that integrates modality-specific belief and uncertainty for improved classification reliability.

Main Results:

• The proposed framework successfully aligns heterogeneous multimodal features into a consistent distribution, reducing interference.
• The Dirichlet distribution mechanism provides interpretability by quantifying prediction belief and uncertainty for each modality.
• The novel fusion strategy effectively integrates diverse data sources, leading to enhanced classification reliability and performance.
• Promising results were achieved on the ADNI and AIBL datasets across four AD-related classification tasks.

Conclusions:

• The developed framework effectively addresses multimodal heterogeneity and reliability issues in Alzheimer's disease classification.
• The approach demonstrates significant potential for improving diagnostic accuracy and robustness in AD detection.
• This method offers a reliable strategy for multimodal data fusion in neurodegenerative disease research.