Incomplete multi-modal representation learning for Alzheimer's disease diagnosis.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Alzheimers disease (AD) diagnosis and treatment are critical research areas.
• Multi-modal data representation learning is emerging for AD research.
• Incomplete multi-modal data poses a significant challenge for existing algorithms.

Purpose of the Study:

• To propose a novel framework for learning common representations from incomplete multi-modal data for AD diagnosis.
• To address the limitations of current multi-modal learning algorithms in handling missing data.

Main Methods:

• Developed an Auto-Encoder based Multi-View missing data Completion framework (AEMVC).
• Mapped complete data views to a latent space using auto-encoders.
• Complemented incomplete views using latent representations and kernel matrices.
• Incorporated graph regularization and Hilbert-Schmidt Independence Criterion (HSIC) for structural and associative information preservation.
• Applied a kernel-based multi-view method for common representation acquisition.

Main Results:

• The AEMVC framework effectively learns common representations from incomplete multi-modal data.
• Experimental validation on Alzheimers Disease Neuroimaging Initiative (ADNI) datasets demonstrated the method's effectiveness.
• The proposed approach successfully handles partially available multi-modal data for AD diagnosis.

Conclusions:

• The AEMVC framework offers a robust solution for multi-modal data analysis in AD research.
• This method advances the field of representation learning for complex diseases like Alzheimers.
• The findings highlight the potential of advanced AI techniques in improving early AD diagnosis.