Joint Multi-Modal Longitudinal Regression and Classification for Alzheimer's Disease Prediction

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Informatics

Background:

• Alzheimer's disease (AD) is a growing global health concern with increasing prevalence.
• Understanding the biological mechanisms driving AD development is crucial for effective intervention.
• Current computational approaches face challenges in integrating diverse clinical data for AD research.

Purpose of the Study:

• To present a novel computational method for identifying Alzheimer's disease (AD) biological mechanisms.
• To assess cognitive status and underlying mechanisms using multi-modal data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.
• To develop a flexible and efficient approach for AD research and potentially other neurodegenerative conditions.

Main Methods:

• Developed a Joint Multi-Modal Longitudinal Regression and Classification (jmmlrc) method.
• Integrated genetic information and brain scans using advanced regularization techniques to identify AD-relevant biomarkers.
• Designed an efficient iterative algorithm to solve the non-smooth optimization problem and proved its convergence.

Main Results:

• The jmmlrc method accurately predicted cognitive scores and clinical diagnosis in the ADNI cohort.
• Demonstrated the method's ability to identify key biological mechanisms underlying Alzheimer's disease.
• Experimental results validated the effectiveness, benefits, and flexibility of the proposed approach.

Conclusions:

• The novel jmmlrc method offers a powerful tool for Alzheimer's disease research.
• The approach successfully integrates multi-modal data to uncover disease mechanisms and predict patient outcomes.
• The open-sourced code facilitates broader application in clinical research beyond AD.