Instance-Based Representation Using Multiple Kernel Learning for Predicting Conversion to Alzheimer Disease.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Early detection of Alzheimer's disease (AD) is challenging for machine learning due to missing data and interpretability issues.
• Predicting conversion from mild cognitive impairment (MCI) to AD is crucial for timely intervention.

Purpose of the Study:

• To develop a machine learning methodology for predicting MCI to AD conversion using structural brain MRI.
• To address challenges of missing data and enhance result interpretability in Alzheimer's disease progression modeling.

Main Methods:

• Instance-based feature mapping using morphological measures of brain structures to handle missing follow-up data.
• Combining multiple feature mappings via convex combination of reproducing kernels.
• Tuning kernel weights using the centered-kernel alignment criterion.

Main Results:

• The proposed method accurately detects MCI and predicts conversion to Alzheimer's disease within three years.
• Experimental validation performed on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Identified brain structures with higher weights are linked to memory and cognitive functions, aiding interpretability.

Conclusions:

• Instance-based representation with multiple kernel learning offers a robust approach for Alzheimer's disease progression prediction.
• The methodology effectively handles missing data in longitudinal neuroimaging studies.
• The model provides insights into key brain structures influencing cognitive decline in Alzheimer's disease.