Feature selective temporal prediction of Alzheimer's disease progression using hippocampus surface morphometry

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Alzheimer's disease (AD) progression prediction is crucial for treatment strategies.
• Previous models used general MRI features, cognitive scores, and demographics.
• Hippocampal structure is vital in AD, suggesting localized analysis may improve predictions.

Purpose of the Study:

• To develop a more accurate method for predicting cognitive decline in Alzheimer's disease.
• To integrate advanced hippocampal surface analysis with machine learning for enhanced predictive power.

Main Methods:

• A novel multivariate morphometric surface map of the hippocampus (mTBM) was developed.
• mTBM features were combined with traditional MRI features and demographic data.
• A predictive multi-task machine learning method (cFSGL) was employed for simultaneous prediction across multiple time points.

Main Results:

• The combined approach generated a significantly larger feature space (2100+ features) from hippocampal surface maps.
• This integrated method improved the prediction accuracy of Alzheimer's Disease Assessment Scale (ADAS) cognitive scores.
• Predictions were enhanced at 6, 12, 24, 36, and 48 months from baseline.

Conclusions:

• Combining cFSGL with mTBM hippocampal surface features significantly improves AD cognitive score prediction.
• This approach offers a more sensitive method for tracking AD progression using neuroimaging.
• The findings support the utility of detailed hippocampal analysis in predicting Alzheimer's disease outcomes.