Predicting Future Brain Atrophy Based on Longitudinal MRI

Area of Science:

• Neuroimaging
• Machine Learning
• Neurodegenerative Diseases

Background:

• Neuron loss and brain atrophy are hallmarks of neurodegenerative diseases like Alzheimer's disease (AD).
• Magnetic resonance imaging (MRI) is crucial for detecting brain atrophy, essential for AD research.
• Accurate measurement of brain atrophy is vital for understanding disease progression and evaluating interventions.

Purpose of the Study:

• To forecast annual percentage changes in brain volumes (hippocampus, ventricles, total gray matter) using machine learning.
• To compare the predictive power of baseline versus longitudinal MRI data combined with risk factors.
• To assess the utility of predicted atrophy rates for predicting clinical status progression in AD and related dementias.

Main Methods:

• Developed an elastic net linear regression model to predict future annual brain volume changes.
• Evaluated two approaches: baseline (single-time-point) and longitudinal (multiple time points).
• Compared MRI-only models with models incorporating risk factors (age, sex, APOE4, diagnosis), validated on external datasets.

Main Results:

• The longitudinal MRI + risk factor model achieved high prediction accuracy (e.g., 0.62 for hippocampus).
• Longitudinal models consistently outperformed baseline models; models with risk factors outperformed MRI-only models.
• Predicted atrophy rates were superior to current volumes for predicting progression to mild cognitive impairment and dementia.

Conclusions:

• Future atrophy prediction using longitudinal data and risk factors is a valuable tool for assessing cognitive decline risk.
• This approach can aid in identifying individuals for clinical trials targeting disease-modifying therapies for AD.
• Predicted atrophy rates offer a more sensitive marker of disease progression than static volume measurements.