Predicting Brain Age at Slice Level: Convolutional Neural Networks and Consequences for Interpretability.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Chronological aging is linked to brain degeneration, increasing risks for stroke and dementia.
• The brain age gap, a discrepancy between chronological and predicted brain age, may signal early aging processes.
• Machine learning models for brain age prediction often lack interpretability.

Purpose of the Study:

• To develop and evaluate an interpretable brain age regression approach using MRI data.
• To identify specific brain regions and MRI slices contributing to the brain age gap.
• To understand factors influencing brain age prediction models.

Main Methods:

• Convolutional neural networks trained on single MRI slices (gray/white matter) to predict brain age.
• Slice-level prediction analysis to identify regional age discrepancies.
• Evaluation of influences from slice index, plane, participant demographics, and MRI site.

Main Results:

• Prediction error varied significantly based on the specific MRI slice used.
• Stratifying MRI data by collection site minimized site and sex effects on predictions.
• The orientation (plane) of the MRI slice impacted overall model error.

Conclusions:

• Slice-based brain age prediction enhances interpretability compared to whole-brain methods.
• This approach offers a more reliable tool for researchers and clinicians to assess brain age.
• Identifies specific brain regions contributing to age-related changes.