Brain-age prediction: Systematic evaluation of site effects, and sample age range and size

This study focuses on creating a robust model to estimate a person's brain age using structural MRI data. This 'brain-age' can be compared to chronological age to identify atypical brain development or aging. The researchers systematically tested different methods, including site harmonization, age range divisions, and sample sizes, to optimize the accuracy and generalizability of their brain-age prediction model. The final pre-trained model is made freely available on a web-based platform, CentileBrain, aiming to democratize access to this technology for researchers with varying levels of computational expertise.

• 1
  The accuracy of age prediction was higher when no site harmonization (correction for differences between data collection sites) was applied to the brain imaging data.
• 2
  Dividing the participant sample into two age groups (5–40 years and 40–90 years) provided the best balance between model accuracy and how well the model explained age-related differences.
• 3
  The model's accuracy for predicting brain age improved with larger sample sizes, but this improvement leveled off after the sample size exceeded 1600 participants.