OpenMAP-BrainAge: generalizable and interpretable brain age predictor from MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Accurate brain age estimation is crucial for detecting neurodegenerative diseases.
• Robust models are needed for diverse patient populations and varying MRI data.

Purpose of the Study:

• To create an interpretable brain age prediction model.
• To ensure the model is robust against demographic and technological variations in brain MRI.

Main Methods:

• A transformer-based model was developed to analyze 3D T1-weighted MRI scans.
• Model performance was evaluated using Mean Absolute Error (MAE).
• Brain Age Gap (BAG) was analyzed in relation to cognitive status and scores (MMSE, MoCA).

Main Results:

• Achieved state-of-the-art MAE of 3.65 years on ADNI2 & 3/OASIS3 and 3.54 years on AIBL.
• Significant increase in BAG observed in mild cognitive impairment (2.55 years) and dementia (6.12 years) compared to cognitively normal individuals (0.15 years).
• Negative correlation found between BAG and cognitive scores (MMSE, MoCA), indicating association with cognitive decline.

Conclusions:

• The transformer model offers state-of-the-art brain age prediction with enhanced generalizability and interpretability.
• The model's findings correlate with cognitive function, highlighting its clinical relevance for neurodegenerative disease assessment.