Probing multiple algorithms to calculate brain age: Examining reliability, relations with demographics, and...

Area of Science:

• Neuroscience
• Radiology
• Gerontology

Background:

• Brain age from structural MRIs is a promising aging biomarker.
• Discrepancies between brain age and chronological age may predict health outcomes.
• Technical complexities exist in calculating brain age using various algorithms.

Purpose of the Study:

• To systematically compare three common brain age algorithms: XGBoost, brainageR, and DeepBrainNet.
• To assess the reliability, collinearity, and predictive power of these algorithms.
• To identify predictors of cognitive impairment across different brain age calculations.

Main Methods:

• Utilized multiple datasets with repeated structural MRI scans for reliability analysis (intraclass correlations, Bland-Altman bias).
• Examined correlations between brain age, chronological age, sex, and image quality.
• Employed machine learning to identify predictors of cognitive impairment.

Main Results:

• All three algorithms demonstrated excellent reliability (r > 0.9).
• brainageR and DeepBrainNet showed moderate correlation; XGBoost correlated strongly with image quality.
• XGBoost brain age was more sensitive in detecting clinical diagnoses of cognitive impairment.

Conclusions:

• Common brain age algorithms offer reliable estimations of brain age.
• Algorithm choice impacts sensitivity to cognitive impairment detection, with XGBoost showing an advantage.
• Recommendations are provided for future brain age research utilizing these findings.