Quantifying uncertainty in brain-predicted age using scalar-on-image quantile regression.

Area of Science:

• Neuroimaging
• Biostatistics
• Neurodegenerative Diseases

Background:

• Brain age prediction from MRI offers insights into neurodegenerative diseases.
• Current methods often overlook prediction uncertainty.
• Quantifying this uncertainty is crucial for reliable clinical application.

Purpose of the Study:

• To develop a novel method for brain age prediction that incorporates uncertainty.
• To apply functional data analysis, specifically penalized functional quantile regression, for this purpose.
• To predict brain age and its uncertainty in individuals with Mild Cognitive Impairment (MCI) and Alzheimer's Disease (AD).

Main Methods:

• Utilized penalized functional quantile regression on brain structure data from cognitively normal (CN) subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.
• Modeled age as a function of brain anatomy.
• Extended the model to predict brain age for MCI and AD subjects.

Main Results:

• The proposed model generates prediction intervals for brain age, unlike traditional point-prediction machine learning methods.
• This provides a more nuanced estimation of brain age, reflecting inherent uncertainties.
• The method was successfully applied to predict brain age in MCI and AD populations.

Conclusions:

• Functional quantile regression offers a robust framework for brain age prediction with uncertainty quantification.
• Prediction intervals enhance the interpretability and clinical utility of brain age as a biomarker.
• This approach advances the assessment of brain changes in neurodegenerative conditions.