Quantitative Analysis of BrainAge using Neuroimaging Data to Estimate Radiation-Induced Brain Aging and Associations with Cognitive Changes in Brain Metastases Patients.

PURPOSE/OBJECTIVE(S): Radiotherapy is critical in brain tumor management yet causes accelerated aging via irreparable damage to white matter, cortex, and subcortical areas. BrainAGE is an open-source machine learning tool that analyses morphology from raw quantitative structural MRI to generate a predictive 'BrainAge' value, enabling analyses of patients with neurodegenerative diseases. We assessed the utility of this algorithm in measuring brain age morphology in brain metastases patients on a clinical trial, analyzed differences in BrainAge and chronological age, and explored associations with cognitive performance.

MATERIALS/METHODS: We analyzed pre-SRS high-resolution T1-weighted noncontrast volumetric MRI in patients with brain metastases (n = 57, median age 63 years, range 20-87) treated on a prospective clinical trial of cognitive-sparing SRS. The BrainAGE machine learning tool processed images to derive a BrainAge; it reliably generates this value using a Gaussian Processes regression, with a robust neurotypical training cohort. We examined correlations between BrainAge and chronological age with regression. Brain-predicted age gap (Brain-Gap) was calculated, defined as the difference between a patient's BrainAge and chronological age. We assessed cognitive function at baseline pre-SRS with a comprehensive battery of validated tests across multiple domains performed by a neuropsychologist, including memory (HVLT total), executive function (COWAT category fluency), and language (COWAT letter fluency). Raw test scores were scaled to T-scores adjusting for age, sex, and education level per testing norms. We examined if Brain-Gap was associated with cognitive performance pre-SRS.

RESULTS: BrainAGE software was robust, estimating a BrainAge for all input MRI sequences. We found a strong correlation between BrainAge and chronological age (r-squared = 0.82). The Brain-Gap median value was 0.84 and the mean was -0.90, in years (range -18.30-11.36 years). Brain-Gap was not significantly associated with cognitive performance across the three tests at the pre-SRS timepoint. Among patients with impaired cognition (T<35), there was no significant difference in BrainAge relative to chronological age.

CONCLUSION: To our knowledge, this is the first study utilizing the BrainAGE model as a biomarker of neuroanatomical age for brain tumors patients. Existing publications had lesion-free cohorts. Our results show that BrainAge was reliably associated with chronologic age in brain metastases patients pre-SRS. The Brain-Gap difference was minimal at baseline, and we found no significant association with pre-SRS cognitive performance. Future studies will examine this biomarker after SRS to look for individual aging radiation effects and prediction for cognitive changes after therapy.