Radiomics-based brain aging prediction with multi-modal magnetic resonance imaging: detecting regional biomarkers in...

We developed a radiomics-based model to classify brain MRI scans into two age groups: younger (<30 years) and older (>65 years) adults using multi-modal magnetic resonance imaging (MRI; T1 and T2 sequences), identify regional biomarkers that distinguish younger and older adults, and identify radiomic signatures across brain regions and MRI modalities.

In this cross-sectional study, brain MRI data were obtained from the Information eXtraction from Images dataset and segmented using Freesurfer. Radiomic features were extracted from seven key brain regions using PyRadiomics software, adhering to the Image Biomarker Standardisation Initiative guidelines. A multi-step feature selection process was employed to identify significant radiomic signatures, which were used to develop logistic regression models for brain age prediction. Significant radiomic features were identified as indicative of brain aging across various regions.

The cohort included 122 participants. The T1 + T2 Rad-Score demonstrated high predictive accuracy, particularly for the hippocampus (area under the curve [AUC] = 0.96). Regional variability in model performance was observed, with the thalamus, caudate, and hippocampus showing the highest AUCs (0.95, 1.00, and 0.96, respectively). Radiomic signature importance varied across brain regions and MRI modalities: the thalamus, caudate, and hippocampus/amygdala were the most important in T1, T2, and T1 + T2 model images, respectively. The robustness and reproducibility of our findings were assured.

Radiomic signatures from T1 and T2 MRI data provided robust brain aging predictors, with models demonstrating high discriminative performance. We demonstrated the efficacy of a radiomics-based approach using multi-modal MRI for predicting brain aging and identifying regional biomarkers.