A comparative evaluation of brain MRI modalities, modality-specific measures, brain regions, and cognitive tests for brain age prediction.
Authors
Affiliations (4)
Affiliations (4)
- Department of Neuroscience and Addiction Studies, School of Advanced Technologies in Medicine, Tehran University of Medical Sciences, Tehran, Iran.
- School of Mathematics, Statistics, and Computer Science, College of Science, University of Tehran, Tehran, Iran.
- Department of Neuroscience and Addiction Studies, School of Advanced Technologies in Medicine, Tehran University of Medical Sciences, Tehran, Iran. [email protected].
- BrainEE Research Group, Tehran University of Medical Sciences, Tehran, Iran. [email protected].
Abstract
Brain age estimation using machine learning has gained significant attention as a promising approach to assess cognitive health and aging. By analyzing structural and functional MRI data, these models can predict an individual's brain age, providing insights into neurological conditions and potential interventions. The purpose of this study is to compare the performance of different MRI modalities in predicting brain age and to determine the relative importance of specific brain regions. In this study, features extracted from T1-weighted MRI (cortical thickness and volume), rs-fMRI (amplitude of low-frequency fluctuations and functional connectivity matrix), diffusion-weighted MRI (diffusion tensor imaging parameters), and cognitive test scores were used in a ridge regression model to predict brain age. Feature selection, dimensionality reduction, hyperparameter tuning, and age-bias correction were performed within the training data of each outer fold to reduce information leakage. Rs-fMRI-derived features, specifically the correlation values of the functional connectivity matrix, achieved the lowest mean absolute error (MAE = 7.60 ± 0.94), followed by ALFF (MAE = 8.29 ± 1.09) and cognitive tests (MAE = 9.14 ± 1.27). Selecting the top 20% of features improved performance for cortical thickness and highlighted modality-specific regional contributions. However, connectivity-based models also showed a stronger residual association between the brain age gap and chronological age, indicating that prediction error and age bias should be interpreted jointly. These findings highlight the differential contributions of structural, functional, diffusion, and cognitive measures to brain age prediction. Systematic comparison of multiple imaging modalities and cognitive features provides valuable insight into their relative predictive utility and their potential role in characterizing brain aging. Future studies in larger and more diverse populations are warranted to determine whether these measures can serve as reliable biomarkers of neurodegenerative and abnormal aging processes.