Brain Age Estimation Using Universum Learning-Based Kernel Random Vector Functional Link Regression Network

Brain age serves as a vital biomarker for detecting neurological ailments like Alzheimer’s disease (AD) and Parkinson’s disease (PD). Magnetic resonance imaging (MRI) has been extensively explored with deep neural networks to estimate brain age. The discrepancy between the predicted age and chronological age (real age) can be instrumental in identifying brain-related issues and assessing overall brain health. In this study, we have developed a brain age estimation framework utilizing a ResNet-50 deep neural network and a universum learning-based kernel random vector functional link (UKRVFL) network based on MRI images. A novel formulation of universum-KRVFL is introduced for regression tasks that capitalizes on prior knowledge through supplementary data samples. The universum data samples originate from the same domain as training samples but have different distributions. The proposed work efficacy is substantiated by conducting experiments on publicly available datasets. The model performance is quantified through metrics such as the mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination ( \(R^2\) $R^2$ ), where lower MAE and RMSE values and a higher \(R^2\) $R^2$ indicate greater accuracy in age prediction. The proposed age prediction model achieved an MAE of 2.68 years and 3.53 years of RMSE on healthy control (HC) test subjects. To further assess the significance of the brain age gap (BAG) as a biomarker for brain health, studies are conducted on mild cognitive impairment (MCI), PD, and AD groups. For MCI, PD, and AD groups, age estimation model yielded an RMSE of 4.13, 4.86, and 6.60 years, respectively. The experimental results demonstrate that the brain age gap in brain function is notably wider within AD group, indicating an acceleration of brain aging among those with neurodegeneration.