Research on Brain Age Prediction Based on Dual-Pathway 3D ResNet

Brain aging is a complex process, while its mechanism remains unknown. Due to the accuracy and high reliability of brain age as a phenotype, which can be used to track the trajectory of human brain aging, it has garnered considerable interest among researchers as a potential means to evaluate this process. In this study, we proposed a novel brain age prediction model based on a dual-path 3D ResNet architecture, which leverages two pathways to extract deep features from both grey and white matter structural MRI for the purposes of brain age prediction. In the testing dataset, this model produced improved prediction results, with mean absolute error (MAE) of 4.4396 years, root mean square error (RMSE) of 5.7886 years, and coefficient of determination (R2) of 0.8860. These findings demonstrate that our dual-path model significantly enhanced the accuracy of brain age evaluation using structural magnetic resonance imaging (sMRI) as compared to models solely relying on grey matter or white matter images as input. Thus, our findings make a valuable contribution to the area of brain aging prediction.