Brain Age Estimation by Convolutional Neural Network Based on T1-Weighted MRI and Gender Feature

Abstract

Aging is a very complicated process, which is influenced by a combination of genetic and environmental factors. Brain age is a biomarker that can be used to measure brain aging. Accurate methods for estimating brain age can deepen our understanding of the aging process in the brain and help us recognize neurodegenerative diseases. Since convolutional neural networks have shown better performance in regression tasks, applying convolutional neural networks to magnetic resonance images can conveniently estimate brain age. Most previous studies on brain age estimation have used only MRI images as input to the model. In this study, considering the differences in the aging process of the brains of different genders, we proposed a brain age estimation model based on 3D-ResNet34 using T1-weighted MRI images and gender features as inputs. This model performed better on the test set with the mean absolute error (MAE) of 4.85 years and the coefficient of determination (R2) of 0.886, compared to the model using only T1-weighted MRI images as input (MAE = 4.99, R2 = 0.882).