Human Age Estimation with Surface-Based Features from MRI Images

Over the past years, many efforts have been made in the estimation of the physiological age based on the human MRI brain images. In this paper, we propose a novel regression model with surface-based features to estimate the human age automatically and accurately. First, individual regional surface-based features (thickness, mean curvature, Gaussian curvature and surface area) from the MRI image were extracted, which were subsequently used to construct combined regional features and the brain networks. Then, the individual regional surface-based features, brain network with surface-based features and combined regional surface-based features were used for age regression by relevance vector machine (RVM), respectively. In the experiment, a dataset of 360 healthy subjects aging from 20 to 82 years was used to evaluate the performance. Experimental results based on 10-fold cross validation show that, compared to the previous methods, age estimation model with combined surface-based features can yield a remarkably high accuracy (mean absolute error: 4.6 years and root mean squared error: 5.6 years) and a significantly high correlation coefficient (r = 0.94), which is the best age estimation result as far as we know and suggests that surface-based features are more powerful than other features used in previous methods for human age estimation.