Age estimation using effective brain local features from T1-weighted images

This paper proposes a simple method of selecting effective brain local features for age estimation from T1-weighted MR images. We also employ the high-resolution AAL atlas, which is defined by 1,024 local regions, to improve the accuracy of age estimation. We evaluate performance of the proposed method using 1,099 T1-weighted images from a large-scale brain MR image database of healthy Japanese, and demonstrate that the proposed method exhibits efficient performance of age estimation compared with conventional methods.

I. Introduction

Studies on statistical analysis of brain T1-weighted images have found that brain development follows a specific pattern of morphological changes during the normal aging process. Volumes of brain tissues such as gray matter (GM), white matter (WM) and Cerebrosphinal fluid (CSF) have changed in the process of brain development and healthy aging [1]–[4]. GM volume monotonically decreases with age from 20s to 70s, WM volume shows small changes, and CSF monotonically increases with age from 20s to 70s in contrast with GM volume [4]. On the other hand, brain disorders such as Alzheimer's disease (AD) have changed the pattern due to brain atrophy. The above fact allows us to estimate the age of subjects from T1-weighted images and identify age-related brain disorders by evaluating the difference between the actual and estimated age.