Age estimation using cortical surface pattern combining thickness with curvatures

Brain development and healthy aging have been proved to follow a specific pattern, which, in turn, can be applied to help doctors diagnose mental diseases. In this paper, we design a cortical surface pattern (CSP) combining the cortical thickness with curvatures, which constructs an accurate human age estimation model with relevance vector regression. We test our model with two public databases. One is the IXI database (360 healthy subjects aging from 20 to 82 years old were selected), and the other is the INDI database (303 subjects aging from 7 to 22 years old were selected). The results show that our model can achieve as small as 4.57 years deviation in the IXI database and 1.38 years deviation in the INDI database. Furthermore, we employ this surface pattern to age groups classification and get a remarkably high accuracy (97.77 %) and a significantly high sensitivity/specificity (97.30/98.10 %). These results suggest that our designed CSP combining thickness with curvatures is stable and sensitive to brain development, and it is much more powerful than voxel-based morphometry used in previous methods for age estimation.