WaveletBrain: Characterization of human brain via spectral graph wavelets

Abstract

Early diagnosis of Alzheimer's disease plays a key role in understanding the degree of the patient's mental decline and determining preventive therapies. In this study, we introduce WaveletBrain, a novel representation of the white and gray matter surfaces of the cortex. The proposed framework innovates by deriving localized shape information from a global harmonic representation, that can be used in large-scale population studies of surface data. Results show that WaveletBrain leads to statistically significant improvements in comparison to the ShapeDNA representation in a variety of experiments including (i) classification of Alzheimer's disease, normal aging, and mild cognitive impairment, (ii) sex classification and (iii) age prediction of subjects. We performed our analysis on 719 patients and 2,876 surfaces. While this work focuses primarily on Alzheimer's disease diagnosis, our proposed framework can be used to address general surface analysis problems in neuroscience.