Extending local binary patterns to 3D for the diagnosis of Alzheimer´s Disease

Neuroimaging has shown great potential for the computer-aided diagnosis (CAD) of both Alzheimer´s disease (AD) and Mild Cognitive Impairment (MCI). However, the texture of such images has been little explored. In this paper, we explore the discriminative power of the Local Binary Patterns (LBPs) texture descriptor to diagnose AD and MCI from 3D brain images. For this purpose, we propose a novel extension of LBPs to full 3D data that, unlike previous approaches, makes no approximations to the underlying concepts of uniformity and rotation invariance. Experimental results obtained using FDG-PET images from the Alzheimer´s Disease Neuroimaging Initiative (ADNI) showed that the new feature was able to improve the system´s performance when compared to the raw Voxel Intensities (VI) and to the standard 2D LBP version applied to axial cuts of the PET volume.