Automatic Classification of Alzheimer’s Disease based on CPD Brain Point Cloud Registration for Feature Extraction

In this paper, a new scheme of feature extraction based on the registration of two point clouds that model the brain surface area and a cerebral baseline, is presented. The aim is to design an automatic Random Forest classifier based on brain-shape pattern extraction, in order to discriminate among Control subjects, patients with Mild Cognitive Impairment, and patients with Alzheimer's disease using their MRI data. The features used were rotation and displacement parameters obtained from the rigid-register transformation matrix between each subject's point cloud model and the brain baseline cloud. Additional features consisted of distance metrics performed over the aligned clouds and the baseline; those metrics attempt to estimate the cerebral cortex deformations. The training set utilized 60 MRI grouped into subsets of 20 studies a class; while the test consisted of 30 not seen data (10 per class). The proposed method achieved up to 87.2% accuracy in the final validation stage. These results suggest that our proposed scheme could be useful to discriminate among several pathologies that affect brain gray matter morphology. Moreover, the proposed features might be used as brain deformation quantifiers.