The Integrated Voxel Analysis Method (IVAM) to Diagnose Onset of Alzheimer's Disease and Identify Brain Regions through Structural MRI Images.

Magnetic Resonance Imaging (MRI) provides three-dimensional anatomical and physiological details of the human brain. We describe the Integrated Voxel Analysis Method (IVAM) which, through machine learning, classifies MRI images of brains afflicted with early Alzheimers Disease (AD). This fully automatic method uses an extra trees regressor model in which the feature vector input contains the intensities of voxels, whereby the effect of AD on a single voxel can be predicted. The resulting tree predicts based on the following two steps: a K-nearest neighbor (KNN) algorithm based on Euclidean distance with the feature vector to classify whole images based on their distribution of affected voxels and a voxel-by-voxel classification by the tree of every voxel in the image. An Ising model filter follows voxel-by-voxel tree-classification to remove artifacts and to facilitate clustering of classification results which identify significant voxel clusters affected by AD. We apply this method to T1-weighted MRI images obtained from the Open Access Series of Imaging Studies (OASIS) using images belonging to normal and early AD-afflicted individuals associated with a Client Dementia Rating (CDR) which we use as the target in the supervised learning. Furthermore, statistical analysis using a pre-labeled brain atlas automatically identifies significantly affected brain regions. While achieving 90% AD classification accuracy on 198 images in the OASIS dataset, the method reveals morphological differences caused by the onset of AD.