Fast brain-wide search of highly discriminative regions in medical images: an application to Alzheimer's disease

We propose a fast algorithm for the identification of localised brain regions from medical images that discriminate between two groups of individuals. The method is based on a combination of penalised regression and a data resampling procedure. We apply this approach to both MRI and PET images for the classification of subjects with Alzheimer's disease and mild cognitive impairment. We show that the voxels selected by the algorithm form connected brain regions which are well known to be affected by Alzheimer's disease. A linear statistical classifier trained on the selected voxels achieves cross-validated classification results that are comparable to those obtained by current state-of-the-art methodologies.