Early Alzheimer disease detection with bag-of-visual-words and hybrid fusion on structural MRI

In this paper, we tackle the problem of recognition of Alzheimer's disease (AD) in structural MRI images using visual similarity. AD yields visible changes in the brain structures. We aim to recognize patient category such as AD, or prodromal stage of the AD called Mild Cognitive impairment (MCI), or normal control subject (NC). We use visual local descriptors and the bag-of-visual-words approach on the most involved regions in AD (Hippocampus and Posterior Cingulate Cortex) in MRI. The Content-Based Visual information retrieval (CBVIR) approach is then applied to recognize patient category. The contribution of the paper is in the fusion of visual signatures and of classification results obtained on characteristic brain regions. The performance of image retrieval is improved by 10% using the early and hybrid fusion with regard to the use of hippocampus region only.