Classification of Alzheimer's disease in MRI using visual saliency information

Computational visual atention models aims to emulate the Human Visual System performance in selecting relevant features for efficient visual scene processing. As a result, visual saliency maps highlights relevant visual patterns in an image, possibly associated with objects or specific concepts. In the analysis of medical images, this allows the radiologist or clinical expert to focus the attention on image anormalities or specific patterns that could suggest the presence of a pathology. This paper presents an initial exploration of the effect of visual saliency models in the extraction of pathology-related relevant patterns, suitable for classification of Magnetic Resonance images of normal controls and probable Alzheimer's disease patients. By adjusting the saliency models to work on medical images, and combining this process with a Support Vector Machine for classification, the preliminar results shows a maximum performance of 85% in accuracy and 0.9 in the area under the ROC curve. In comparison with previous approaches, an increment of about 4% in the classification performance, suggesting that the visual saliency information could be promising for AD discrimination.