Active learning based segmentation of Crohn´s disease using principles of visual saliency

We propose a active learning (AL) approach to segment Crohn´s disease (CD) affected regions in abdominal magnetic resonance (MR) images. Our label query strategy is inspired from the principles of visual saliency which has similar considerations for choosing the most salient region. These similarities are encoded in a graph using classification maps and low level features. The most informative node is determined using random walks. Experimental results on real patient datasets show the superior performance of our approach and highlight the importance of different features to determine a region´s importance.