Paper Title
Deep Pattern-based Classification with Entropy Coding for MRI Segmentation

Abstract
After several decades, it still hard to efficiently segment MRI data namely when the latter are affected by image artifacts such as noise and intensity non uniformity. Aiming at enhancing Magnetic Resonance Image (MRI)-based diagnosis or cerebral tissue evolution, data volumes must be correctly segmented even with high levels of MRI specific artifacts. In this paper, a new machine learning-based method for MRI segmentation is proposed, where we experiment different models with different set of features. However, instead of raw MRI data, we introduce some pattern- features based on spatial entropy of intensities. For such a set of features, two distinct classifiers, including a deep neural network, are tested. Obtained results, from wide experimentation, show that the new proposed entropy-based features, when it is used with the suited classifiers, allows an accurate MRI segmentation. Keywords - MRI, Pattern-Based, Entropy of Intensities, Classification, Deep Neural Classifier.