Multinomial level-set framework for multi-region image segmentation

Abstract

We present a simple and elegant level-set framework for multi-region image segmentation. The key idea is based on replacing the traditional regularized Heaviside function with the multinomial logistic regression function, commonly known as Softmax. Segmentation is addressed by solving an optimization problem which considers the image intensities likelihood, a regularizer, based on boundary smoothness, and a pairwise region interactive term, which is naturally derived from the proposed formulation. We demonstrate our method on challenging multimodal segmentation of MRI scans (4D) of brain tumor patients. Promising results are obtained for image partition into the different healthy brain tissues and the malignant regions.