Segmentation of Medical Images with a Combination of Convolutional Operators and Adaptive Hidden Markov Model

Abstract

Medical image segmentation is a problem of fundamental importance in medical image processing. The accurate segmentation of a medical image can provide important information for the diagnosis and treatment of many diseases. Since a medical image often contains noises and the objects in it are inherently complex in general, methods that can accurately segment an arbitrary medical image are still unavailable. In this paper, a new approach that combines convolutional operators and an adaptive Hidden Markov Model is developed for the segmentation of medical images. Specifically, the features associated with each pixel in a medical image are obtained with a set of convolutional operators. The semantic and spatial correlations among pixels in the image are then progressively captured by an adaptive Hidden Markov Model. The labels of the pixels can be efficiently obtained with a dynamic programming algorithm in linear time. Our experimental results show that this approach can achieve segmentation results with improved accuracy on a set of brain medical images.