Improving the Accuracy of MR Image Segmentation through the use of Local Gradient Information

Segmentation is a core technology in medical image analysis, providing a route to tissue volume estimation that can be used in a wide range of applications, such as monitoring the progress of, or the effects of drug therapies on, tumour growth or the effects of atrophic diseases. In general, the more image information we can extract to use in segmentation, the more accurate the results will be. In previous work we have proposed a unified mathematical framework for incorporating local image gradient into feature-space based segmentation algorithms. In this paper we demonstrate, using simulated MR images of the normal brain, that the additional information present in the gradients can significantly improve segmentation accuracy.