Three-dimensional labeling of vulnerable regions in carotid plaques using Graph-Cuts

Carotid plaque morphology assessed through high resolution ultrasound is nowadays an important diagnostic approach in the evaluation of stroke risk. The literature describes several methods for the characterization of plaque echogenicity and texture by using overall statistics across the carotid plaques. However, this averaged indicators may not be enough for a correct diagnosis of plaque instability. Therefore, it has been suggested that a local characterization of plaque morphology, where the extension and location of vulnerable regions inside the plaques is taken into account, could lead to significant improvements in the assessment of stroke risk. In this paper, a new method to characterize the morphology of carotid plaques in three-dimensions (3D) is proposed. The local characterization of carotid plaques is performed by using a 3D Graph-Cuts (GraphC) robust labeling method. This methodology allows to binary segment the data by minimizing an energy function that uses spatial correlation among neighboring pixels/voxels in order to remove small misclassified regions. The method is efficient since it is able to find the global minimum of the energy function in a very short time interval. Results show that this labeling procedure is less noisier and favors clustering, being more meaningful from a clinical point of view than the one obtained with simple thresholding. This paper shows that the use of GraphC may improve the 3D morphological characterization of carotid plaques, namely by providing a more appropriate identification of unstable foci inside the plaque.