IVUS Image Segmentation By Using Expectation-Maximization Approach

Abstract: Now a day’s heart attack is one of the cause of human deaths. It mainly comes from the atherosclerotic plaques. Segmentation of coronary arteries of atherosclerosis is one important process prior to many analyses and visualization tasks for intravascular ultrasound (IVUS) images. The algorithm used in this project includes K-means clustering, Fuzzy C Means (FCM) clustering, and Expectation-Maximization (EM). K-means uses standard Euclidean distance metric, which is usually insufficient for image clustering. Instead in FCM, weighted distance metric utilizing pixel co-ordinates, RGB pixel color and/or intensity and image texture is commonly used. As the datasets scale increases rapidly it is difficult to use K-means and FCM to deal with massive data and Sensitive to noise. This noise could lead to serious inaccuracies in the segmentation result. To overcome this limitation, this project gives a new Expectation-Maximization called Gaussian Mixture Model using Expectation-Maximization (GMM-EM). GMM-EM is implemented through probabilistic approach for smoothening and clustering. Experiment results show that the new algorithm yields better segmentation results.

Keywords: Medical imaging; Segmentation; Intravascular ultrasound (IVUS); Gaussian mixture model using expectation-maximization (GMM-EM).