Classification of contrast ultrasound images: improvement of the GMM using gaussian filter

Abstract

Contrast agent microbubbles play an important role in ultrasound images. These agents are small and safe, and are administered intravenously in the systemic circulation to enhance the vascular zone of interest. The backscattered signals issued from the agent area are not considerable enough to be differentiated from the backscattered signals derived from the surrounding tissues. Therefore, applying further image processing techniques is mandatory to improve the visualization of the contrast ultrasound images, to achieve a satisfactory classification, and even to quantify the agent concentration in the perfused zone. The nonlinear physical property of the agent results in a relatively high backscattered signal with respect to the tissue characterized by a quasi-linear response. In this paper, a Gaussian mixture model (GMM) is used to discriminate between the biological tissues and the contrast agent. The efficiency of the GMM classifier is obviously consolidated if the data distribution fits to Gaussian. Therefore, we propose to apply a Gaussian filter as a pre-processing phase that allows the nonGaussian distribution to match slightly the Gaussian. Consequently, Gaussian filter lends positive impact on the quality of the image and improves the performance of the GMM classification of the contrast-ultrasound images.