Ultrasound b-scans image denoising via expectation maximization-based unsharp masking

Abstract

In this paper, we present an unsharp masking-based approach with subsequent bilateral filtering stage to noise smoothing of ultrasound (US) image. At our first processing stage, we propose image segmentation via EM to segregate two pixels populations instead of separating original image into the low- and high-frequency components. Our proposed method then enhances the edge by shifting the mean of the two pixels populations away from each other. This is similar to the conventional unsharp masking structure, except that the concept is reformulated and worked in probabilistic setting. At our second stage, we use bilateral filtering to attenuate the retained noise in the flat areas. Performance of synthetic and real clinical B-scan US images based on several dominant image quality measures, e.g., signal-to-noise ratio (SNR) and contrast-to-noise-ratio (CNR), is evaluated. The performance is improved over the conventional US image despeckling methods. The CNR-SNR performance tradeoff is also addressed here for the first time.