Speckle Reduction of Ultrasound B-mode Image using Patch Recurrence

Abstract

In this paper, a novel speckle reduction method using the patch recurrence (SRPR) of the medical ultrasound B-mode image with the self-similarity is proposed. The SRPR utilizes the additive white Gaussian noise to model the speckle and provides the despeckled ultrasound B-mode image from the similar patches based on minimum mean square error estimation (MMSE). It also improves the performance of the edge preservation and speckle reduction using the local variance of the patch which adjusts the threshold of the Euclidean distance that determines the similarity between the patches. In the MMSE process, the proposed SRPR has a low computational complexity since it uses pre-calculated local information on the 2D input image when extracting statistical information from 3D data set composed of similar patches, thus it offers advantages for hardware logic implementation and real-time processing. From the simulation study, the proposed SRPR shows improved results of signal-to-noise ratio (SNR) and similarity quality measurement (SSIM) compared with conventional methods. The visual assessment and contrast-to-noise ratio (CNR) of real ultrasound B-mode images (e.g. thyroid images) also show that the proposed method is superior to the previous methods.