Ultrasound Image Despeckling and Enhancement using Modified Multiscale Anisotropic Diffusion Model in Non-Subsampled Shearlet Domain

Abstract

Ultrasound imaging is undoubtedly the most used imaging modality for diagnostic purposes. Unfortunately, it is accompanied by speckle which can degrade texture information by obscuring fine details like boundaries and edges. This work presents a method for despeckling ultrasound images by treating them with multiscale modified speckle reduction anisotropic diffusion model and Non-Subsampled shearlet transform (NSST). The method involves division of images using a non-subsampled Laplacian pyramid. This results in low and high frequency image components. Modified anisotropic diffusion is used on the low frequency part. The high frequency component, as subjected to shearlet function, generates noisy coefficients in various directions. These coefficients are further subjected to NSST thresholding. The denoised low and high frequency image components are then recombined to obtain the enhanced image. This multidimensional and multidirectional method improves the qualitative characteristics of ultrasound images by not just removing speckle noise but also by preserving edges, thus resulting in effective image enhancement. Performance of the method is analysed on synthetic and real medical ultrasound images. Results reveal that the proposed method exceeds the state-of-the-art methods in the context of edge preservation and structural similarities, and thus, it is an effective aid to radiologists in their clinical diagnosis by providing an enhanced denoised image.