Kidney Extraction from Ultrasound Images Based on Multi-Scaling and Multi-directional Filters and Shape Model

Abstract

Detecting the kidney tissue in ultrasound images has important role in preventing and diagnosing kidney diseases and complications. Because of the intrinsic noise and low lucidity of such images, kidney segmentation is a difficult and complicated task. We offer an automatic method to extract the kidney tissue from ultrasound images, which is consisted of two parts: pre-processing and post-processing. In preprocessing phase, we use the multi-direction sticks filter to remove the intrinsic noise in ultrasound images, a local function to intense illumination changes and the Curvelet Transform (CT) to smooth the image and noise removal. Finally, in post-processing phase, to segment the kidney’s tissue, shape prior and signed distance functions are used to extract variable shape model. In this shape-based algorithm, segmentation is carried out using calculating the parameters of shape model to minimize the energy function. Using minimizing of energy function, image divides into two regions, textures with low and high variances, inside and outside of the border curve. Many errors occur in methods based on the kidney tissue segmentation with aid of feature extraction because of high similarity between the kidney central tissue and the background tissue in ultrasound images. A brilliant point of our algorithm is to segment the kidney tissue into medial and lateral regions thereby the defects in similar methods are modified considerably. The proposed algorithm is evaluated on a standard set of the kidney ultrasound images. The obtained segmentation results show that more than 89% of pixels inside of kidney’s tissue are extracted correctly.