Segmentation of kidney in 3D-ultrasound images using Gabor-based appearance models

This paper presents a new segmentation method for 3D ultrasound imag es of the pediatric kidney. Based on the popular active shape models, the algorithm is tailored to deal with the particular challenges raised by US images. First, a weighted statistical shape model allows to compensate the image variation with the propagat ion direction of the US wavefront. Second, an orientation correction approach is used to create a Gabor - based appearance model for each landmark at different scales. This multiscale characteristic is incorporated into the segmentation algorithm, creating a hierarchical approach where different appearance models are considered as the segmentation process evolves. The performance of the algorithm was evaluated on a dataset of 14 cases, both healthy and pathological, obtaining an average Dice's coefficient of 0.85, an average point -to-point distance of 4.07 mm, and 0.12 average relative volume difference. \ Index Terms ² Segmentation, Statistical Shape Model, Kidney, Hydronephrosis, Ultrasound. 1. INTRODUCTION Ultrasound (US) imaging is one of the most widely and conveniently used medical imaging methods. The non -ionizing and non -invasive properties of sonography, along with its real -time nature, safety, and relatively low cost, make US imaging especially useful in the pediatric population. Thus, renal US is one of the most common pediatric US studies, allowing to observe the state of the kidneys and the urinary tract quickly and safely. In particular, the most common abnormal finding in these studies is hydronep hrosis, the dilation of the renal pelvis and calyces due to obstruction of the urinary tract, affecting 2 -2.5% of children [1]. In this context, a n early diagnosis is very important in order to distinguish those kidneys that require surgery from those that do not. Additionally, the accurate parameterization and segmentation of the kidney anatomy plays an important role in the diagnosis of renal diseases [2], [3] and intervention planning. Although the quality of US images has increased in recent years, they still suffer from low signal -to-noise ratio, speckle, signal attenuation and dropout, and missing boundaries due to the orientation dependence of acquisition. These peculiarities make the detection of organs and object of interest form US images particula rly challenging, even when performed manually by a trained expert [4]. The improvement in image quality achieved in recent years has led to an increasing interest in developing new segmentation methods for sonographic images [5]. However, unlike other appl ication areas like echocardiography, or transrectal ultrasound (TRUS), kidney segmentation from renal US has received limited attention from the scientific community. To the best of our knowledge, only a few kidney segmentation approaches have already been reported [6] -[8], most of them focused exclusively on 2DUS. Xie et al [6] presented a 2D segmentation method based on texture and shape priors. From the set of features extracted by means of a Gabor filter bank, they create a texture model using an expect ation -maximization Gaussian -mixture approach, and a shape model to improve the robustness of the method. Though additional experiments with other type of images are provided, the validation of the method was limited. More recently, the work presented by Me ndoza et al. [8 ] takes into consideration the image formation process in US to combine the traditional Active Shape Models (ASM) [9] with a novel texture orientation correction. However, in spite of the promising results reported, the method is only app lied over 2D longitudinal kidney sections, which must be manually selected by an expert radiologist. Martin -Fernandez and Alberola -Lopez [7] proposed a segmentation approach for the kidney in 3DUS images using a probabilistic Bayesian method. However, the slices of the 3D volume are processed independently, trying to identify the kidney contour on each 2D image, and thus, loosing relevant information about the true 3D shape of the organ. Second, the significant user intervention required to adjust the templ ate reduces the automation and the reproducibility of the method. In this paper we present a new variant of the classic ASM, Gabor -based Appearance Models (GAM). This new approach addresses the inherent limitations of the original ASM when dealing with 3DU S renal images. First, a weighted statistical shape model correction compensates the image dependency with the propagation direction of the US wavefront. Second, a new multiscale Gabor based texture model is incorporated in the algorithm, reducing the spec kle noise effect and allowing to detect the kidney contours at 978-1-4673-1961-4/14/$31.00 ©2014 IEEE 633