Real time kidney ultrasound images segmentation: a prospective study
S. Dahdouh
*a
, E. Frenoux
a
, A. Osorio
a
a
Laboratoire d’Informatique pour la Mécanique et les Sciences de l’Ingénieur, Orsay, France.
ABSTRACT
Segmentation of ultrasound kidney images represents a challenge due to low quality data. Speckle, shadows, signal
dropout and low contrast make segmentation a harsh task. In addition, kidney ultrasound imaging presents a great
variability concerning the organ’s shape on the image. This characteristic makes learning methods hard to use. The aim
of this study is to develop a real time kidney ultrasound image segmentation method usable during surgical operations
such as punctures. To deal with real time constraints, we decided to focus on region based methods and particularly split
and merge algorithm. In this prospective study, the selection of the interesting area in the initial image is made by the
physician, drawing a coarse bounding box around the organ. A pre-processing phase is first performed to correct image’s
artefacts. This phase is composed of three major steps. First, an image specification is made between the image to
segment and a reference one. Then, a Haar wavelet filtering method is applied on the resulting image and finally an
anisotropic diffusion filter is applied to smooth the result. Then, a split and merge algorithm is applied on the resulting
image. Both split and merge criteria are based on regions statistics. Our method has been successfully applied on a set of
22 clinical images coming from 10 different patients and presenting different points of view regarding kidney’s shape.
We obtained very good results, for an average computational time of 8.5 seconds per image.
Keywords: kidney ultrasound image, segmentation, speckle, split and merge
1. INTRODUCTION
Ultrasound (US) imaging gained wide spread acceptance in human organs visualization over the past few years. Indeed,
it is a very low cost imaging and non invasive technique with no risk of injury for the patient and no side effects.
However, images acquisition and interpretation are subordinated to manipulator’s judgment and experience; the
interpretation and even quality of an US image may change from a clinician to another. The variations in diagnosis can
be explained in some cases by the very low quality of images which are subject to many artifacts such as low contrast,
speckle, attenuation, etc. Because of all these features, the segmentation of US images using classical methods coming of
artificial vision domain, like active contours or region growing, often fails. In this paper, we focus on kidney images
segmentation in order to help surgeons during kidney punctures.
Kidney images are characterised by the fact that tissue-tissue boundaries of kidney are relatively more difficult to
localize in US images than for other tissues or organs [Xie]. In the recent past, many researches have been conduced on
kidney ultrasound image segmentation, but in a significant percentage of cases, they deal with in vitro kidneys [Bakker,
Matre]. During punctures, kidney is in vivo and parameters are less controllable than in in vitro context. Indeed, in vitro
kidney is merged in a liquid and there is therefore a clear echographical transition between liquid and organ which is not
clearly the case in in vitro situation [Matre]. Moreover, in in vitro situation, there is usually no need to deal with “vital”
movements (e.g. breathing) that degrades images quality. These problems make in vitro methods unreliable for an
application during the operation. Research about segmentation of US kidney images has mainly been conduced these
past years on two directions: region growing and active contours. Because of kidney images characteristics, classical
region growing and active contours methods are unreliable for segmenting those kinds of images. Kidney ultrasound
images are noisy, with a poor signal-to-noise ratio [Bommanna Raja] and kidney boundaries are usually difficult to
locate.
To deal with this problem, some methods using a priori knowledge have been developed, like in [Xie], where authors
present a segmentation framework based on texture and shape a priori knowledge, assuming the fact that a prior model
of anatomical structure will ease the segmentation of those structures on US images. Although, in this paper, it is said
*
sonia.dahdouh@gmail.com
Medical Imaging 2009: Ultrasonic Imaging and Signal Processing, edited by Stephen A. McAleavey, Jan D'hooge
Proc. of SPIE Vol. 7265, 72650E · © 2009 SPIE · CCC code: 1605-7422/09/$18 · doi: 10.1117/12.812493
Proc. of SPIE Vol. 7265 72650E-1