Cadaver validation of intensity-based ultrasound to CT registration

Abstract

A method is presented for the rigid registration of tracked B-mode ultrasound images to a CT volume of a femur and pelvis. This registration can allow tracked surgical instruments to be aligned with the CT image or an associated preoperative plan. Our method is fully automatic and requires no manual segmentation of either the ultrasound images or the CT volume. The parameter which is directly related to the speed of sound through tissue has also been included in the registration optimisation process. Experiments have been carried out on six cadaveric femurs and three cadaveric pelves. Registration results were compared with a “gold standard” registration acquired using bone implanted fiducial markers. Results show the registration method to be accurate, on average, to 1.6 mm root-mean-square target registration error.

Introduction

Recent years have seen the emergence of image-guidance systems for orthopaedic surgery (Amiot and Poulin, 2004, Barger et al., 1998, DiGioia et al., 1998, Jaramaz et al., 1999, Nabeyama et al., 2004). All of these systems require a registration between physical and image space. Most systems achieve this by identifying point landmarks (either fiducial markers or anatomical positions) or by delineating surfaces using a tracked pointer. In order to achieve accurate registrations, bony landmarks or fiducial markers attached to bone should be located. Implanting fiducials or exposing additional bone surface for registration is invasive and may cause additional pain (Honl et al., 2003, Nogler et al., 2001) and increase the risk of infection, whereas limiting surface information to regions exposed in standard procedures may adversely affect registration accuracy (Heger et al., 2005). Therefore, there is a trade off between the invasiveness and the accuracy of the registration process. The drive to develop less invasive registration methods will increase with the adoption of minimally invasive surgical techniques (Berger, 2003, DiGioia et al., 2003). This is very important for computer assisted orthopaedic surgery (CAOS) systems, as one of the main arguments for their use is that they have the potential to greatly reduce the invasiveness of procedures (DiGioia and Nolte, 1998).

A number of authors have proposed using imaging devices to acquire information on the position of bony structures in the operating room. The use of X-ray or fluoroscopy images has been put forward with promising results (Guéziec et al., 1998, Livyatan et al., 2003). Proposals to use ultrasound (US) for registration in image-guided surgery go back approximately a decade (Barbe et al., 1993, Ault and Siegel, 1994, Lavallée et al., 1995). Tracked US has advantages over X-ray imaging as it can acquire 3D information, rather than 2D projections, is typically cheaper and more portable than X-ray equipment, and is non-ionising. However, there are a number of image artefacts associated with US: speckle noise, saturation of the reflected echo at the bone-tissue boundary and variation of the speed of sound in different tissues can all make precise location of the bone surface difficult.

Methods to match B-mode US images to bone have been reported by a number of authors. The main application areas have involved registrations on vertebrae (Barbe et al., 1993, Lavallée et al., 1995, Muratore et al., 2002, Brendel et al., 2002, Ionescu et al., 1999, Kowal et al., 2003), pelves (Amin et al., 2003, Ionescu et al., 1999, Tonetti et al., 2001) and long bones (Ault and Siegel, 1994, Brendel et al., 2003, Jaramaz et al., 2003). The work presented in this paper differs from previously published work in three main ways:

• We have used an intensity-based algorithm, therefore no segmentation is required in either the US or CT volume. All previously published US to CT bone registration algorithms have required segmentation of the CT volume. This does not affect the clinical utility of such approaches, as the segmentation can be carried out prior to the procedure, and automated segmentation methods exist (Kang et al., 2003). However, segmentation errors can occur, particularly at joint interfaces, which may affect registration accuracy. Some algorithms also require segmentation of the US images. This is more problematic to the clinical process as, due to time constraints during a procedure, the segmentation must be carried out quickly. Manual segmentation is therefore not feasible, and accurate automated segmentation of US images is a challenging problem, although some techniques have recently been published (Daanen et al., 2004). Another incentive for developing an intensity-based technique is that we believe it has the potential to produce more accurate results than techniques which require segmentation of one or both modalities. The improved accuracy of intensity-based compared to surface-based registration techniques has been shown for CT to MR registration of head volumes (West et al., 1999).

• The algorithm presented here also optimises one of the probe calibration parameters: the parameter which is directly related to the average speed of sound within the US imaged medium.

• Our validation strategy uses human cadavers, results are compared to an independently calculated “gold standard” registration based on bone implanted fiducial markers, and we use clinically realistic starting positions. Much of the previous work has either used dry or plastic bones in water baths (Barbe et al., 1993, Ault and Siegel, 1994, Lavallée and Szeliski, 1995, Muratore et al., 2002), where the lack of soft tissue structures and easy access to all of the bone surface greatly simplifies the registration process. The use of human cadavers enables us to acquire a set of US images which should closely represent a clinical dataset in two ways: firstly in terms of the individual image characteristics, e.g., speckle, distortion, speed of sound variations; and secondly in terms of in which regions of human femur and pelvis is it possible to obtain clear images of the bone surface using US. Previous studies have used a number of validation methods: quoting residual errors (Lavallée and Szeliski, 1995), visual inspection (Lavallée and Szeliski, 1995, Brendel et al., 2002), positioning surgical instruments (Barbe et al., 1993), comparison with an image-guided surgery system (Amin et al., 2003, Kowal et al., 2003) and the use of fiducial markers (Muratore et al., 2002, Jaramaz et al., 2003). However, fiducial markers have only been used previously with either dry bones in water baths (Muratore et al., 2002) or animal cadavers (Jaramaz et al., 2003). An ultrasound-based computer-assisted surgery system has been used clinically for the positioning of iliosacral screws on four patients (Tonetti et al., 2001). Postoperative CT volumes were used to verify screw positioning, and hence also provided a validation of registration accuracy. Promising results were obtained, however the registration algorithm required manual segmentation of the US images, a process which they describe as being “very delicate” and time consuming.