3D RECONSTRUCTION AND VISUALIZATION OF FEMUR BONE STRUCTURES

Transcription

1 3D RECONSTRUCTION AND VISUALIZATION OF FEMUR BONE STRUCTURES Pedro M. B. Torres 1,2, Paulo J. S. Gonçalves 1,2, Jorge M. M. Martins 2 1 Polytechnic Institute of Castelo Branco, School of Technology, Av. Empresário, Castelo Branco, Portugal. 2 Technical university of Lisbon, IDMEC/IST, Av. Rovisco Pais, Lisboa, Portugal. {pedrotorres paulo.goncalves}@ipcb.pt, martins@dem.ist.utl.pt Abstract: Visualization of 3D volume data, obtained from reconstruction of 2D medical images is a very important concept in different areas, e.g., surgical robotics. Perception of anatomical structures of the human body, without the use of evasive techniques, is an important aspect that can increase the rate of precise diagnosis. The usage of Computer-Assisted Orthopedic Surgical (CAOS) techniques by surgeons is rapidly growing. CAOS enhance the Hip Resurfacing or Hip Replacement process, since automatic systems increase surgical tasks precision and accuracy. The surgeon, to achieve a better implant alignment, should use a precise 3D bone visualization tool. The work, presented in this paper describes the techniques of volume rendering of ultrasound (US) and CT images of femur bones, essential for navigating a robot for orthopedic surgery. Maximum Intensity Projection (MIP) is one of the techniques used. MIP consists of projecting the voxel with the highest attenuation value on every view throughout the volume onto a 2D image. For each XY coordinate only the pixel with the highest Hounsfield number along the z-axis is represented so that in a single bi-dimensional image all dense structures in a given volume are observed. The Visualization Toolkit (VTK) and MATLAB are used to implement the methods and analyse the results. Keywords: Visualization, Surface Rendering, Volume Rendering, Medical Imaging, Surgical Robotics. 1. Introduction Recent years have been marked by the development of robotic systems for medical applications in many areas, ranging from rehabilitation systems to surgical robotic applications. The use of robots to automate medical tasks increases its reliability and accuracy. One of the key parts towards the success of surgical robotic systems is the visualization component, useful in the reconstruction of virtual models of the human organs, identify features in the images acquired, identify tumours, traumatisms, and so on. These components are crucial, for surgical planning and robot navigation control. The work presented in this paper was developed in the scope of the HIPROB project. This project consists on the development of a robotic solution for orthopedic surgery, solving some problems caused by current systems, e.g., ROBODOC [1], ACROBOT [2], that use fiducial markers in order to find the bone in space. This approach is currently the most accurate, but leaves lesions in the patients, manifested after surgery by several pains. With HIPROB is expected to obtain the bone 3D location using information from the ultrasound (US) images acquired intraoperatively and CT images acquired on a preoperative scenario. The paper focus on the threedimensional (3D) reconstruction of several medical image datasets implemented on using the Visualization Toolkit (VTK) and Matlab. Two types of rendering methods are implemented, surface rendering and volume rendering. Different algorithms were used to study both reconstruction methods. This paper is organized as follows. In section II, the concepts of 3D rendering methods are presented. The next section, III, describes a brief description of the VTK. Section IV, describes the experimental evaluation and results. Finally, conclusions and future work are presented. 2. 3D Rendering Methods Three-dimensional rendering is the process of creating a realistic 2D image that intuitively conveys 3D relationships. This section describes the techniques of Surfacing Rendering and Volume The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No

2 Rendering, used in this work. Surface Rendering interprets the datasets by generating a set of polygons that represent the anatomical surface, and displaying a three-dimensional model representation. Polygons representing the outer surface of an object can be identified by running an isosurface detection algorithm (e.g. a Marching Cubes algorithm [3]). This method show 3D relationships most effectively, but suffer from artefacts and fail to effectively display lesions hidden behind overlying bone or located beneath the bone cortex. It also does not display the most important information in the image dataset, because simplifies the data into a binary form, classifying each pixel as either bone or not bone. In other hand, Volume Rendering represents 3D objects as a collection of cube-like building blocks called voxels, or volume elements. Each voxel is a sample of the original volume, a 3D pixel on a regular 3D grid. Each voxel has associated one or more values quantifying some measured or calculated properties of the original object, such as transparency, luminosity, density, flow velocity or metabolic activity. The main advantage of this type of rendering is its ability to preserve the integrity of the original data throughout the visualization process. Using this method is possible show subcortical lesions, minimally displaced fractures, and hidden areas of interest with few artefacts. Depending on the degree of surface shading and opacity, multiple overlying and internal features can be shown, and the displayed intensity is related to the amount of bone encountered along a line extending through the volume. This technique, however, requires huge amounts of computational time and is generally more expensive than conventional surface rendering technique. The present paper is focused in the Maximum Intensity Projection (MIP) Algorithm. This method, first introduced by Jerold Wallis [4] is a Volume Rendering method for 3D data, which projects in the visualization plane the voxels with maximum intensity. The technique consists of projecting the highest intensity captured by the rays perpendicular to the projection of the image, i.e. the method only displays the highest intensity value seen through each pixel. This type of projection is used to highlight the important parts of the image. Figure 1 shows a basic illustration of algorithm: for every pixel (1) in the viewing plane (2) a ray (3) is cast through the volume (4). Although other voxels are intersected by the ray, only the voxel with the highest intensity value (5) is projected onto the viewing plane at that pixel position [5]. Considering a 3D volume dataset f, and project parallel to the z- axis (axial projection) by computing the maximum value. The result is denoted by equation (1): MIP algorithm has been developed and currently there are several variants of the MIP such as Multiresolution Maximum Intensity Projection (MMIP) algorithm [6], Depth-shaded Maximum Intensity Projection (DMIP) [7] or Local Maximum Intensity Projection (LMIP) [8][9], which allow multiresolution, and different levels of transparency. Figure 1 shows an illustration of MIP algorithm in its simplest form. (1) Figure 1 a) Maximum Intensity Projection overview. Figure extracted from [5]. b) example of projection of voxel intensity. 3. Brief Description of the VTK The Visualization Toolkit (VTK) [10] is an opensource, object-oriented software system for 3D computer graphics, image processing and visualization. It is provided as a C++ library with interfaces to the interpreted languages such as Tcl, Python and Java. In this work, Microsoft Visual 52 The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No. 41

3 Studio 2010, with VTK, is adopted to reconstruct the 3D images using the 2D CT image sequence in DICOM format. Conventional 3D visualization tools usually have important limitations such as low efficient code execution, poor computing capacity. VTK overcomes these limitations, because has constantly evolved, has fast algorithms and computationally efficient. 4. Experimental Results This section presents the bone reconstruction results, performed with VTK and Matlab, for CT and ultrasound (US) images. For CT reconstruction, femur bone reconstruction is performed by 360 CT two-dimensional (2D) crosssection images, with scanning interval of 0.75 mm between them, acquired by a Siemens SOMATON Sensation 16 CT Scanner, with image resolution of The images were stored in DICOM (Digital Imaging and Communications in Medicine) standard format, the most common in medical images management. The use of this standard facilitates the communications between different types of medical equipment and it is possible to store images in different datasets, useful to reconstruct different views. With the reconstruction is possible to visualize the bone in 360º, for complete visualization, in its total length (270 mm) and identify possible lesions, if existing. To reconstruct the volume of the bone it is necessary to load images (vtkdicomimagereader class), rendering the images (vtkrenderer class) and finally create a volume (vtkvolume class). Figure 2 shows a CT slice and a US image of the femur bone. Figure 3, represent the volume reconstruction of bone, performed with the MIP algorithm for the CT images. Figure 2 a) An original CT image and b) US image. Figure 3 Femur Bone Reconstruction, with a high level of transparency (MIP). The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No

4 The application developed allows combine colours and opacity of the representation and thus highlight the parts that are most interesting finding in the representation. Figure 4 presents an example where it intends to highlight more the bone definitions, giving less importance to the interior, unlike the results of Figure 3, which is very transparent. Figure 4 Femur Bone Reconstruction, with highlight bone definitions. Figure 5 shows a section of the bone highlighting the cross section and view inside the bone. These representations have a good image quality showing a perfect bone constitution, inside and outside. It is visible the nodules of the femur in a perfect and realistic reconstruction. To compare the reconstruction results, obtained from CT images and US images, of the femur, extensive tests were performed using Matlab and VTK. For this, the Volume Rendering and Surface Rendering algorithms, described in section II, were implemented. US images, of the femur, were acquired with Aloka Prosound 2 equipment with a 5 MHz probe placed in the end effector of a robot, where the probe was moved at a constant speed. 489 Figure 5 Cross section Femur Bone Reconstruction. US images of the femur were acquired, with 0.55 mm between them. After processing the images, reconstruction was performed. Figure 6 shows the Volume Rendering of CT images (a) and US images (b). As ultrasound does not penetrate the bone, in this type of image is only possible to view the upper surface of the femur. The results obtained from the Matlab implementation, for the volume reconstruction with the CT images, have less quality 54 The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No. 41

5 than the results achieved with VTK and the computational cost is much higher. Surface Rendering technique was only implemented in order to compare the results with Volume Rendering technique. Figure 7 shows the results of surface rendering obtained with CT images and US images. To the reconstruction with CT images unable to reconstruct the total dimension of the bone, because computational cost was very high and Matlab was not enough memory to process so much data. Figure 6 Bone Reconstruction (Volume Rendering) with a) CT images and b) US images. Figure 7 Bone Reconstruction (Surface Rendering) with a) CT images and b) US images. 5. Conclusions and Future Work The reconstruction based on VTK shows much better results than other types of software and is best suited to reconstruct medical images. This paper discusses 3D reconstruction and gives an example of reconstruction and visualization from CT images and US images. Volume rending plays an important part in reconstructing the bone and perform better results than Surface rendering. The results demonstrate that VTK is a robust tool to integrate the visualization component of HIPROB project. As The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No

6 future work is expected to perform bone reconstruction in VTK with ultrasound images and integrate the components of image registration between CT datasets and US datasets. Acknowledgments This work was supported by: Strategic Project, reference PEst-OE/EME/LA0022/2011, through FCT (under the Unit IDMEC - Pole IST, Research Group IDMEC/LAETA/CSI): by FCT, under the project PTDC/EME-CRO/099333/2008;, IDMEC; by the European Union FP7 Project: ECHORD, experiment HIPROB. 6. References [1] Kazanzides, P.; Zuhars, J.; Mittelstadt, B.; Williamson, B.; Cain, P.; Smith, F.; Rose, L.; Musits, B.;, "Architecture of a surgical robot," Systems, Man and Cybernetics, 1992., IEEE International Conference on, vol., no., pp vol.2, Oct [2] Davies, B.L.; Fan, K.L.; Hibberd, R.D.; Jakopec, M.; Harris, S.J.;, "ACROBOT - using robots and surgeons synergistically in knee surgery," Advanced Robotics, ICAR '97. Proceedings., 8th International Conference on, vol., no., pp , 7-9 Jul [3] William E. Lorensen, Harvey E. Cline: Marching Cubes: A high resolution 3D surface construction algorithm. In: Computer Graphics, Vol. 21, Nr. 4, July 1987 [4] Wallis, J.W.; Miller, T.R.; Lerner, C.A.; Kleerup, E.C., "Three-dimensional display in nuclear medicine," Medical Imaging, IEEE Transactions on, vol.8, no.4, pp , Dec [5] Stefan Bruckner, Performing Maximum Intensity Projection with the Visualization Toolkit, Seminar Paper, Austria, 2002; [6] Roerdink, J.B.T.M.;, "Multiresolution maximum intensity volume rendering by morphological adjunction pyramids," Image Processing, IEEE Transactions on, vol.12, no.6, pp , June 2003; [7] Heidrich, W.; McCool, M.; Stevens, J.;, "Interactive maximum projection volume rendering," Visualization, Visualization '95. Proceedings., IEEE Conference on, vol., no., pp.11-18, 433, 29 Oct-3 Nov [8] Yoshinobu S., Nobuyuki S., Shin N., Shinichi T., Ron K., Local Maximum Intensity Projection (LMIP): A New Rendering Method for Vascular Visualization, in Journal of Computer Assisted Tomography, Vol. 22, N. 6, pp , November/December [9] Stefan Bruckner and M. Eduard Gröller, Instant Volume Visualization using Maximum Intensity Difference Accumulation. Eurographics / IEEE-VGTC Symposium on Visualization, Vol. 28, n. 3, [10] seen in June 2012; 56 The Romanian Review Precision Mechanics, Optics & Mechatronics, 2012, No. 41