COMPUTER ANIMATION AND VIRTUAL WORLDS Comp. Anim. Virtual Worlds 2014; 25:255–269 Published online 16 May 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/cav.1586 SPECIAL ISSUE PAPER A haptic-enabled novel approach to cardiovascular visualization Shamima Yasmin1*, Nan Du2, James Chen3 and Yusheng Feng4 1 2 3 4 Computer Science and Engineering, Arizona State University, Tempe, AZ, USA Electrical and Computer Engineering (ECE), University of Texas at San Antonio, San Antonio, TX, USA Division of Cardiology, Department of Medicine, University of Colorado, Denver, CO, USA Mechanical and Bioengineering, University of Texas at San Antonio, San Antonio, TX, USA ABSTRACT Intravascular ultra sound (IVUS) imaging technique is widely used for the detection of plaque deposit inside coronary artery wall. Adequate detection of plaque deposit is necessary for further treatment of the patient, but image by image analysis, nevertheless, is cumbersome. In this paper, we proposed a fully automatic novel method for coronary artery visualization, which takes as an input a number of intravascular ultrasound images to construct the 3D model of coronary artery wall where the plaque deposit is demarcated in a rapid 3D rendering environment. This 3D visualization of coronary artery model has been enhanced in a haptic environment where the user is not only able to visualize the fatty plaque deposit but is also able to feel and differentiate stiffness of plaque deposit as well as soft, elastic property of normal artery through a virtual tour along the artery pathway. Copyright © 2014 John Wiley & Sons, Ltd. KEYWORDS visualization; IVUS images; coronary artery; 3D modeling; haptic *Correspondence Shamima Yasmin, Computer Science and Engineering, Arizona State University, Tempe, AZ, USA. E-mail: shamima.yasmin@asu.edu 1. INTRODUCTION Intravascular ultrasound (IVUS) is a catheter-based medical imaging technique that provides high resolution, cross-sectional images of the interior of blood vessels in real time. IVUS is most frequently used by cardiologists to diagnose the interior details of coronary artery of a patient. Atherosclerosis is a disease characterized by a deposit of plaque in coronary arterial wall which may become hardened over time. Soft plaque may rupture and is considered to be the most frequent cause of heart attack and sudden cardiac death. Hence, IVUS is a valuable imaging tool in the diagnosis and analysis of the development of plaque deposit inside the coronary artery wall. It has been shown that atherosclerosis neglected during angiography can be detected precisely with IVUS. IVUS can generate thousands of cross-sectional images that possess valuable information about the extent of stenosis to facilitate subsequent treatment planning, e.g., stent deployment. However, it is very time consuming and challenging for a physician to analyze on-line a great number of 2D images. Although a subset of images can be selected for analysis, it is still awkward to manually quantify and outline the boundaries and compositions of Copyright © 2014 John Wiley & Sons, Ltd. plaque in each image; especially when the plaque morphology is very irregular in nature. In the clinical application, physicians will be the utmost users to use the software utility. They generally wish a one-button approach so that the required results can be calculated and shown. Hence, an automated technique is needed to quickly and accurately process a large number of IVUS cross sections within an acceptable time frame. 2. BACKGROUND Segmentation of IVUS cross-sectional images of coronary artery refers to the identification of the lumen/intima and media/adventitia boundaries. On-line or real-time segmentation of IVUS images is quite challenging because their differences in gray levels are not prominently shown in a uniform fashion between consecutive images. A number of semi-automatic/automatic methods have been developed to perform segmentation of IVUS images. Some methods consider the first image of the series to be manually outlined and served as an active contour or a trained model to extract the desired region-of-interests in the rest of the series of the images [1,2]. Specifically, subsequent contours are gradually 255 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular evolved at discrete intervals with regard to the active contour until the optimum energy function is met so that corresponding boundaries can be captured. The active model based approaches may fail when there exist substantial differences among the images as it may necessitate the building of a new active contour model. Unal et al. [3] used prior knowledge of the expected IVUS pattern by building an active shape model, which is the average shape obtained from training data set. Mean lumen shape and mean adventitial shape are constructed and deviation of each image from the mean shape is observed by principal component analysis. Hence, each shape is defined by the mean shape with varying weighted factors (i.e. distance to the catheter or radial shift, negative/positive amplification of the curving, and angular shift) associated with each of them. Edge enhancement by anisotropic diffusion was incorporated in a number of methods [4,5] from which approximate initial edge is determined first. Afterwards the final contour evolved from the approximate contour either by statistical discriminant measure of weighted image separability or parametric/geometric deformation in the form of energy minimization. Unfortunately, such a method may not be able to create the initial contour properly when noises in image are too high. Papadogiorgaki et al. [6] generated intensity and texture information by multilevel discrete wavelet frames decomposition to initialize a contour. The process of contour initialization involves several iterations followed by visual evaluation of filtered image at different iteration levels. Low-pass filtering and radial basis functions are introduced for producing smooth contours. This involves some preprocessing for noise removal, visual selection of optimal contour, and may fail where a branching pattern takes place. Brusseau et al. [7] modeled the image brightness based on Rayleigh distribution to facilitate separation of blood from tissue. Optimal separation of luminal region is carried out by comparing different statistical properties in term of the likelihood function when it reaches the maximum. However, the proposed technique was lack of the process of media adventitia border detection and may not work properly if catheter-center difference between images was not aligned. Cardinal et al. [8] employed a fast marching method in which the boundary of the region propagated according to a speed function dependent on the magnitude of the local image gradient. Upon approaching to the image boundary, the speed value became low while the gradient magnitude became high. As a result, the regions on the consecutive IVUS images were semi-automatically identified by the fast marching propagation method. Wennogle et al. [9] proposed an improvement over this method by using a directional gradient velocity term instead of gradient magnitude in the speed function followed by a level-set technique as post processing for smooth contour generation. Apart from manual involvement in both of these methods, segmentation in 3D environment may fail when catheter centers among the images are not aligned. Too much compromising to align the images may also result in loss of information. 256 All the methods discussed above have some advantages versus shortcomings under different image contents. As the number of images is large, it would be useful if a 3D model can be automatically generated and well visualized from a series of 2D IVUS images to examine the morphology and composition of plaque in detail. Such a 3D environment not only can minimize the time required for reviewing the images, but also improve understanding of the details of coronary artery plaque to facilitate faster and better treatment plan. Apart from visualization in the 3D environment, if the model can be haptically enabled, it can be greatly enhanced as a tool where a hands-on training can be developed to emulate interactions with tissue properties of vessel lumen, that is, normal, semi-normal, or fully plaqued region, the extent of artery blockage, delivery of stent, and so on. In this paper, the technique of a fully automatic 3D construction of a patient-specific coronary artery model from a series of 2D IVUS images is proposed, from which the coronary plaque deposit can be delineated properly in a 3D visual-haptic environment. 3. PROPOSED METHOD The proposed visual-haptic process is divided into the following three major steps: • Image pre-processing, • 3D reconstruction, and • Haptic modeling. Each step will be discussed in detail in the following paragraphs. 3.1. Image Pre-processing This step can be further divided into the following two sub-steps: (1) extraction of media/ adventitia /vessel contour and (2) extraction of lumen contour. 3.1.1. Extraction of Media/Adventitia/Vessel Contour Two observations have been made for the extraction of vessel contour. Firstly, the center of catheter may not locate at the center of the vessel in IVUS images and secondly, the average vessel diameter approximately varies from 85 to 90 pixels (Figure 1). As the image is initially thresholded, a faint circular region of low pixel value (approximate pixel intensity less than 70) surrounded by high pixel value (approximate pixel intensity greater than 140) in the middle is detected (Figure 2). The circular region represents the vessel boundary that needs to be segmented out. It is observed that in Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 1. Intravascular ultrasound coronary artery images with catheter center (left) aligned and (right) not aligned with the vessel boundary. Figure 2. Initial thresholding of the intravascular ultrasound image to get a better understanding of further segmentation. images where the center of the vessel is aligned with catheter center, this circular region roughly coincides with the vessel boundary, but there are cases where catheter center is far off from the vessel center. In that case, vessel center needs to be determined first in order to extract the vessel boundary. A circle with an approximate artery radius is considered with initial center positioned at catheter center, that is, the center of the image. The circle center is gradually shifted at discrete intervals within a small region around the catheter center. As the center is gradually shifted, the point at which the center of the circle coincides with the center of the vessel is called the optimum position (Figure 3). At optimum position, the following two properties of image contents can be concluded as: Firstly, the difference in peripheral pixel intensity for a small increase and for a small decrease in radius becomes the maximum (Figure 4). Secondly, there is a uniform increase in pixel intensity along the periphery for a small increase in radius (Figure 5). Let Arij be the sum of pixel intensities within a circle with radius r and center (i, j). Similarly, Aijr Δr is the sum of pixel intensities for a circle centered at (i, j) with radius ‘r ∆r’. LetRij Δr ¼ Arij Aijr Δr be the difference in peripheral pixel intensity for a small decrease in radius. Similarly, the difference in peripheral pixel intensity for a small increase in radius ‘∆r’ is defined as RþΔr ¼ AijrþΔr Arij. Thus, the optimum radius r* ij at optimum center (i, j)* is given by the following  argmax RþΔr ij r∈½rbegin ;r end Š;ij∈½ijmin ;ijmax Š Rij Δr  where rbegin , rend approximate range of vessel radius and ‘∆r’ represents discrete step interval of 2 pixels. Similarly, center Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 257 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 3. Alignment of the catheter center with the vessel center. Figure 4. Measurement of peripheral pixel intensity difference at optimum position. (i, j) varies from 15 to +15 along x and y directions with initial center at catheter center, that is, origin (0, 0). After achieving the optimum center for an optimum radius, the initially thresholded higher pixel region is shaved off. The convex hull of the interior lower pixel region yields the vessel boundary (Figure 6). 3.1.2. Extraction of Lumen Contour After the vessel boundary is extracted, the optimum center and radius for each vessel boundary are used to extract the new region of interest (ROI), as shown in Figure 7(b). Hence, for the extraction of lumen contour, further processing is carried out only on the region of interest and the outer part of the image is removed. In order to extract the contour of the lumen region, a series of operations, that is, smoothing, thresholding are applied followed by morphological operations which will be discussed in brief next. Smoothing operation on the extracted ROI is performed first (Figure 8 and Figure 9). For each pixel p(i, j) in an image matrix p, the average of its eight immediate neighbors is calculated as follows: 258 1 8 average¼ ½pði 1; j 1Þ þ pði; j 1Þ þ … þ pði þ 1; j þ 1ފ The intensity of each pixel is updated by the difference between its value and the average of its eight immediate neighbors. The smoothing process for each pixel p(i, j) is carried out as follows: pði; jÞ ¼ s
pði; jÞ þ ð1 sÞ
average where if s < 1, the image is blurred s = 1, the image is unchanged s > 1, the contrast is heightened Once a smoothed image is obtained (Figure 10(a)), a global threshold is selected in an iterative manner by using ‘Isodata’ algorithm [10]. The essence of this method is to begin by dividing the histogram of the ROI into two parts using a starting threshold value, that is, the sample mean of the ROI. Two sample means of the gray values associated with Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 5. A uniform increase in pixel intensity defined by blue color region along the periphery of the artery boundary for a small increase in radius, that is, ‘Δr’. Figure 6. Gradual extraction of artery boundary. the foreground pixels and with the background pixels are computed. A new threshold value is then computed as the average of these two sample means. Based upon the new threshold, the process is repeated until the threshold value becomes unchanged. The selected threshold value is then used to convert the grayscale image into a binary image as shown in Figure 10(b). Based on the binary image, a series of morphological operations are applied to extract the lumen region. Firstly, an erosion operation is performed followed by a dilation operation, to roughly extract the lumen region (Figure 11(a)). An Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 259 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 7. (a) Original raw image, (b) The extracted region of interest (ROI). opening operation is then used to select and preserve the intensity patterns while attenuating others (Figure 11(b)). In order to refine and smooth the boundary of the lumen region, a median filter and a dilation operation are used. The result is shown in Figure 11(c). In order to remove the catheter-induced artifacts, the convex hull of the extracted lumen region is computed, and the difference between the convex hull and the extracted lumen region is calculated to identify and fill the gap induced by the catheter (Figure 12). Next, the connected components are labeled and the lumen boundary is extracted (Figure 13). 3.2. 3D Reconstruction Figure 8. Each pixel p(i, j) and its eight neighbors are considered while carrying out the smoothing operation. After the image processing steps as described in the previous section, a series of vessel and lumen boundaries are extracted (Figure 14). IVUS needs to be associated with angiography so that image alignment follows the proper course of artery pathway. Due to lack of alignment information associated with each IVUS cross section, all the extracted boundaries Figure 9. Smoothing operation is carried out on region of interest at different ‘s’ values. 260 Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 10. (a) The resultant image after applying the smoothing process. (b) The binary image after thresholding the smoothed image in (a). Figure 11. Gradual extraction of lumen region. Figure 12. The process of identifying gap induced by the catheter. are simply stacked up along the corresponding center of the image plane as shown in Figure 14(b). Afterwards a surface reconstruction process for both inner lumen and outer vessel boundaries is performed. For surface reconstruction, we need to model the inner (lumen) and the outer (vessel) layers separately from the extracted lumen and vessel boundaries, respectively. Two consecutive boundaries, that are lumen boundaries or vessel boundaries, may not have equal number of points. For smooth transition between two consecutive boundaries, four control points are located along four quadrants of each boundary (Figure 15). Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 261 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 13. (a) The original region of interest and (b) the segmented lumen. Each quadrant consists of equal number of points. After equalizing the boundary points between two consecutive boundaries, triangulation is performed (Figure 16). For smooth transition between two consecutive boundaries, two consecutive boundaries are interpolated and a number of intermediate boundaries are constructed in-between the source and target boundaries shown as ‘Boundary 1’ and ‘Boundary 2’, respectively, in Figure 17. Extracted lumen boundaries can be very irregular in shape and may not be fully continuous within the vessel boundary as shown in Figure 18. In the case of partial plaque formation, some portion of lumen boundary resides within the vessel, whereas the rest consists of soft vessel tissue. For ease of surface reconstruction, extracted vessel boundary is given a small thickness and the outer vessel boundary is formed. Next, discontinuous portions of lumen boundary are connected and extended to be in-between the inner and the outer vessel boundaries (Figure 19). The extended lumen boundary remains hidden in-between the inner and the outer vessel boundaries. This way, nonplaque portions of soft tissue and irregular plaque portion inside the vessel boundary can be properly outlined and separated in the surface reconstruction process. Figure 14. (a) Vessel (red) and lumen (yellow) contours overlaid on an image and (b) all vessel and lumen contours are stacked up. Figure 15. Equalization of boundary points. 262 Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 16. Triangulation is performed after equalizing the boundary points between two consecutive boundaries. Figure 17. Formation of intermediate interpolated boundaries from source (Boundary 1) and target (Boundary 2) boundaries performed in-between two consecutive lumen boundaries when irregular, discontinuous plaque is formed inside the vessel wall. 3.3. Haptic Modeling Figure 18. Discontinuous lumen boundary (yellow) and continuous vessel boundary (red). While surface reconstruction, extended lumen boundary along with the original extracted portion is considered as a whole and divided into four quadrants (Figure 20) as explained before. Thus, surface reconstruction is After 3D reconstruction of the model, haptic properties are assigned to the constructed vessel and lumen surface. Vessel surface consists of inner and outer boundaries. Depending on the plaque formation, lumen surface may totally be confined within the interior vessel boundary or some part of lumen surface may be hidden in-between the inner and the outer vessel boundaries as can happen when plaque deposit becomes irregular, discontinuous. In both cases, one haptic lumen surface is formed. For irregular plaque configuration, only intercepted portion of the lumen surface is perceived haptically. A normal artery without any plaque deposit feels elastic and soft. The region of plaque in the artery Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 263 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular 4. RESULTS Figure 19. Discontinuous lumen boundary is extended to be continuous. Figure 20. Vessel contour (red) and extended lumen contour (yellow). will feel rough and stiff. Thus, irregular plaque deposit intermingled with soft vessel tissue can be differentiated haptically (Figure 21). A series of thirty IVUS images have been tested. Figure 22 shows the result of the extracted vessel boundaries for 30 IVUS images. Figure 23 shows the extracted lumen boundaries of the corresponding images. While extracting lumen boundary, only a portion of the image demarcated with a square of side ‘r’ centered at the vessel center has been considered. The proposed algorithm shows promising result as commented by cornonary imaging expert. Deviation from the actual vessel boundary is almost negligible. Further work is being carried out to make the lumen boundaries work perfect and error-free for irregular or partial-plaqued IVUS images. Figure 24 (a) shows the normal artery model (without plaque formation) constructed from the extracted vessel boundaries. A small thickness has been assigned to the artery wall. Another interior tube is constructed from the lumen boundaries as shown in Figure 24 (b). In the figure, plaque deposit has been marked yellowish. Figure 24(b) shows the combined constructed model where the inner lumen tube has been placed inside the artery model. The transverse cross-section to the right of Figure 24(b) shows that the artery is heavily plaqued. Figure 25 shows the haptically constructed 3D coronary artery model where a catheter is inserted to simulate the navigation through the artery tunnel. A user can make a virtual tour through the artery tunnel to have better and closer interaction in an immersive haptic environment. While probing through the artery tunnel, the haptic catheter can distinguish and differentiate normal, soft artery tissue from plaque. Normal artery tissue feels soft and elastic. With the haptically constructed 3D model, the user can feel the extent of plaque deposit. The user can also examine the physical properties of plaque deposit. HLAPI interface has been used to represent two polygonal models: lumen tube and vessel tube. In order to avoid buzzing noise while Figure 21. Formation of irregular and stiff plaque deposit inside soft vessel wall. 264 Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 22. Extracted vessel boundaries for all 30 images. Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 265 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 23. Extracted lumen boundaries for all thirty images. 266 Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 26. Fanning in coronary artery while branching out. Figure 24. A 3D coronary normal artery model (a) without plaque deposit and (b) with plaque deposit and transverse Xsection. haptic interaction, a coarse polygon defined by its oriented bounding box has been rendered haptically, whereas for visualization, polygons have been displayed with minute details. The anatomy of coronary arterial model looks like a tree structures, that is, a main trunk (e.g., left descending artery or left circumflex artery) with a few secondary branches (e.g., diagonal artery, septal artery, or obtuse marginal artery, etc.). When the tip of IVUS catheter is near the bifurcation or junction of the main and side-branch region, the fan beam signal comes across from the target vessel to the side-branch as shown in Figure 26. Fanning may cause a sudden change in the vessel diameter. The proposed algorithm can consider the branching of artery successfully. The approximate vessel diameter while working on the extraction of vessel boundary is given a wider range to meet this issue so that the cross-sectional change in vessel can be correctly determined. Figure 27 shows how fanning can be handled automatically with the proposed algorithm. In the figure, four consecutive images have been considered where an artery with a small diameter gradually merges with a larger-diameter artery. The sudden change in vessel boundary in the fourth image has been correctly outlined with the proposed algorithm. The algorithm has also been tested with stent-implanted IVUS dataset. As the vessel boundary looks very irregular with the stent-induced artifacts, the extraction of vessel boundary seems more challenging. The proposed algorithm is found to be able to correctly extract the vessel boundary from stent-implanted IVUS images (Figure 28). 5. CONCLUSIONS AND FUTURE WORK In this paper, the technique of an automatic 3D visualization of coronary artery model based on IVUS images has been proposed. The processing time is acceptable as the algorithm takes only a few iterations. As the plaque deposit increases over time, artery loses its elasticity and plaque rupture may take place. Different stages of plaque formation models can be created and examined in a 3D haptically immersive environment based on corresponding IVUS datasets, that not only makes the diagnosis faster but also the treatment plan easier. Figure 25. A haptic catheter is probing the coronary artery; (a) normal view and (b) tunnel view. Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 267 S. Yasmin et al. A haptic-enabled novel approach to cardiovascular Figure 27. Branching results in change in vessel diameter as well as the extracted vessel boundary. Figure 28. Extraction of vessel boundaries from stent-implanted intravascular ultrasound images. Further work on testing different IVUS datasets with the proposed algorithm will be carried out. The proposed 3D software utility can be used by cardiologists for examining coronary IVUS images in a user-friendly 3D environment and be employed as a training tool for medical students/practitioners as well. 5. ACKNOWLEDGEMENT 6. This work is supported by the National Science Foundation grant NSF/HRD-CREST#0932339. 7. REFERENCES 1. Moraes MC, Furuie SS. An automatic mediaadventitia border segmentation for IVUS images. Computing in Cardiology 2010; 37: 389–392. 2. Mendizabal-Ruiz EG, Rivera M, Kakadiaris IA. Robust segmentation of the luminal border in intravascular ultrasound using a probabilistic approach, Technical Report Number UH-CS-11-02, 2011 3. Unal G, Bucher S, Carlier S, Slabaugh G, Fang T, Tanaka K. Shape-driven segmentation of intravascular ultrasound images, Proc. of First International Workshop on Computer Vision for Intravascular and Intracardiac Imaging (CVII 2006) , pages 50–57, 2006 4. Taki A, Najafi Z, Roodaki A, et al. Automatic segmentation of calcified plaques and vessel borders in IVUS 268 8. 9. 10. images, International Journal of Computer Assisted Radiology and Surgery 2008; 3: 347–354. Koga T, Ichiyama S, Uchino E, Suetake N, Hiro T, Matsuzaki M. Fully automatic boundary extraction of coronary plaque in IVUS image by anisotropic diffusion and T-S type fuzzy inference. Soft computing in Industrial Applications, AISC 2010; 75: 139–147. Papadogiorgaki M, Mezaris V, Chatzizisis YS, Giannoglou GD, Kompatsiaris I. Image analysis techniques for automated IVUS contour detection. Ultrasound in medicine and biology journal 2008: 34(9): 1482–1498. Brusseau E, de Korte CL, Mastik F, Schaar J, van der Steen AFW. Fully automatic luminal contour segmentation in intracoronary ultrasound imaging- a statistical approach, IEEE Transactions on Medical Imaging 2004; 23(5): 554–566. Cardinal MR, Meunier J, Soulez G, Maurice RL, Therasse E, Cloutier G. Intravascular ultrasound image segmentation: a three-dimensional fastmarching method based on gray level distributions. IEEE Transactions on Medical Imaging 2006; 25(5): 590–601. Wennogle M, Hoff W. Three dimensional segmentation of intravascular ultrasound data, ICIAR 2009. LNCS 2009; 5627: 772–781. Ridler TW, Calvard S. Picture thresholding using an iterative selection method, IEEE Trans. System, Man and Cybernetics, SMC-8, 630–632,1978 Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav S. Yasmin et al. A haptic-enabled novel approach to cardiovascular AUTHOR’S BIOGRAPHIES Shamima Yasmin received her bachelor’s degree in Civil Engineering (Structural) from Bangladesh University of Engineering and Technology. Later, she received her Master’s in Computer Science from the University of New South Wales, Australia, in 2004. She obtained her PhD degree in Computer Graphics from Universiti Sains Malaysia in 2010. Her research interests include 3D modeling, visualization, virtual reality, and parallel computing. Currently, she is working as a postdoc researcher at Arizona State University. Nan Du received her bachelor’s degree in Information Engineering from Beijing University of Post and Telecommunications in 2008. She is currently a doctoral student at Digital Image Processing Laboratory at the Department of Electrical and Computer Engineering (ECE), University of Texas at San Antonio (UTSA). She is also a research assistant at the Department of ECE, UTSA. She is the author of two papers and her main interests are in image processing and medical imaging. S. James Chen received his BS degree from the Department of Computer Science and Information Engineering from the National Chiao-Tung University, Taiwan, in 1982 and his MS and PhD degrees from the Department of Computer Science from Northwestern University, Evanston, Illinois, in 1986 and 1991, respectively. In 1996, Chen joined the Division of Cardiology, Department of Medicine, University of Colorado Denver (Health Sciences Center), as an Assistant Professor. He is now an Associate Professor of Medicine and directing the 3-D Coronary Imaging Laboratory. He has collaborated with Dr John D. Carroll for more than 18 years to develop various computer-assisted utilities to facilitate diagnostic and therapeutic procedures in the cardiac catheterization laboratories. Such a research project has received the ‘Melvin Judkins Young Clinical Investigators International Award’ in Cardiovascular Radiology in the Conference of American Heart Association 1997. Currently, several important research projects are under development including (a) on-line 4D imaging to assist cardiac intervention for structural heart diseases, (b) multi-modality fusion (computed tomography, X-ray, and ultrasound) and advanced quantitative estimates, and (c) kinetic and deformation analyses on vascular structures and implantation devices. Yusheng Feng is a Professor of Mechanical and Biomedical Engineering at the University of Texas (UT) San Antonio and the Director and coFounder of NSF-Sponsored Center for Simulation Visualization and Real-Time Prediction. His research areas are in computational bioengineering, mathematical modeling, computer simulation, and hapticsenabled visualization for surgical simulation and training. Feng received his PhD in computational mechanics from the UT at Austin. He has two Master’s degrees in mechanical engineering and applied mathematics. Comp. Anim. Virtual Worlds 2014; 25:255–269 © 2014 John Wiley & Sons, Ltd. DOI: 10.1002/cav 269