New Approaches to Computer-based Interventional Neuroradiology Training

Transcription

1 New Approaches to Computer-based Interventional Neuroradiology Training Xunlei WU, Vincent PEGORARO, Vincent LUBOZ, Paul F. NEUMANN, Ryan BARDSLEY, Steven DAWSON, Stephane COTIN The Simulation Group, CIMIT, MGH, Harvard Medical School Abstract. For over 20 years, interventional methods have substantially improved the outcomes of patients with cardiovascular disease. However, these procedures require an intricate combination of visual and tactile feedback and extensive training periods. In this paper, a prototype of endovascular therapy training system is presented. A set of core simulation components applicable to most vascular procedures has been designed and integrated into a real-time high-fidelity interventional neuroradiology training system for the prompt treatment of ischemic stroke. We believe it will improve the quality of training and the speed of learning without putting patients at risk. 1 Introduction 1.1 Background Stroke is the third leading cause of death in the US. Each year, more than 750, 000 strokes result in over 150, 000 deaths [1]. The main therapy for ischemic stroke is catheterization, allowing direct lytic therapy to dissolve the clot and restore the flow. Following femoral arterial puncture, a guidewire-catheter combination is advanced under fluoroscopic guidance through the iliac arteries, into the aortic arch, and inside the common carotid artery. This allows entry into the internal carotid artery and the cerebral circulations in the brain. This guidance is provided by intravascular angiogram, obtained during contrast agent (CA) propagation, which defines the abnormal areas, guides the instrument movement, and verifies the treatment. Because the treatment is delivered only under image-based guidance, the dedicated skill of instrument navigation and the thorough understanding of vascular anatomy are critical to avoid irreversible complications. Unfortunately, the best training environments have been actual patients who have had a stroke. Additionally, the shortage of trained interventional neuroradiologists means that many ischemic stroke patients do not have access to care. In 2004, a decision by the FDA regarding appropriate levels of training for physicians who perform high-risk procedures in the cerebral circulation mandated that physicians train to proficiency before treating humans. For the first time, a major part of this mandated training includes simulation. Future trends in procedural education are therefore likely to increase the role of high-fidelity simulation for medical training. 1.2 Previous Work A few VR systems focusing on interventional radiology (IR) have been developed or commercialized [10, 7, 9, 3]. The core architecture has remained the same. In this paper, we propose

2 2 Xunlei Wu et al. Figure 1: Left: Segmented vasculature. Right: Geometric skeleton generated from the vascular isosurface. new approaches for rendering, physics-based modeling, and anatomical representations that will lead to a real-time high fidelity simulation system for more effective IR skill training. 2 Methodology 2.1 Semi-automatic Segmentation At first we need to segment the cerebrovascular system from a CTA scan. The segmentation task is particularly challenging due to the small vessel diameter and the close proximity of vessels to the skull. We used a combination of anisotropic filtering and morphological operators, e.g. dilation and erosion provided in Amilab 1 to eliminate bony structures, skin, and sinuses. The result of the segmentation on a CTA dataset of mm 3 can be seen on the left side of Figure 1. A few discontinuities in the network due to artifacts will be resolved to generate a complete 3D vascular model. After obtaining the isosurface of the vascular network, the medial axis is computed. Geometric skeletons are defined as the locus of the centers of maximal spheres within a shape, and provide a mean of describing the topology of the model. This information is required by the one-dimensional flow model in Section 2.4 and contrast agent propagation algorithm in Section 2.5. We also used Amilab to generate the centerlines and, with the help of a physician, to label each vessel as shown in the right half of Figure Incremental Finite Element Model for Catheter/Guidewire In this simulator, we have developed physics-based flexible instrument models representing a variety of intravascular devices, suitable for vessels from 3 to 20mm diameter. Current methods represent flexible instruments as a set of connected rigid elements [5], or by linear elastic FEM [9]. A common drawback of current methods is the inability to capture the essential characteristics of wire-like structures, including high tensile strength and low resistance to bending. To improve the accuracy of previously proposed models, we have developed new mathematical representations based on three-dimensional beam theory, where the large number of parameters for each element will allow the representation of a library 1

3 New Approaches to Computer-based Interventional Neuroradiology Training 3 of commonly employed catheters or guidewires. Key characteristics to model mechanical properties include: cross-sectional area, effective shear area, Young s modulus, Poisson s ratio, and cross-section/polar moment of inertia. By using an incremental approach to update the stiffness matrix at every time step, highly non-linear behavior of catheter/guidewire can be realisticly simulated. Furthermore by controlling the number of elements, we are able to ensure real-time deformation while maintaining reasonable accuracy. 2.3 Sequential Quadratic Programming for Collision Response Navigating through the vasculature of this system requires an instrument tracking device and a collision response (CR) method between the vasculature and instrument models. In our previous work [3],as well as that of others [8, 4], once contact has been determined, forces or boundary conditions are applied to the deformable catheter/guidewire model. Since the catheter has many contacts against the vessels, an Octree based collision detection (CD) algorithm has been implemented. Because sliding occurs at the point of contact, Lagrange multiplier techniques or penalty forces will not constrain the flexible body correctly and may induce oscillations. Our approach relies on sequential quadratic programming (SQP), combining FEM and CR together. A set of inequality constraints, based on the distance between catheter nodes and neighboring vessel triangles, are added to the set of FEM equations. Our catheter model and CR are shown in the left half of Figure 2. Figure 2: Left: Model the catheter/guidewire deformation by 3D beam elements with collision response. Right: The optical tracking device for catheter/guidewire can be concealed inside a patient mannequin. 2.4 One-dimensional Vascular Flow Computation Clinically turbulent flow is a diagnostic aid for IR procedures, but rarely observed under stroke therapy where the small diameter of vessels combined with limited flow rates reduces the occurrence of turbulence. In addition, the average resolution of a fluoroscopic image is on the order of 0.3mm, making it difficult to observe turbulence or transaxial flow pattern. Hence, one-dimensional fluid flow model along the vessel axes should provide enough information for a training system. To compute such vascular flow, we have implemented a simplified one-dimensional FEM representation proposed in [3].The blood flow in each element is modeled as an incompressible viscous fluid flowing through a cylindrical pipe, and can be calculated from the Naiver-Stokes equation. The resulting equation, called Poiseuille

4 4 Xunlei Wu et al. Law [6], is solved by linear FEM. This computation runs very efficiently and provides realtime simulation of vascular flow. The left half of Figure 3 shows the flow computation results with vessel median axes with radii ranging from 2 to 6mm. Figure 3: Left: Real-time 1D vascular flow simulation using vascular graph. Right: Contrast agent propagation computed in a simplified bifurcation according to an advection diffusion model D Particle-based Real-time Angiography Simulation Contrast agent (CA) is often used during medical imaging to highlight specific parts of the body under X-ray, CT, and MRI. Upon injection, CA is carried by blood cells and circulates through the vasculature until it is eliminated in the kidneys and liver. We computed the transportation of CA by an advection equation in terms of the CA concentration C(x, t) distribution parameterized by the curvilinear coordinate x and time t, C(x, t) t C(x, t) + u(x, t) x = r(t) (1) where r(t) is the injection rate of contrast agent and u(x, t) is the averaged laminar flow velocity along the axial direction of each vessel. Mixture transition of multiple fluids and the transversal velocity profile are not considered. We numerically solved (1) by forward-in-time and center-in-space finite difference schemes whose accuracy is linear in temporal domain and quadratic in space. To visualize an angiogram, we used a set of equally spaced 3D particles to represent the vessel s internal volume as in Section 2.1. The intensity values of particles are then mapped to C(x, t) at each sampling point along the vascular graph. The final rendering stage combines one updated particle system for CA propagation and one volumetric texture map for the surrounding anatomy as described in Section 2.6. This approach uses mainly GPU power and runs at an interactive frame rate. 2.6 Real-time Simulation of Fluoroscopic Rendering We have developed a new volume rendering approach for the simulation of fluoroscopic images directly using CT. With OpenGL as a rendering library, a ray casting method using specific blending is implemented to approximate the X-ray attenuation process. The volume rendering algorithm creates a set of parallel slices onto which the corresponding part of the

5 New Approaches to Computer-based Interventional Neuroradiology Training 5 volume intersects. The slices are rendered as a 2D texture-mapped polygon by approximating the beam attenuation described by the discretized Beer s Law, I = I 0 e P j µ jd j (2) with I 0 the input intensity, µ j the linear attenuation coefficient sampled at slice j, and d j the slice thickness along the ray. This method greatly reduces computation times by using OpenGL and specific texture map operations. 2.7 Catheter/Guidewire Motion Tracking Device We have developed an optic catheter/guidewire tracking device that passively recreates haptic sensation, since most feedback in IR is visual. The tracking device is embedded inside a full-size patient mannequin, therefore naturally recreates the OR environment as shown in the right of Figure 2. This improves the level of realism during training. The virtual vasculature is accessed through a standard sheath on the left or right femoral artery. Once a real catheter is inserted into the sheath, the system starts tracking the instrument s 2 DOF s motion. Passive haptic feedback friction is provided by a set of anatomically correct Teflon tubing phantoms. Currently, only one instrument - catheter or guidewire - can be tracked at a time. A major difficulty, when tracking two or more nested devices, is that access to the inner device is mechanically impossible without modifying the instrument [7]. This challenge will be thoroughly investigated in the future as well as the validation of tracking accuracy. 3 Results We have developed new approaches for fluoroscopic rendering, physics-based instrument modeling, and high resultion anatomical representations and integrated them into a real-time interventional neuroradiology simulator for skill training. Our simulator is implemented on a P4 3.0GHz PC equipped with 1GB main memory and NVIDIA GeForce FX5900 Ultra graphic card. It is capable of rendering 256 slices of a volume with more than one million 3D particles at 29F P S depending on user interaction. This frame rate includes fluoroscopic rendering, collision detection and response, catheter/guidewire deformation, and CA advection. The left half of Figure 4 shows the full simulation system and the right half Figure 4: Left: The high-fidelity interventional neuroradiology skill training simulator. Right: 3D fluoroscopic rendering of CA propagating through a cerebrovascular model within a CT head model.

6 6 Xunlei Wu et al. illustrates the simulated contrast agent propagation through a volumetric cerebro-vasculature model. 4 Conclusions and Future Work In summary, a set of endovascular simulation components have been developed and integrated into a training system for the treatment of stroke. This system allows multiple levels of skill acquisition using interventional instruments. This prototype emphasizes high fidelity visual feedback and physically accurate CR and CA propagation. Due to its cost-effective design and system compactness, such a system would lead to cross-specialty interventional training. Also, this platform increases accessibility to medical institutions and hospitals. In the future, we will extend the catheter model to interventional balloon and stent by mass-spring network or as a set of radial elements [2]. Upon completion, our prototype will undergo face validity from our hospital collaborators and then comprehensive validation study. We will also incorporate an educational curriculum providing various scenarios. These improvements should reveal full potentials of this simulator including procedural planning. 5 Acknowledgments This work was supported by Telemedicine and Advanced Technology Research Center (TATRC) through grant number DAMD We thank Dr. Karl Krissian from SPL and Dr. James Rabinov in MGH for providing us valuable feedback on anatomical segmentation. References [1] American Heart Association. Heart and stroke facts statistics: Statistical supplement. American Heart Association, Dallas, Texas, [2] R. Balaniuk and K. Salisbury. Soft-tissue simulation using the radial elements methods. In Proc. International Symposium on Surgery Simulation and Soft Tissue Modeling, pages 48 58, [3] S. Dawson, S. Cotin, D. Meglan, D.W. Shaffer, and M.A. Ferrell. Designing a computer-based simulator for interventional cardiology training. Catheterization and Cardio. Intervention, 51(4): , [4] A. Deguet, A. Joukhadar, and C. Laugier. A collision model for deformable bodies. In International Conference on Itelligent Robots and Systems. IEEE, [5] R. Featherstone. The calculation of robot dynamics using articulated-body inertias. International Journal of Robotics Research, 2(1):13 30, [6] A. Guyton and J. Hall. Textbook of Medical Physiology. Saunders Elsevier, 10th edition, [7] U. Hoefer, T. Langen, J. Nziki, F. Zeitler, J. Hesser, U. Mueller, W. Voelker, and R. Maenner. Cathi - catheter instruction system. In Computer Assisted Radiology and Surgery (CARS), 16th International Congress and Exhibition, pages , Paris, France, [8] P. Meseure, J. Davanne, L. Hilde, L. France, F. Triquet, and C. Chaillou. A physically-based virtual environment decidated to surgical simulation. In Surgery Simulation and Soft Tissue Modeling (IS4TM), pages 38 47, June Juan-les-pins, France. [9] W.L. Nowinski and C.K. Chui. Simulation of interventional neuroradiology procedures. In Int. Workshop on Medical Imaging and Augmented Reality (MIAR), pages IEEE Computer Society, [10] N. Subramanian and T. Kesavadas. A prototype virtual reality system for preoperative planning of neuroendovascular interventions. In Proceedings of Medicine Meets Virtual Reality 12th, pages , 2004.