Int J CARS (2008) 3 (Suppl 1):S86–S93 DOI 10.1007/s11548-008-0178-9

CARDIOVASCULAR SURGERY

Towards navigation on the heart surface during coronary arterty bypass grafting Gnahm C.1, Hartung C.1, Friedl R.2, Hoffmann M.3, Dietmayer K.1 1 Ulm University, Measurement, Control and Microtechnology, Ulm, GERMANY 2 University Hospital of Ulm, Heart Surgery, Ulm, GERMANY 3 University Hospital of Ulm, Diagnostic Radiology, Ulm, GERMANY Keywords CABG, registration, optical tracking system Purpose Coronary artery bypass grafting (CABG) is the standard treatment for advanced coronary artery diseases and is widely performed on patients with significant stenoses of the coronary arteries. In this surgical procedure, the blood flow to the affected ischaemic region is improved by a vessel graft, rerouting the blood flow around the stenosis. For heart surgeons, localising the diseased vessel, possibly covered by epicardial fat, and finding an appropriate anastomosis site for the bypass graft is of particular importance. Moreover, it is desirable to accelerate the localisation of the anastomosis site and thus reduce the ischaemic time of the heart and the duration of extracorporeal circulation. To address these problems, the Cardio-Pointer project aims at developing a surgical navigation system to assist the surgeon in the optimal placement of the bypass graft. The proposed system is based on the fusion of preoperative MSCT (multi slice computer tomography) data with intraoperative data of an optical tracking system. During surgery, its purpose is to provide a patient-specific map of the coronaries in which the current position of a surgical pointing device (Cardio-Pointer) is visualised. To enable navigation, pre- and intraoperative modalities have to be registered. Although CABG on the beating heart is increasingly performed, most procedures are performed during ischaemia on a non-beating heart connected to the heart lung machine. This work is therefore focused on an appropriate registration method for preoperative MSCT data and optical tracking data recorded intraoperatively during ischaemia. The registration algorithm is based on mutually shared anatomical point landmarks and vessel paths on the heart surface. It has been validated retrospectively on real patient data sets recorded during CABG surgery. Methods The registration approach is based on mutually shared anatomical point landmarks on the heart surface and vessel paths visible during surgery. As parts of the heart surface and therewith the coronary arteries may be covered with epicardial fat, the number of actually visible anatomical landmarks is limited and different in each patient. Prior to registration, pre- and intraoperative data have to be recorded. Preoperatively, an individual map of the patient’s coronary arteries is generated. For this purpose, the coronary artery tree as well as plaque positions are segmented from mid-diastolic (75% RR) MSCT data. Additionally, prominent features which are likely to be observable during surgery are marked. Such potential point landmarks can be vessel bifurcations or crossing points of different vessels. Intraoperatively, data for the registration process are recorded by means of an optical tracking system. For this purpose, the surgeon is equipped with a trackable pointing device (Cardio-Pointer). While pointing at relevant landmarks, the position of the Cardio-Pointer tip

123

is recorded. Additionally, the course of registration-relevant vessels is retraced with the Cardio-Pointer. The registration algorithm consists of three steps: First, the data sets are coarsely matched using a rigid body transformation; Second, the matching is refined by applying an enhanced weighted ICP algorithm which corrects elongations and shortenings in the optical tracking data due to the deformation of the heart; Third, a torsion correction is performed. In the first step, mutually shared anatomical point landmarks are utilised for the rigid body transformation. In the second step, the resulting transformation is employed to initialise the enhanced weighted iterative closest point (ICP) algorithm. This algorithm uses anatomical point landmarks as well as vessel paths to refine the matching. If suitable, the utilised data points can be weighted individually. The non-beating heart is considerably deformed compared to the beating heart upon which the preoperative map of the coronaries is based. As soon as the non-beating heart is manually brought into an appropriate position for bypass grafting, major deformations occur. To some extent, they can be corrected with non-isotropic scaling of the optical tracking data. Therefore, the ICP algorithm was enhanced to consider a different scaling factor for each spatial direction (axial, sagittal, and coronal). In the third step, the torsion of the non-beating heart along the course of the relevant vessel is corrected. The torsion of the transformed optical tracking data resulting from step two is calculated along the vessel course with respect to a heuristically estimated torsion axis of the heart. The variation of the torsion angle along the vessel is piecewise linearly approximated. This torsion angle approximation and an accordingly rescaled vessel length allow a further correction of the vessel tracking data. Results The described registration process was tested on real patient data recorded on ischaemic hearts during different phases of the bypass grafting procedure. For the initialisation, different numbers and types of landmarks were used. After completion of the three registration steps, the intraoperatively recorded vessel paths showed good accordance with the preoperative map of the coronaries. To evaluate the performance of the registration algorithm, two values were considered: First, the RMS (root mean square) fiducial registration error of all mutually shared data points not involved in the registration process was calculated. Second, the average distance of corresponding vessel paths taken from the preoperative MSCT data and the transformed intraoperative tracking data was calculated. A RMS error of 4–6 mm was obtained for the mutually shared data points. The average distance of the vessel paths after complete registration was 2– 5 mm. Conclusion The registration process presented in this work is capable of matching the relevant parts of a map of the coronaries which is extracted from a preoperative CT scan with intraoperatively recorded optical tracking data. This allows the visualisation of the optical tracking data in a patient-specific coronary artery map. The registration process can be utilised as a basis for a surgical navigation system which is intended to assist the CABG surgeon in the localisation of the optimal bypass

Int J CARS (2008) 3 (Suppl 1):S86–S93 graft anastomosis site. The surgeon will point at potential distal anastomosis sites utilising the Cardio-Pointer. On the patient-specific map of the coronaries, the current pointer position is shown. Furthermore, it displays luminal narrowing and wall plaque formations at this potential site and decision making can be guided accordingly. Lumen and plaque information would subsequently become available prior to dissection and opening of the vessel of interest. This may augment both quality and speed of CABG surgery. Acknowledgments This work is supported by the Federal Ministry of Education and Research (BMBF), project 01EZ0614.

Inside the Beating Heart: An in vivo Feasibility Study on Fusing Pre- and Intra-operative Imaging for Surgical Guidance Cristian A. Linte*1,2, John Moore1, Chris Wedlake1, Daniel Bainbridge2,3, Ge´rard M. Guiraudon1,2,3, Douglas L. Jones2,3 Terry M. Peters1,2,3 1 Imaging Research Laboratories, Robarts Research Institute, London, Ontario, CANADA 2 Schulich School of Medicine and Dentistry, University of Western Ontario, Calgary, Alberta, CANADA 3 Canadian Surgical Technologies and Advanced Robotics, London, Ontario, CANADA Abstract In response to the global ongoing theme on reducing morbidity during minimally invasive cardiac surgery, we developed an interventional system to assist surgeons with therapy delivery inside the beating heart, in absence of direct vision. Our system features a virtual reality environment that integrates pre-operative anatomical information, real-time intra-operative imaging via 2D trans-esophageal ultrasound, and models of the surgical tools tracked using a magnetic tracking system. To further enhance visualization, we supply the surgeons with detailed 3D dynamic cardiac models constructed from high-resolution pre-operative MR data and registered within the intra-operative imaging environment. Here we report our experience on employing a feature-based registration technique to fuse the pre- and intra-operative data during an in vivo intracardiac procedure on a porcine subject. This method is suitable for in vivo applications as it relies on easily identifiable landmarks, and also ensures a good alignment of the pre- and intra-operative anatomy within the virtual reality environment. Given its extensive capabilities in providing surgical guidance in the absence of direct vision and with no exposure to ionizing radiation, we believe that our virtual environment constitutes an ideal candidate for performing off-pump intracardiac surgery. Keywords minimally invasive cardiac interventions, intra-operative imaging, pre-operative anatomical modeling, virtual augmented reality Introduction In the context of cardiac interventions, minimizing invasiveness has inevitably led to more limited access to the surgical targets. In spite of the pervasive use of medical imaging and both robotic and laparoscopic technologies, the transition to off-pump intracardiac interventions has been hampered by the lack of adequate visualization and guidance inside the beating heart. In an effort to reduce morbidity in cardiac interventions through the use of less invasive techniques, we developed a novel surgical guidance system designed to facilitate the delivery of intracardiac therapy on the beating heart [1]. Our system relies on a virtual reality (VR) environment that augments intra-operative, 2D trans-esophageal echocardiography (TEE) images with pre-operative anatomical models [2], and virtual representations of the surgical instruments tracked in real time using a magnetic tracking system [1]. As a result, the intra-procedure TEE information can be interpreted within its 3D anatomical context to enhance procedure planning and navigation. This current work complements our previous endeavours in the context of an in vivo porcine study. It

S87

Fig. 1 a) Cardiac image at mid-diastole (MD); b) Static pre-operative cardiac model showing the left ventricle (LV), left atrium (LA) and right atrium and ventricle (RA/RV); c) Intra-operative 2D US image at MD also showing the LV, LA and the mitral valve annulus (MVA) focuses on generating dynamic, subject-specific anatomical models of the heart from pre-operatively acquired magnetic resonance (MR) images, and fusing them with the intra-operative 2D trans-esophageal ultrasound (US) images to generate a robust virtual display of the surgical field that compensates for the lack of direct intra-procedure visualization. Methodology Intra-operative imaging We employed 2D trans-esophageal echocardiography (TEE) for real-time, intra-operative image guidance. The acquisition was gated to the animal’s ECG, and moreover, and the TEE transducer was tracked in real-time using the NDI Aurora magnetic tracking system, using a 6 degree-of freedom (DOF) magnetic sensor coil attached to the US probe and calibrated using a Zstring approach [3]. Thus, the 2D US images were encoded both spatially and temporally (Fig. 1). However, although 2D TEE offers the flexibility of acquiring good-quality images while eliminating the interference between the probe manipulation and the surgical workflow, the main disadvantage of the 2D US images is their inadequate representation of the anatomical targets and surgical instruments required for procedure guidance. To address these limitations, we augment the 2D US images with anatomical context provided via preoperative cardiac models. Pre-operative modeling A set of 20 high resolution (1.09 x 1.09 x 2.0 mm3) 3D MR images of a pig’s heart were acquired throughout the cardiac cycle using prospective ECG gating. A static anatomical model of the pig’s heart (Fig. 1b) was obtained by manually segmenting the left ventricular myocardium (LV), left atrium (LA) and the right atrium and ventricle (RA/RV) from the mid-diastole (MD) cardiac MR image (Fig. 1a). A 3D free-from deformation field that describes the trajectories of all points in the surface model was extracted via a non-rigid image registration technique [4]. Using the mid-diastole heart image as reference, the frame to frame motion vectors (T0-k, where k = 1 to 19) were computed by nonrigidly registering the MD image (corresponding to k = 0) to the remaining images in the 4D dataset. A dynamic model was obtained by sequentially propagating the static model throughout the cardiac cycle using the motion vectors previously estimated (Fig. 2). Model-to-subject registration We employed a feature-based registration algorithm to augment the intra-operative US images with the pre-operative cardiac models, based on easily identifiable features – the mitral (MVA) and aortic (AVA) valve annuli extracted from both the pre- and intra-operative datasets [5]. The pre-operative annuli were segmented manually under the guidance of an experienced cardiologist. Similarly, the intra-operative annuli were also segmented manually by an echocardiographer, by sweeping the image fan of the magnetically tracked 2D TEE probe across each annulus. Ultimately, the model-to-subject registration was achieved as a result of the alignment of the pre-operative AVA and MVA defined in the model, with those identified intra-operatively [5]. Results This pre- to intra-operative registration technique proved suitable for cardiac interventions, as the selected valvular structures were not only

123

S88

Fig. 2 Dynamic cardiac model depicting the heart at different cardiac phases and obtained by propagating the static model throughout the cardiac cycle using the motion vectors extracted using non-rigid image registration.

Fig. 3 a) Intra-operative 2D US image augmented by pre-operative cardiac model; b) Complete augmented reality environment including pre-operative anatomical model, intra-operative US imaging, and virtual models of the surgical instruments, US transducer, and US image fan

easily identifiable in both the pre-operative and intra-operative datasets, but also ensured a good alignment of the pre- and intra-operative surgical targets, as well as their surrounding regions. On average, we achieved a * 4.8 mm accuracy in aligning the pre-operative model with the in vivo intra-operative US images for regions located in the vicinity the mitral and aortic valvular region (basal region of LV, LA and RA/RV). This adequate anatomical alignment enables us to employ these techniques in a variety of image-guided interventions inside the heart, including mitral or aortic valve implantation, atrial septal defect closure or ablation therapy for atrial fibrillation. As a result, during intra-procedure guidance, the surgeon has access to the complete augmented reality environment as opposed to just the 2D intra-operative US images (Fig. 3). Conclusions This initial work has demonstrated the tremendous potential of multimodality imaging, combined with tracking of tools and real-time US for providing the capability to both visualize and assess the surgical intervention in a manner that will ultimately be superior to direct vision, within its inherent limitations. References 1. Linte CA, Moore J, Wiles AD, Wedlake C and Peters TM. Virtual reality-enhanced ultrasound guidance: A novel technique for intracardiac interventions. Computer Aided Surgery. 13(2): 82–94. 2008. 2. Linte CA, Wierzbicki M, Moore J, Guiraudon GM, Little SH and Peters TM. Towards subject-specific models of the dynamic heart for image-guided mitral valve surgery. Proc Med Image Comput Comput-Assist Interv. (MICCAI). Lect Notes Comput Sci. (LNCS). 4792: 94–101. 2007. 3. Gobbi DG, Comeau RM and Peters TM. Ultrasound probe tracking for real-time ultrasound/MRI overlay and visualization

123

Int J CARS (2008) 3 (Suppl 1):S86–S93 of brain shift. Proc Med Image Comput Comput-Assist Interv. (MICCAI). Lect Notes Comput Sci. (LNCS). 1679: 920–27. 1999. 4. Wierzbicki M, Drangova M, Guiraudon GM and Peters TM. Validation of dynamic heart models obtained using non-linear registration for virtual reality training, planning and guidance of minimally invasive cardiac surgeries. Med Image Anal. 8: 387– 401. 2004. 5. Linte CA, Wierzbicki M, Moore J, Guiraudon GM, Jones DL and Peters TM. On enhancing planning and navigation of beatingheart mitral valve surgery using pre-operative cardiac models. Proc IEEE Eng Med Biol Soc.: 475–8. 2007.

Value of augmented reality enhanced transesophageal echocardiography (TEE) for determining optimal annuloplasty ring size during mitral valve repair Jacobs S.1, Holzhey D.1, Falk V.1 1 Heartcenter Leipzig, Heartsurgery, Leipzig, GERMANY Keywords Model guided Therapy, 3D TEE, Minimally Invasive Surgery Purpose Aim of the study was to investigate the potential value of augmented reality enhanced transesophageal echocardiography (TEE) for determining optimal annuloplasty size. Intraoperative transesophageal echocardiography (TEE) for mitral valve repair is a class I indication to detect the underlying pathology and to grade the severity of mitral regurgitation. Implantation of an annuloplasty ring is mostly performed independently from the type of the reconstruction technique (i.e.resection or implantation of neo-chords. Usually the sizing of the ring is performed using commercial sizer models during cardioplegic cardiac arrest comparing the distance between both commissures and the height of the anterior mitral leaflet.. In most patients this technique functions quite well, but there are some clinical conditions, where this technique may be incorrect (i.e. impaired exposure of the native valve with the possibility of annuloplasty mismatch). Therefore in this present study we compared a TEE- guided sizing of the annuloplasty ring with the routine method of direct sizing. Methods In patients undergoing elective mitral valve repair, a 3 dimensional reconstruction of the mitral valve was performed using TEE. A modified 4D valve assessment software was used to create 3D CAD-models of standard annuloplasty-rings (28 to 36 mm) which were stored in a digital database. These virtual 3D annuloplasty ring templates were overimposed on the preoperative 3D-TEE reconstructions of the mitral valve. Post-hoc comparison to conventional sizing under direct vision was performed. The investigator, who determined the potential size of the annuloplasty ring was blinded for the implanted ring size. Postoperatively, the size of the virtual ring templates and the implanted annuloplasty prostheses were compared to validate the accuracy of the virtual ring templates. Results 50 patients were included in the study. The correlation between the implanted annuloplasty ring size and the size, evaluated preoperatively with the virtual 3D annuloplasty ring templates was 0.83. 30 virtual ring templates showed no differences and 20 rings showed ± 2 mm differences in size. Postoperatively determined correlation of 0.94 between the size of the virtual ring templates and the implanted annuloplasty prostheses demonstrates, that this CAD model reflects real dimensions. Discussion The low perioperative risk for mitral valve repair and the recently published results of conservative therapy lowers the threshold for surgical intervention in patients with low grade mitral insufficiency.

Int J CARS (2008) 3 (Suppl 1):S86–S93 Therefore it is important to ensure a high success rate of mitral valve repair. The adequate size of the implanted annuloplasty ring is essential for the postoperative result. A mismatch in sizing may lead to either a residual regurgitation if the annuloplasty ring is too large, or in the case of a too small ring size in huge tension to the sutures that fix the annuloplasty ring with the risk of late mitral repair failure. Another consequence of a too small ring size can be the impaired flow through the mitral valve creating a mitral stenosis per se or producing a Systolic Anterior Motion of the anterior leaflet of the mitral valve with subsequent left ventricular outflow tract stenosis. This occurs especially in patients with a reduced C-Sept distance and a relatively large posterior leaflet. The routine method of sizing is the use of a tester in the cardioplegic arrested heart, which may not reflect the beating heart conditions during the TEE-examination. In our study we investigated the feasibility of TEE-guided sizing as compared to the surgical sizing. The excellent postoperatively determined correlation of 0.94 inbetween the size of the ring models and the implanted annuloplasty prostheses demonstrates, that this computer model reflects real dimensions. The total agreement of the sizing in both methods on 30 of the 50 patients can be explained by the experience of the surgeons in a high volume center, where more than 350 mitral repairs are performed per year. In one patient where the implanted ring size was 4 mm smaller than the TEE guided ring, a large resection of the PML and the AML was performed. The question weather the difference of 2 mm in ring size inbetween both methods is clinically important can not be answered by this retrospective study. The poor image quality in 13 patients may be one explanation.. Probably this problem can be solved in the future with the availability of real-time 3D transesophageal echocardiography. The surgical technique of mitral valve repair may not have a great influence regarding the agreement of both methods except for those patients with large resection of both leaflets. Although in our study population the majority of the patients was operated using the loop technique. Conclusions Augmented reality enhanced TEE for determining optimal annuloplasty ring size during mitral valve repair correlates well with the surgical sizing. A prospective randomized study is necessary to evaluate the clinical value of this new approach to determine the optimal annuloplasty-ring.

A MRI compatible robotic delivery module for transapical aortic valve replacement Li M.1, Mazilu D.1, Horvath K.1 1 National Institutes of Health, National Heart, Lung and Blood Institute, Bethesda, UNITED STATES Keywords MRI compatible, robot-assisted surgery, aortic valve replacement Purpose Transapical aortic valve replacement is a new minimally invasive approach that allows the placement of a bioprosthetic valve via a trocar that is inserted into the apex of the beating heart. Because the aortic valve lies in close proximity to both of the mitral valve and the coronary ostia, the position and orientation of the implanted valve is critical. Misalignment of the prosthesis could result in mitral valve damage or cardiac ischemia. Use of real-time MR imaging for guidance allows continuous evaluation of the delivery of the prosthesis, excellent views of valvular and annular anatomy, and can be performed on the beating heart without requiring ventricular unloading. Our group performed the first set of real-time MRI guided transapical aortic valve replacement in an animal study [1, 2]. There are several obstacles to the current MR guided procedure. Access to the operative field inside the magnet can be difficult for the surgeon. A surgical assistant is required to hold a long delivery device in an

S89 appropriate position and angle providing the surgeon a direct short trajectory toward the target area. In order to manipulate the delivery device properly, a coordinated effort between the surgeon and the assistant is critical. To better assist the performance of this minimally invasive beating heart procedure under real-time MRI guidance, we developed a fully MR compatible robotic system. Methods This work presents the development of a robotic module that provides precise and reproducible positioning of a bioprosthetic aortic valve into a beating heart. A newly developed 3-DOF valve delivery module can be attached to a MRI compatible robotic arm (Innomotion, German, [3, 4]) or to a passive arm. This compact robotic module is designed for manipulating and placing the valve into a beating heart inside the MRI scanner by remote control. The valve delivery module consists of two components; a sterile disposable valve delivery unit, and an active mechanism that provides the essential manipulation of the valve delivery device for a valve placement. All parts of the prototype delivery module are made from non-conductive plastic materials. Pneumatic actuators and optical sensors are used for operating and positioning each joint. A PIV (proportional position loop integral and proportional velocity) controller runs on a XMP-SynqNet-PCI board is used for servoing the valve delivery module movement. The function of the valve delivery module allows three movements: 1) translation stage: advance or retract the delivery device through the trocar, 2) rotation stage: rotate the delivery device around its axis to change the orientation of the valve relative to coronary ostia before it is deployed, and 3) insertion stage: advance the valve out of the delivery device to place and affix the valve. The operations of these stages are independent. The order of operation is interchangeable. Results The prototype of the valve delivery module is shown. The robotic module allows precise and reproducible positioning of a bioprosthetic valve in the correct location. The maximum measured force load of the translation stage, under 75 psi operation air pressure, was 34 N; and the maximum torque of the rotation stage was 0.4Nm. With operational velocity of 10 mm/s, the accuracy of translation stage was found to be 0.19 ± 0.14 mm. With operational velocity of 5 deg/s, the accuracy of the rotation stage was found to be 0.46 ± 0.27°0. With a load of 10 N (a reasonable force if we consider the friction and reaction force from real case), the module moved smoothly and stably and the accuracy was in the same range. Accuracy is not the critical requirement for this stage of the procedure, especially with the surgeon in the control loop. The MR compatibility of the robotic module was studied on a 1.5T Siemens Espree scanner. A steady-state free precession (SSFP) sequence was used with following scanning parameter: TR = 436.4 ms, TE = 1.67 ms, echo spacing = 3.2 ms, imaging flip angle = 45 deg, slice thickness = 4.5 mm, field of view (FOV) = 340x283 mm, and matrix = 192x129. This protocol was similar to the one we used in MR scanning for the cardiac intervention. The presence and motion of the robotic device inside the bore was found to have no noticeable disturbance in the image. Furthermore, the observed SNR loss was 6.1% to 6.5% for the valve delivery module placed in the bore and the valve delivery module in motion respectively. Conclusion We have developed and evaluated a MR compatible robotic valve delivery module for a surgeon to remotely control transapical aortic valve replacement. This valve delivery module along with a 5-DOF Innomotion positioning arm will provide a direct access to the aortic valve and allow the surgeon to remotely control bioprosthetic valve delivery with MRI guidance. Preliminary evaluation of the parameters of the prototype of the valve delivery module in ex-vivo experiment shows the robotic valve delivery module can provide sufficient

123

S90 capabilities to successfully assist the surgeon. Our immediate goal is to evaluate this system in a phantom study and in a large animal study. References

Int J CARS (2008) 3 (Suppl 1):S86–S93 left ventricular lead implantation in a clinical CRT case. 120 projection images were obtained over 4 seconds using a 110 degree rotation. The coronary venous tree was opacified in a retrograde fashion using manual iodinated contrast agent injection (Schering

1. McVeigh ER, Guttman MA, Lederman RJ, et al.: Real-time interactive MRI-guided cardiac surgery: Aortic valve replacement using a direct apical approach. Magnetic Resonance in Medicine, 2006, 56:958–964 2. Horvath KA, Guttman M, Li M, et al.: Beating heart aortic valve replacement using real-time MRI guidance. Innovations, 2007, 2:51–55 3. Hempel, E., H. Fischer, L. Gumb, et al.: An MRI-compatible surgical robot for precise radiological interventions. Computer Aided Surgery, 2003, 8(4):180–191 4. Melzer A., Guttmann B, Remmele TH, et al.: MRI and CT Compatible Robotic System for Percutaneous Image Guided Interventions: Principles and Evaluation, IIIE Engineering in Medicine and Biology, 2008 in press

Multimodality fusion of soft-tissue imaging volumes with vascular angiograms based on anatomical models for cardiac resynchronization therapies Manzke R.1, Tournoux F.2, Handschumacher M.D.2, Singh J.2, Chan R.1 1 Philips Research North America, Briarcliff Manor, UNITED STATES 2 Mass General Hospital, Boston, UNITED STATES Keywords electrophysiology, interventional guidance, multi-modality fusion Purpose In cardiac resynchronization therapies (CRT) procedures, a bi-ventricular pacemaker is implanted. The pacemaker paces the right atrium (lead screwed in right atrial tissue), the right ventricle (lead screwed into septal wall close to right ventricular apex) and the left ventricle (lead implanted in the coronary venous tree). Cardiac electrophysiology (EP) applications such as CRT require detailed knowledge of cardiac anatomy (vein tree, ventricle) and functional parameters (ventricular synchronicity). It is thought that with optimal placement of the left ventricular lead and optimal programming of the pacing sequence one can optimize cardiac output to mitigate heart failure symptoms. Functional and anatomical measurements from the heart are generated on separate imaging systems. X-ray is the mainstay imaging modality for interventional EP guidance whereas cardiac Ultrasound is the modality of choice for characterization of ventricular function. Fusion of complementary information from these modalities is a first step towards fully utilizing the available information during interventional guidance. Fusion of different modalities such as X-ray and Ultrasound data has been difficult due to lack of shared features (X-ray visualizes venous anatomy based on contrast agent delivery whereas ultrasound images myocardial characteristics without venous definition). In this paper, we present an image-fusion approach which employs a computer model of the coronary sinus anatomy to register X-ray and Ultrasound images. No external tracking devices are required, since this is an image-based approached. Methods 12 cardiac CT scans were evaluated to generate an average computer model of the coronary sinus anatomy including the proximal 3 cm relative to 4 fiducial points of the mitral valve annulus available in Ultrasound and CT (see Figs. 1 and 2). During the intervention, rotational X-ray vein imaging was performed using a Philips Allura X-Per FD10 flat detector system during

123

Fig. 1 Statistical model generation based on delineation of the proximal 3 cm of the coronary veinous centerline relative to 4 mitral valve fiducial points visible in cardiac CT and ultrasound. The 3D locations of 4 mitral valve fiducial points (purple markers in lower left plot) are determined from multiplanar reformatted slices of 12 cardiac CT volumes. The centerline location of the proximal 3 cm of the coronary veins is also defined (green markers) for each patient. These markers are all mapped into a common reference space and the mean position of the 3D coronary venous centerline is computed (red markers). The red centerline represents the inferred proximal vein centerline location relative to the mitral valve fiducials which are readily identifiable in 3DUS datasets

Fig. 2 The same mitral valve landmarks measured in cardiac CT volumes are easily identified in US volume data and are used to register the left ventricular shell from cardiac echo with the statistical model of the proximal coronary vein. We averaged the coronary vein measurements from the 12 patients to build the model shown. The vein model centerline is the mean 3D position over 12 patients whereas the model diameter represents one standard deviation of the centerline position at each segment location

Int J CARS (2008) 3 (Suppl 1):S86–S93 Ultravist, 20 ml) through the catheter at the time of the image acquisition. Subsequently, coronary vein models were generated using a 3D centerline approach. Immediately after the procedure 3D ultrasound images of the heart were taken (Philips iE33 cardiac 3D Ultrasound, X3-1 transducer) having the bi-ventricular pacemaker switched on and off. A 3D parametric mesh describing the ventricular shape and synchronicity was calculated using QLab (Philips Medical Systems, Andover, MA). The computer model of the average coronary sinus and mitral valve anatomy was registered to the patient data from 3D Ultrasound using Procrustes method and the 4 mitral valve points (Fig. 2). Subsequently, the 3D coronary vein, reconstructed from the X-ray data was registered with the average coronary sinus location of the computer model (Fig. 3). Hence, functioanatomical 3D Ultrasound data was successfully fused with anatomical X-ray data via the average computer model derived from CT (see Fig. 4). Results Functional meshes from 3D Ultrasound and X-ray reconstructions of the coronary veins of 10 cases have been fused using the computer model. The registration of the mitral annulus points of the model with 3D Ultrasound was achieved with an average error of less than 3 mm. The registration of the coronary venous tree with the average coronary sinus anatomy was inspected visually side-by-side with X-ray angiography as shown in Fig. 4.

Fig. 3 Registration of ultrasound and X-ray spaces based on spatial transformation of the proximal vein model in US space into the corresponding segment of the coronary vein present in X-ray space

Fig. 4 Final result showing rotational X-ray projection on the left and corresponding fused LV shell (from 3DUS) and vein model (from rotational X-ray) on the right

S91 Conclusion Combined X-ray and 3D Ultrasound imaging can be used as a tool for interventional guidance in CRT EP procedures. Registration of both image modalities can be performed using average computer models of the coronary sinus and mitral valve plane. Intra-procedural 3D image acquisition and registration, in particular intra-procedural ultrasound, remains a significant challenge due to the sterile operating field (3D ultrasound) and procedure time constraints (3D coronary vein modelling). Coronary sinus anatomy, however, varies significantly from patient to patient. Hence a more accurate model based on more patient data would be required to increase the feasibility and accuracy of the technique.

Image Guidance for Coronary Artery Bypass Grafting Christine Hartung1, Claudia Gnahm1, Reinhard Friedl2, Martin Hoffmann3, Klaus Dietmayer1 1 Institute of Measurement, Control and Microtechnology; University of Ulm, GERMANY 2 Dept. of Heart Surgery; University Hospital of Ulm, GERMANY 3 Dept. of Diagnostic Radiology; University Hospital of Ulm, GERMANY Keywords Coronary Artery Bypass Grafting; Image Guidance; Navigation; Registration Purpose Open heart surgery with coronary artery bypass grafting (CABG) is the standard treatment for advanced cases of coronary heart disease [1]. Optimal placement of the bypass graft anastomosis is of utmost importance for the success of the procedure. Therefore, detailed and precise knowledge about the path and morphology of the target vessel is crucial for the operating surgeon. To provide such information during the procedure, the Cardio-Pointer project aims at developing a surgical navigation system for open heart CABG surgery. In other surgical disciplines navigation systems are already in clinical use [2–4]. For cardiac applications, research has been done prototypically in minimally invasive surgery [5]. It is likely that an intraoperative navigation system would lead to a higher precision in CABG surgery by standardizing the difficult and often subjective process of determining the anastomosis site, reduce the ischeamic time of the heart, and therefore lead to a safer operation with less risk for the patient and better outcome. The proposed system merges a patient-specific vessel map of the coronary arteries, generated from multi-slice computed tomography (MSCT), with intraoperative data recorded with a stereo camera system. A registration method based on mutually-shared anatomical landmarks on the heart surface is presented. Successful registration enables surgical navigation assistance, such as visualization of the coronary arteries and plaque positions in the video images. Methods In order to establish a reliable registration database, mid-diastolic (75% RR) MSCT image data with a resolution of 0.4 9 0.4 9 0.5 mm3 are acquired utilizing retrospective gating methods. The coronary artery centerlines and the plaque positions are segmented from the CT data to generate a three dimensional model of the coronary arteries. Intraoperatively, stereo image data are obtained with a stereo camera system which is mounted on a robotic arm placed above the patient (see Fig. 1 right). Single images from the stereo camera system are used for image-based extraction of landmarks, while 3D surface information is extracted from the image pairs. The registration results are visualized by projecting the coronary artery model to the video images obtained with the stereo camera system. See Fig. 2 left for a schematic overview of the data processing pipeline. The stereo image data are recorded at the beginning of the procedure, directly after opening of the pericardium and before bypass grafting starts. The heart is not yet connected to the heart-lung

123

S92

Fig. 1 Left: overview of the data processing pipeline. Right: intraoperative recording of stereo image data

Fig. 2 Registration of preoperative MSCT data with intraoperative video images. Left: CT image with vessel centrelines and landmarks used for registration. Right: intraoperative video image with registered CT vessel centrelines and landmarks. Vessels visible in the video image are marked in green, CT centrelines are marked in red. Deviations increase with the distance from the LAD, while the target area on the LAD itself shows good accordance

machine. Hence, in this phase of the operation the shape of the heart can be expected to have the smallest deviations from its preoperative shape, which is the situation recorded by MSCT. For better access to the left anterior descending artery (LAD), which usually is the most important target vessel for bypass grafting on the front side of the heart, the heart is slightly luxated to the right side by pericardial stay sutures. To avoid delays while transfering the electrocardiogram (ECG) signal to the computer system, synchronization between the video images and the ECG signal is achieved by means of a fashing LED which marks each QRS peak directly in the video images. Corresponding to the MSCT data, a 75% RR image pair is used for registration. As external fiducial markers cannot be employed for the registration process, anatomical point landmarks on the heart surface are used to determine point correspondences between the datasets. There are two groups of potential anatomical landmarks: firstly artery bifurcations and contact points between different vessels, and secondly the appendages of the atrial auricles. Vessel bifurcations and contact points are sharply defined and contribute to an exact registration result, while the appendages of the auricles introduce a larger registration error due to their broad based rounded shape. Depending on the amount and distribution of epicardial fat, landmarks may be covered. Visible point landmarks available for the registration have to be determined for each patient individually and are manually selected in the corresponding datasets. Registration is

123

Int J CARS (2008) 3 (Suppl 1):S86–S93 performed in two steps. First, Procrustes methods [6] are applied to achieve a rigid 3D-3D registration including isotropic scaling. For this registration step, all available landmarks are used to determine the approximate global pose of the heart. Afterwards, the CT data are transformed to the stereo coordinate system. Using the projection matrix of the calibrated camera, the CT data are rendered from the perspective of one of the stereo cameras, yielding comparable video and CT images (see Fig. 3). After projecting the 3D point landmarks to the images, the registration is refined by a 2D-2D rigid image registration using only point landmarks on the LAD itself. The second registration step is performed to improve the registration result for the LAD region, for which a small target registration error is of utmost importance. The accuracy of the registration increases with the number of available landmarks on the LAD. Limitations arise if too few landmarks are found on the LAD. This situation can be addressed by using extended registration methods that are currently under investigation. Results The presented methods were evaluated on real patient data consisting of corresponding MSCT and stereo image data. The stereo image data were recorded directly in the operating room during CABG procedures and examined in retrospective fashion. In a first step, the datasets were inspected for mutually-present anatomical point landmarks. For datasets with a sufficient number of LAD landmarks, a successful registration was realized. For these datasets, a RMS fiducial registration error \ 5 mm arose from the 3D-3D Procrustes registration step. For the projected landmarks, the RMS error in image coordinates amounts to less than 7 pixel, corresponding to 4 mm in x-y-direction. After the second registration step, the image error for the LAD landmarks was reduced to 3 pixel, that is 2 mm, while the error for the auricle landmark increased. A qualitative evaluation of the registration accuracy for the target area was obtained by labeling visible vessels in the video image and projecting the corresponding vessel centerlines segmented in the CT data to the same image. As can be seen in Fig. 2, the path of the LAD and the diagonal branch (DB) centerlines obtained from MSCT is well aligned with the path of the corresponding vessels labeled in the intraoperative image. Figure 3 shows how successful registration can be used to visualize preoperative knowledge about path and morphology of the coronaries to assist the surgeon in finding the optimum anastomosis site. A model of the LAD and the DB was projected to the video image on the right. Furthermore, the 3D position and extension of the plaque as displayed in the curved planar reformation (CPR) image on the left was extracted from the CT dataset. The corresponding region on the model of the LAD was marked in a different color to visualize this information in the video images. Conclusion Registering a preoperative MSCT vessel map with intraoperative data is the crucial step to providing additional information during CABG

Fig. 3 Left: CPR image of the LAD with plaque. Right: a model of the LAD is projected to the image, the plaque region is colored in red

Int J CARS (2008) 3 (Suppl 1):S86–S93 surgery that is not available in current clinical practice. This information can assist the surgeon in a fast and accurate localization of the optimum anastomosis site. Registration can be performed using anatomical point landmarks on the heart surface. In the context of the Cardio-Pointer project, further improvement of this registration algorithm will be explored. Such extended registration methods will explore additional anatomical features like the shape and path of visible vessels, and surface models generated from CT and stereo camera data. The research is based on real patient datasets which are acquired directly in the cardiac operating room on a regular basis. Acknowledgments This work is supported by the Federal Ministry of Education and Research (BMBF), project 01EZ0614. References 1. Mustafa Zakkar and Phillip Hornick. Surgery for coronary artery disease. Surgery, 25(5):231–237, 2007.

S93 2. Terry M. Peters. Image-guided surgery: From X-rays to virtual reality. Comput Methods Biomech Biomed Engin. 4(1):27–57, 2000. 3. E. Grimson, M. Leventon, G. Ettinger, A. Chabrerie, F. Ozlen, S. Nakajima, H. Atsumi, R. Kikinis, and P. Black. Clinical experience with a high precision image-guided neurosurgery system. MICCAI (LNCS, vol. 1496):63–73, 1998. 4. Branislav Jaramaz, Mahmoud A. Hafez, and Anthony M. DiGioia. Computer-assisted orthopaedic surgery. Proceedings of the IEEE, 94(9):1689–1695, 2006. 5. Fabien Mourgues, Thierry Vieville, Volkmar Falk, and E`ve Coste-Manie`re. Interactive guidance by image overlay in robot assisted coronary artery bypass. MICCAI (LNCS, vol. 2878):173–181, 2003. 6. J.C. Gower. Generalized procrustes analysis. Psychometrika, 40:33–51, 1975

123