Image-Guided Transapical Aortic Valve Implantation

ORIGINAL ARTICLE

Image-Guided Transapical Aortic Valve ImplantationSensorless Tracking of Stenotic Valve Landmarks in Live Fluoroscopic

Images

Denis R. Merk, MD, PhD,*† Mohamed Esmail Karar, MSc,‡ Claire Chalopin, PhD,‡David Holzhey, MD, PhD,† Volkmar Falk, MD, PhD,§ Friedrich W. Mohr, MD, PhD,†

and Oliver Burgert, PhD‡

Objective: Aortic valve stenosis is one of the most frequently acquiredvalvular heart diseases, accounting for almost 70% of valvular cardiacsurgery. Transapical transcatheter aortic valve implantation has recentlybecome a suitable minimally invasive technique for high-risk andelderly patients with severe aortic stenosis. In this article, we aim toautomatically define a target area of valve implantation, namely, thearea between the coronary ostia and the lowest points of two aorticvalve cusps. Therefore, we present a new image-based tracking methodof these aortic landmarks to assist in the placement of aortic valveprosthesis under live 2D fluoroscopy guidance.Methods: We propose a rigid intensity-based image registration tech-nique for tracking valve landmarks in 2D fluoroscopic image se-quences, based on a real-time alignment of a contrast image includingthe initialized manual valve landmarks to each image of sequence. Thecontrast image is automatically detected to visualize aortic valve fea-tures when the aortic root is filled with a contrast agent.Results: Our registration-based tracking method has been retrospec-tively applied to 10 fluoroscopic image sequences from routinetransapical aortic valve implantation procedures. Most of all testedfluoroscopic images showed a successful tracking of valve land-marks, especially for the images without contrast agent injections.

Conclusions: A new intraoperative image-based method has beendeveloped for tracking aortic valve landmarks in live 2D fluoro-scopic images to assist transapical aortic valve implantations and toincrease the overall safety of surgery as well.

Key Words: Aortic valve, Biomedical image processing, Image-guided interventions, Minimally invasive valvular cardiac surgery,X-ray fluoroscopy.

(Innovations 2011;6:231–236)

Aortic valve stenosis is the most frequently acquired val-vular heart disease and is responsible for approximately

70% of all valve surgery.1 Between 2% to 4% of the popu-lation older than 65 years in the United States and 43% of allpatients with valvular diseases in Europe suffer from degen-erative aortic stenoses.2,3

Aortic valve replacement via sternotomy and cardio-pulmonary bypass (CPB) is the gold standard in treatingaortic valve pathologies.2,4 However, nearly 30% of elderlypatients with severe and symptomatic aortic stenosis are notsuitable candidates for standard surgical replacement of theaortic valve due to their comorbidities and are therefore highsurgical risk.5 Consequently, a new approach such as mini-mally invasive transcatheter aortic valve implantation (TAVI)presents a good treatment solution for high-risk patients.6–9

Compared with the standard treatment of aortic valvestenosis, the TAVI limits the surgical access to either aminithoracotomy (transapical TAVI) or a transfemoral TAVIapproach. Another advantage of the TAVI is the possibility toperform beating heart surgery without any CPB support andtherefore decrease the side effects of CPB.10

The main advantage of using the transapical TAVItechnique is the direct access to the aortic valve, eliminatingthe need of a large peripheral vascular access as needed forthe transfemoral approach and therefore being advantageousfor patients with peripheral vascular disease, small tortuousvasculature, a history of major vascular complications, and/orprevious vascular interventions.2 In transapical TAVI, astented aortic valve prosthesis (AVP) is temporarily crimpedupon a balloon catheter and inserted via a left lateral minitho-racotomy through the apex into the aortic root. After the valve is

Accepted for publication June 30, 2011.From the *Department of Cardiothoracic Surgery, School of Medicine,

Stanford University, Stanford, CA USA; †Department of CardiothoracicSurgery, Heart Center Leipzig, University of Leipzig, Leipzig, Germany;‡Innovation Center Computer Assisted Surgery (ICCAS), University ofLeipzig, Leipzig, Germany; and §Division of Heart and Vascular Sur-gery, University Hospital Zurich, Zurich, Switzerland.

Supported by German Academic Exchange Service (DAAD) in cooperationwith the Egyptian Supreme Council of Universities (ESCU) under scholar-ship no. A0690520. The Innovation Center Computer Assisted Surgery(ICCAS) at Faculty of Medicine, University of Leipzig, is funded byGerman Federal Ministry of Education and Research (BMBF) and SaxonMinistry of Science and Fine Arts (SMWK) in the scope of the initiative“Unternehmen Region” with grants 03 ZIK 031 and 03 ZIK 032.

Disclosure: The authors declare no conflict of interest.

Address correspondence and reprint requests to Mohamed Esmail Karar,MSc, Innovation Center Computer Assisted Surgery (ICCAS), Univer-sity of Leipzig, Semmelweisstrasse 14, D-04103 Leipzig, Germany.E-mail: [email protected].

Copyright © 2011 by the International Society for Minimally InvasiveCardiothoracic SurgeryISSN: 1556-9845/11/0604-0231

Innovations • Volume 6, Number 4, July/August 2011 231

considered to be positioned correctly, the balloon-expandableprosthesis is deployed under rapid ventricular pacing to its finaldiameter, fixing the prosthesis in the aorta and pressing thenatural calcified valve in the aortic annulus (Fig. 1a).

Once the AVP has been deployed, it cannot berepositioned. Hence, exact valve positioning is crucial. Aprosthesis positioned too high can obstruct the coronaries.On the other hand, a valve positioned too low can result indamaged mitral valve leaflets and/or leads to severe para-valvular aortic regurgitation.

Other complications occurring during TAVI can affectthe conducting system resulting in atrioventricular block (10%–30%) or an insufficient valve with regurgitation or a paravalvu-lar leakage (2%–35%). Further complications include coronaryostia occlusion (0.5%–1%), cardiac tamponade (1%–9%), oracute aortic dissections (0%–4%).11,12 AVP malpositioning dur-ing implantation is rare and occurs with an incidence of 1.2% to5.3% when using the transapical technique.12,13 The 30-day mor-tality for transapical TAVI in Europe is between 5% to 11%.12,14,15

The placement of the AVP is typically carried out underlive 2D x-ray fluoroscopy guidance.10,16 However, tracking ofthe stenotic aortic valve features in intraoperative fluoroscopicimages is a challenging task, because the aortic root includingthe anatomic landmarks is only visible under fluoroscopy guid-ance with contrast agent. On the other side, an excessive usageof contrast agent can lead to renal insufficiency especially in thehigh-risk elderly patient population scheduled for TAVI.

Only a few studies deal with intraoperative image-guided TAVI procedure. A guidance system for assisting theTAVI has been proposed, including a planning system and atracking system for the crimped bioprosthesis.17–19 Anothersystem used interventional angiography and a fluoroscopyC-arm system to overlay a static 3D rendered volume ontolive 2D fluoroscopic images.20 Other systems use roboticsystems with magnetic resonance imaging (MRI) guidancefor image-guided therapy.21,22

The motivation of this study is to automatically find theexact position for the prosthesis without additional usage ofcontrast agent under 2D fluoroscopy guidance. Tracking andcontinuously visualizing the moving anatomic aortic valvelandmarks, such as the coronary ostia and the lowest points ofthe cusps, are the key features for this automation process.

Image registration is a key technology that enables newtechniques for image-guided minimally invasive interventions.23

It is a process that geometrically aligns two images acquiredfrom the same patient at different times or with different imagingmodalities. Real-time alignment of 2D fluoroscopic images atdifferent acquisition times are used to determine the landmarkposition in each fluoroscopic image of live sequences.

To increase the overall safety for TAVI, we present anew method of intraoperative tracking of aortic valve land-marks by defining the boundaries of the target area duringvalve implantation between the coronary ostia and the lowestpoints of two aortic valve cusps. Based on real-time imageregistration during implantation, this new technique assiststhe positioning of the prosthesis under live 2D fluoroscopyguidance as shown in Figure 1b.

METHODSWe retrospectively picked 10 datasets of 2D fluoro-

scopic images for patients receiving a regular transapicalTAVI using Edwards SAPIEN prosthesis (Edwards Life-sciences Inc., Irvine, CA USA). The SAPIEN prosthesis isone of the commercially available transcatheter prosthesesused and approved for the transapical TAVI.14

Figure 2 shows the flowchart of the developed image-based tracking method of four valve landmarks in live 2Dfluoroscopic images. The four necessary anatomic landmarks arethe coronary ostia and the lowest points of two aortic cusps. Theinitial positions of the valve landmarks are manually defined ona contrast image. In this study, the contrast image is automati-cally detected when the aortic root roadmap is clearly visible inone fluoroscopic image of the sequence because of contrastagent injection. Because the contrast-enhanced aortic root in-cluding stenotic valve features shows dark pixels in fluoroscopicimages, a histogram analysis of these fluoroscopic images wasused to capture the required contrast image (Fig. 3b).24

Once the positions of the valve landmarks have beeninitialized, automatic image-based tracking procedure of thelandmarks is started as follows: first, each input image of thesequence and the contrast image have to be preprocessed toreduce the image noise by using median filter.

Second, a rigid intensity-based image registration frame-work between the input image as a fixed image and the contrast

FIGURE 1. a, Schematic view of transapical aorticvalve implantation using Edwards SAPIEN prosthesis.b, Live fluoroscopic image shows important land-marks for valve implantation between the coronaryostia and the lowest points of two cusps (greenpoints) for assisting the transapical approach.

Merk et al Innovations • Volume 6, Number 4, July/August 2011

Copyright © 2011 by the International Society for Minimally Invasive Cardiothoracic Surgery232

image as a moving image is applied, based on the InsightSegmentation and Registration Toolkit (ITK).25 The image reg-istration framework includes four components:

Y A mutual information metric by Mattes et al26 has beenused to define how well the two images match eachother based on similar intensity features between theinput image and the predefined contrast image (Fig. 3).These matched objects are the transesophageal echocar-diography probe, guide wires, the catheter, the ribspreader, and the valve delivery system.

Y A linear interpolation allows intensity estimation of themoving image in a nongrid position after mapping thefixed image space onto the moving image space.

Y A transformation maps the fixed image onto the movingimage by resolving the translational misalignment be-tween the two images to overlap the same objects inboth images.

Y A regular step gradient descent optimizer is applied withfour iterations to explore the optimal values of thesetranslational parameters in x-y coordinates that allow toupdate the registered image transformation in real time.

Third, the initial positions of the valve landmarks in thecontrast image are updated according to the final transform param-eters in current processed image. Finally, the new position of thevalve landmarks is overlaid onto each image of the sequence tovisualize the boundaries of the target area for the valve implantationin the 2D fluoroscopic images during the intervention.

Evaluation of the registration-based tracking procedure isa challenging task, because of the lack of contrast agent in mostfluoroscopic images of the sequence. Moreover, there is noground truth image data for the aortic root including the valvelandmarks. Therefore, all fluoroscopic images have been visuallyassessed and qualitatively inspected to validate the quality score ofdistance localization errors of the target.27 The quality scores of the

FIGURE 2. Flowchart of the image registration-based tracking of aortic valve landmarks trackingwhich are the coronary ostia and the two lowestpoints of the aortic valve cusps (green points) inlive 2D fluoroscopic images.

FIGURE 3. a, Input fluoroscopic image. b, De-tected contrast image have similar intensity-basedfeaturing transesophageal echocardiographyprobe, guide wires, catheter, rib spreader, andvalve delivery system.

Innovations • Volume 6, Number 4, July/August 2011 Sensorless Tracking of Aortic Valve Landmarks

Copyright © 2011 by the International Society for Minimally Invasive Cardiothoracic Surgery 233

tracked valve landmarks are estimated as follows: (a) good isdefined as an excellent alignment to a minimal misalignmentbetween the input sequences and the registered contrast image witha difference of less than 2 mm; (b) moderate is defined when arelatively high misalignment between the two images occurs be-tween 2.0 and 5.0 mm; and (c) insufficient is defined as a significantmisalignment of more than 5.0 mm between both images.

RESULTSThe developed tracking method of aortic valve land-

marks was tested and evaluated on different fluoroscopicimage sequences of 10 patients during routine transapicalTAVI. The dimensions of fluoroscopic images are 1024 �1024 pixels, with a pixel size of approximately 0.2 mm. All

fluoroscopic images sequences were acquired during routineangiographic and fluoroscopic x-ray imaging with a floormounted C-arm system (Artis zeego, Siemens AG, Health-care Sector, Forchheim, Germany).

Figure 4 shows a screenshot of the developed fluoroscopicguidance software for assisting the TAVI. The image-registra-tion tracking algorithm has been implemented using C��programming language being suitable for real-time image pro-cessing with a frame rate of 10 to 15 frames per second.

In Figure 5, two different sets of tested fluoroscopicimages are presented showing successful results of valvelandmark tracking, based on the intensity-based image registra-tion. The contrast images including the manual localization ofaortic valve landmarks are shown in Figures 5a, b. To visualize

FIGURE 4. Screenshot of fluoroscopic guidancesoftware developed by us. The contrast image isautomatically detected and the user locates thefour valve landmarks (the coronary ostia and thelowest points of the aortic cusps).

FIGURE 5. Two successful exam-ples of image registration-basedtracking of aortic valve landmarks.a and b, The contrast images withmanual localization of landmarks. cto f, Checkerboard registration re-sults of contrast images alignmentwith live fluoroscopic images eitherin the presence of contrast agentor without contrast agent.

Merk et al Innovations • Volume 6, Number 4, July/August 2011

Copyright © 2011 by the International Society for Minimally Invasive Cardiothoracic Surgery234

the matching accuracy between the contrast image (movingimage) and the intensity features of each live fluoroscopic image(fixed image), checkerboard images are shown as alternatingblocks from the contrast image and the input fluoroscopic imagein the presence of contrast agent (Figs. 5c, d) and withoutcontrast agent as shown in Figures 5e, f.

Significant errors of valve landmark tracking appearedin the image sequence (seq. 7) among all the tested datasets(Fig. 6). Although the contrast image has been successfullyautomatically detected as shown in Figure 6a, checkerboardregistration visualizes the visible discrepancies between theinput fluoroscopic images and the detected contrast image.

The registration-based tracking algorithm has been qual-itatively evaluated for all tested images of the fluoroscopicsequences as a percentage of the total number of images persequence as shown in Figure 7. In image sequences 6, 7, and 10,insufficiently registered landmarks were observed. The seq. 7showed the maximum percentage of images with an insufficientscore of 25% of all images per sequence, due to severe imagenoise and a small number of similar intensity features betweenthe contrast image and each image of the sequence. The mod-erate score of landmarks tracking did not exceed 40% in seq. 5among all the tested image sequences. However, most images ofthe 10 fluoroscopic sequences showed successful tracking of

landmarks with a good score between 60% and 95% of allfluoroscopic images, such as sequences 2, 3, and 8.

DISCUSSIONThe positioning of the AVP is the most critical step of

the transapical TAVI procedure under 2D fluoroscopy guid-ance. Avoiding usage of additional contrast agent is a keyfactor to solve the current limitations of single-plane x-rayfluoroscopy for the image guidance of TAVI interventions.

We demonstrated in this study that 2D fluoroscopic imageregistration is potentially a feasible method for intraoperativetracking of the aortic valve landmarks to assist the position of theAVP without further contrast agent injections. Moreover, the mainadvantage of image-based tracking methods over optical and elec-tromagnetic tracking systems is that image-based guidance systemsdo not require expensive instrumentation or sophisticated imagingtechnologies. Therefore, our image register tracking-based algo-rithm provides an appropriate solution for real-time tracking ofanatomic valve landmarks in live 2D fluoroscopic images.

The qualitative evaluation of the registration performanceshowed that the quality of registration based on landmark algo-rithm tracking is good as long as the alignment errors are lessthan 2.0 mm as seen in most tested sequences. Three of 10fluoroscopic image sequences showed misalignment in the range

FIGURE 6. Significant misalign-ment results of image sequence 7. a,Contrast image with manual land-marks. b and c, the contrast image isnot correctly aligned in two live fluo-roscopic images with and withoutcontrast agent respectively.

FIGURE 7. Qualitative evaluation of image registration-based tracking method for all tested fluoroscopic image sequences.Insufficient landmarks tracking are only observed in image sequences 6, 7, and 10. However, most of all fluoroscopic imagesshowed a good tracking of valve landmarks on �60% of images per the sequence.

Innovations • Volume 6, Number 4, July/August 2011 Sensorless Tracking of Aortic Valve Landmarks

Copyright © 2011 by the International Society for Minimally Invasive Cardiothoracic Surgery 235

of 1% to 25% insufficient registered images per sequence. Severeimage noise and the small number of similar intensity features suchas transesophageal echocardiography probe and guide wires af-fected the image registration accuracy for seq. 7 (Fig. 6). However,the alignment of fluoroscopic images is still valid and optimized byusing the capabilities of image registration framework.

CONCLUSIONSWe have developed a new sensorless tracking method

of stenotic aortic valve landmarks for assisting the transapicalTAVI under 2D fluoroscopy guidance. This method canprovide a helpful tool for the surgeon by automaticallydefining the desired position of the AVP. In addition, thedeveloped method is used at low levels of contrast agentinjections for controlling the final placement of the prosthesisto reduce long-term negative effects, such as renal failure, inpatients. Therefore, the accuracy of valve implantation willincrease; surgery time can be significantly reduced whileincreasing the overall safety of the surgical procedure.

To minimize the user interaction and to increase theaccuracy of TAVI guidance, we are currently integrating the3D information of the aortic root model including the valvelandmarks from intraoperative C-arm computed tomographicimages28 with live 2D fluoroscopic images.

REFERENCES1. Coeytaux RR, Williams JW, Gray RN, Wang A. Percutaneous heart

valve replacement for aortic stenosis: state of the evidence. Ann InternMed. 2010;153:314–324.

2. Singh IM, Shishehbor MH, Christofferson RD, et al. Percutaneoustreatment of aortic valve stenosis. Cleve Clin J Med. 2008;75:805–812.

3. Iung B, Baron G, Butchart EG, et al. A prospective survey of patientswith valvular heart disease in Europe: The Euro Heart Survey onValvular Heart Disease. Eur Heart J. 2003;24:1231–1243.

4. Bonow RO, Carabello BA, Chatterjee K, et al. 2008 Focused updateincorporated into the ACC/AHA 2006 guidelines for the management ofpatients with valvular heart disease: a report of the American College ofCardiology/American Heart Association Task Force on Practice Guide-lines (Writing Committee to Revise the 1998 Guidelines for the Man-agement of Patients With Valvular Heart Disease): endorsed by theSociety of Cardiovascular Anesthesiologists, Society for CardiovascularAngiography and Interventions, and Society of Thoracic Surgeons.Circulation. 2008;118:e523–e661.

5. Sherif M, Tolg R, Richardt G. Percutaneous aortic valve replacement:results update, future trends and new challenges. Clin Res Cardiol Suppl.2009;4:102–107.

6. Cribier A, Eltchaninoff H, Bash A, et al. Percutaneous transcatheterimplantation of an aortic valve prosthesis for calcific aortic stenosis: firsthuman case description. Circulation. 2002;106:3006–3008.

7. Grube E, Schuler G, Buellesfeld L, et al. Percutaneous aortic valve replace-ment for severe aortic stenosis in high-risk patients using the second- andcurrent third-generation self-expanding CoreValve prosthesis: device suc-cess and 30-day clinical outcome. J Am Coll Cardiol. 2007;50:69–76.

8. Walther T, Falk V, Borger M, et al. Minimally invasive transapicalbeating heart aortic valve implantation—proof of concept. Eur J Car-diothorac Surg. 2007;31:9–15.

9. Webb JG, Pasupati S, Humphries K, et al. Percutaneous transarterialaortic valve replacement in selected high-risk patients with aortic ste-nosis. Circulation. 2007;116:755–763.

10. Walther T, Dewey T, Borger MA, et al. Transapical aortic valveimplantation: step by step. Ann Thorac Surg. 2009;87:276–283.

11. Yan TD, Cao C, Martens-Nielsen J, et al. Transcatheter aortic valveimplantation for high-risk patients with severe aortic stenosis: a system-atic review. J Thorac Cardiovasc Surg. 2010;139:1519–1528.

12. Wendler O, Walther T, Schroefel H, et al. The SOURCE Registry: whatis the learning curve in trans-apical aortic valve implantation? EurJ Cardiothorac Surg. 2011;39:853–860.

13. Al Ali AM, Altwegg L, Horlick EM, et al. Prevention and managementof transcatheter balloon-expandable aortic valve malposition. CatheterCardiovasc Interv. 2008;72:573–578.

14. Thomas M, Schymik G, Walther T, et al. Thirty-Day Results of theSAPIEN Aortic Bioprosthesis European Outcome (SOURCE) Registry:a European registry of transcatheter aortic valve implantation using theEdwards SAPIEN valve. Circulation. 2010;122:62–69.

15. Walther T, Schuler G, Borger M, et al. Transapical aortic valve implan-tation in 100 consecutive patients: comparison to propensity-matchedconventional aortic valve replacement. Eur Heart J. 2010;31:1398:1403.

16. Van de Veire N. Imaging to guide transcatheter aortic valve implanta-tion. J Echocardiogr. 2010;8:1–6.

17. Gessat M, Merk DR, Falk V, et al. A planning system for transapical aorticvalve implantation. Conf Proc SPIE Med Imaging. 2009;7261:72611E.

18. Karar ME, Gessat M, Walther T, et al. Towards a new image guidancesystem for assisting transapical minimally invasive aortic valve implan-tation. Conf Proc IEEE Eng Med Biol Soc. 2009;2009:3645–3648.

19. Karar ME, Merk DR, Chalopin C, et al. Aortic valve prosthesis trackingfor transapical aortic valve implantation. Int J Comput Assist RadiolSurg. In press.

20. John M, Liao R, Zheng Y, et al. System to guide transcatheter aorticvalve implantations based on interventional C-arm CT imaging. MedImage Comput Comput Assist Interv. 2010;13(Pt 1):375–382.

21. Li M, Mazilu D, Horvath K. Robotic system for transapical aortic valvereplacement with MRI guidance. In: Metaxas D, Axel L, Fichtinger G, SzekelyG, eds. Medical Image Computing and Computer-Assisted Intervention—MICCAI 2008. Vol. 5242. Berlin/Heidelberg: Springer; 2008:476–484.

22. Li M, Kapoor A, Mazilu D, Horvath KA. Pneumatic actuated roboticassistant system for aortic valve replacement under MRI guidance. IEEETrans Biomed Eng. 2011;58:443–451.

23. Markelj P, Tomazevic D, Likar B, Pernus F. A review of 3D/2D registrationmethods for image-guided interventions. Med Image Anal. In press.

24. Condurache AP, Aach T, Eck K, Bredno J. Fast detection and processingof arbitrary contrast agent injections in coronary angiography andfluoroscopy. Paper presented at: Bildverarbeitung fur die Medizin 2004(Algorithmen, Systeme, Anwendungen); March 29–30, 2004; Berlin.

25. Kitware. The Insight Segmentation and Registration Toolkit; 2010.Available at: www.itk.org.

26. Mattes D, Haynor DR, Vesselle H, et al. PET-CT image registration in the chestusing free-form deformations. IEEE Trans Med Imaging. 2003;22:120–128.

27. Kolar R, Kubecka L, Jan J. Registration and fusion of the autofluorescentand infrared retinal images. Int J Biomed Imaging. 2008;2008:513478.

28. Zheng Y, John M, Liao R, et al. Automatic aorta segmentation and valvelandmark detection in C-arm CT: application to aortic valve implanta-tion. Med Image Comput Comput Assist Interv. 2010;13(Pt 1):476–483.

CLINICAL PERSPECTIVEThis is a report of an innovative new technique for sensorless tracking of stenotic valve landmarks using 2D fluoroscopic imaging to assist inthe accurate placement of a transapical transcatheter aortic valve. This is a timely report as transcatheter aortic valve implantation (TAVI) isfrequently performed in Europe and has just been approved for inoperable patients in the U.S. Mispositioning during TAVI can have disastrousand, in rare instances, fatal consequences. The advantages of this technique include that it does not require other expensive instrumentation, such asMRI guidance, or additional usage of contrast agents. The disadvantage is that there was insufficient image tracking in 3 of 10 images, and thepercentage of good tracking ranged from under 50% to 95%. Further work will be needed to define the utility of this new tracking methodology.

Merk et al Innovations • Volume 6, Number 4, July/August 2011

Copyright © 2011 by the International Society for Minimally Invasive Cardiothoracic Surgery236