- Email: [email protected]
Virtual positioning of ventricular assist device for implantation planning
Disponible en ligne sur
ScienceDirect www.sciencedirect.com IRBM 36 (2015) 317–323
RITS 2015
Virtual positioning of ventricular assist device for implantation planning S. Collin a,b,∗ , A. Anselmi a,b,c , J.P. Verhoye a,b,c , P. Haigron a,b , E. Flecher a,b,c a INSERM, U1099, Rennes, F-35000, France b Université de Rennes 1, LTSI, Rennes, F-35000, France c Pontchaillou University Hospital, Rennes, F-35000, France
Received 18 January 2015; received in revised form 30 June 2015; accepted 4 September 2015 Available online 9 October 2015
Abstract The use of Ventricular Assist Devices (VAD) is increasing in the context of refractory heart failure. Nevertheless, there is still a high rate of complications. This preliminary work analyzes more precisely the clinical needs and proposes a first solution for preoperative planning of device implantation. The proposed approach consists in representing within a common space the 3D mesh describing the device and the patient CT image, in order to interactively simulate the device positioning and detect collisions between the VAD and different kinds of surrounding anatomical structures (bones and right ventricle). CT scans from 3 adult patients who have previously received a VAD, were used for the experiments. We analyzed the influence of mesh precision on computation time and accuracy of collision detection. Results show that the proposed approach is compatible with fast and interactive simulation of virtual device positioning, in order to preoperatively plan its implantation. Such a solution could also facilitate the decision-making about the choice of the device taking into consideration the feasibility of implantation. © 2015 AGBM. Published by Elsevier Masson SAS. All rights reserved. Keywords: Image processing; Modeling; Simulation; Surgery planning; Implantable device
1. Introduction 1.1. Context Heart Failure (HF) occurs when the heart cannot pump enough blood to meet the body’s needs. This disease is the principal cause of death in Europe and the United States, and is also an economic burden due to the high medical costs (an estimated $32 billion each year for the USA nation [1]). Nowadays, over 23 million of people are living with a HF worldwide. Cardiac transplantation is the first solution for end stage HF but the lack of donors is significant and approximately 300,000 die annually. The number of people living with this disease is rising inexorably with the ageing population while there is stagnation in heart transplantation [2]. * Corresponding author.
E-mail address: [email protected] (S. Collin). http://dx.doi.org/10.1016/j.irbm.2015.09.005 1959-0318/© 2015 AGBM. Published by Elsevier Masson SAS. All rights reserved.
Over the last decades, implantable long-term mechanical Ventricular Assist Devices (VADs) have significantly evolved. These devices offer an alternative to delay or avoid the heart transplantation, in order to remedy the lack of donors, and allow patients a better mobility and quality of life [3]. However, although figures show an increase of their implantation, perand post-operative complications are still reported and are explained by the complexity of the clinical condition of each patient [4] and of the interventional procedure. To limit them patient selection and preoperative management are important [5]. Therefore, there is a need for dedicated preoperative software to help surgeon to plan the implantation of VADs and/or artificial hearts. The aim of this preliminary work was to analyze the clinical needs and to propose a first solution for preoperative planning of device implantation. In the following, we analyze more precisely the clinical needs. Then we present the solution proposed for the simulation of LVAD implantation, and report the results obtained from preoperative patient CT dataset.
318
S. Collin et al. / IRBM 36 (2015) 317–323
Fig. 1. Artificial hearts organization.
1.2. Clinical needs 1.2.1. Artificial hearts Cardiac pumps are composed of two external batteries, an external controller, an internal pump (second generation devices), and a percutaneous driveline that connects the controller to the pump. They are divided into two categories: Total Artificial Heart (TAH) that replaces both ventricles and partial artificial heart more commonly called Ventricle Assist Device, used to support heart function and blood flow. Among VAD, one can differentiate Left Ventricular Assist Devices (LVAD) from Right Ventricular Assist Devices (RVAD) and BiVentricular Assist Devices (Bi-VAD) [6]. An organization of these devices is given in Fig. 1. Our work focuses on two commercially available LVADs implanted at CHU—Rennes (HeartMate II (HM II) – Thoratec Inc., Pleasanton, CA; HeartWare (HW) VAD – HeartWare Inc., Framingham, MA). Their composition includes an inflow cannula placed through the apex of the LV, an outflow cannula anastomosed to the ascending aorta, and the pump with a continuous- or pulsatile-flow rotator, refer to Fig. 2. The appropriate device is chosen by the surgeon considering the patient needs, his metabolism and specificities of each device.
1.2.2. Clinical requirements During the surgical procedure, the pump has to be placed in the attended anatomic location [7]. The proper positioning has to be evaluated for a long-term prevention of infection and damage. Nowadays, surgeons plan it preoperatively by examining patients’ data but some difficulties are still subsisting. They are related to: (i) the positioning of the device. Surgeons have to place the inflow cannula pointing toward the mitral valve. A cannula oriented toward the septum or other free wall facilitates inflow obstruction with potential device malfunction and the creation of thrombus. In the same way, the outflow cannula must be positioned with an orientation that gives the maximum flow into the ascending aorta. (ii) The device collision with anatomical structures. Collisions between the device and specific neighboring organs can cause post-operative complications related to the device functioning or the patient health. On the one hand, during the chest closure a device collision with the thorax can occur and increases the risk of damage, bleeding and malfunctioning, as well as of improper inflow. On the other hand, the RV compression due to the device positioning may facilitate RV dysfunction, which is one relatively common and dreadful complication encountered after a LVAD implantation [8]. Patients in this case may require heart transplantation. (iii) The deformations. Certain parts of the device can be distorted during the implantation. Additionally, the device start-up causes a depletion of the assisted ventricle. These deformations could provoke the displacement of the device, deflecting it from its initial placement. The work presented in this paper is aimed at assessing the collision between the LVAD and two surrounding anatomical structures that are: the chest wall and the RV. 2. Materials and methods This section presents the method proposed for the virtual positioning of LVADs. It involves the devices and patient-specific
Fig. 2. LVAD composition (HeartMate II).
S. Collin et al. / IRBM 36 (2015) 317–323
319
2.1. Data, user interface and interaction
wall and the right ventricle. A collision between the device and these critical organs can increase the risk of damage and death. In the CT images, bone structures that are dense have a high Hounsfield Unit (HU) value and can easily be isolated from other organs. In the proposed solution, the evaluation of the LVAD collision with the thorax is not based on bones segmentation but on the analysis of HU from the CT scan. The anatomical structure of the RV entering in contact with the device is the right ventricular myocardium (muscles). In the neighboring of the heart, there are soft tissues such as liver or abdominal muscles, showing gray values similar to the myocardium ones. To distinguish the myocardium from surrounding tissues we use an approach based on the segmentation of the heart. In this study, the inflow cannula which is inside the LV is not considered.
The first question is related to the description of the patient anatomy. Many segmentation techniques and automatic segmentations of specific structures had been presented in the literature. Their generalization can be complex and can require an important part of interaction with the practitioner. According to the number of anatomical structures required for this study, and because a reconstruction of selected organs will delete much information provided by image data, our approach is to use a 3D slice view (coronal, transversal and sagittal planes) of 3D medical images as representation of the patient and to limit the number of segmented organs for collision detection. For artificial heart implantation, surgeons currently use a CT scan to choose the most appropriate device. Based on the previous analysis of the clinical requirements, the image modality used for this study is a thoracic CT scan, synchronized to the cardiac rhythm in order to minimize artifacts due to heart’s movements. The slice thickness was less than 1 mm and an intravenous injection was used to have a contrast flow in heart ventricles. In order to allow the practitioner to simulate the position of the cardiovascular pump inside the patient, the devices are represented by a 3D mesh and integrated within the 3D slice view of the patient. An ex-vivo rotational imaging of the LVAD (Cone Beam Computed Tomography – CBCT) was used to describe the device. The 3D mesh representing the device was created with the active contour method from ITK-SNAP software (Philadelphia, PA) (Fig. 3) [9]. CamiTK software (Computer Assisted Medical Intervention Toolkit, Grenoble, France) is an open-source modular framework developed in order to be used by clinicians and that can display simultaneously the two kinds of data mentioned previously (3D mesh and CT image). This software was used to implement the virtual positioning application. To allow the user to displace in real-time the device inside the CT scan and to launch the collision detection, a positioning window created with Qt was added. 3D rigid linear transformations were implemented with VTK library.
2.2.1. Right ventricle collision Numerous methods of heart segmentation exist such as Zheng et al. (2008), Ecabert et al. (2008) or Zhuang et al. (2010) [10,11]. However, the reported methods are dedicated to the left atrium, left cavity, left myocardium, right atrium and right cavity. For the right ventricle, myocardium is too slim, epicardium and endocardium are confused. This thinness and the low-contrast in CT images make the segmentation very challenging [12]. The region of the right myocardium entering in contact with the device is located at the apex, where the thickness is regular. Therefore, we decided to describe the RV by its cavities and a margin approximating the myocardium. For the RV segmentation, we chose a method with a simple approach in order to focus on the collision detection. We used ITK-SNAP to perform an interactive segmentation. During preoperative examinations surgeons request an echocardiography that provides information such as myocardium thickness. For more precision, the margin can be fixed by the surgeon according to this exam. The method used for the RV collision detection is represented in Fig. 4. It is based on a distance map implemented by an algorithm based on Danielsson’s Euclidean Distance and developed with the ITK library. Devices are described by a set of points that are characterized by their coordinates x, y and z. To detect parts of the LVAD that are in superimposition with the RV, the process evaluates for each point pi composing the cardiovascular pump, the distance to the right ventricular cavity c via the distance map D(c, pi ). If the point is located inside the region defined by the myocardium thickness parameter, its CT value is checked to confirm that the gray level corresponds actually to soft tissues. Because the heart segmentation is necessary for the assessment of the RV compression, we also use it for the creation of a 3D mesh representing heart chambers. Isosurfaces are extracted with the marching cube module of VTK library and the user can therefore visualize the 3D reconstruction.
2.2. Collision detection
2.2.2. Thorax collision To evaluate parts of the device that are superimposed with the thorax, at each cardiovascular pump’s point, the process evaluates the grayscale value of the CT scan and determines
Fig. 3. 3D mesh of the HeartWare created from a CBCT scan with an active contour method.
anatomy representation, the user interface implementation and the collision detection process.
In this preliminary solution, two surrounding anatomical structures are considered for the collision detection: the thoracic
320
S. Collin et al. / IRBM 36 (2015) 317–323
Several positions of the different devices (HM II and HW) were tested in order to validate the collision detection. Parts of the LVAD in conflict with the thorax are highlighted in red, while those in conflict with the right ventricle are highlighted in blue. The red region of the HM II in Fig. 5b represents the collision with the chest wall, and the blue region in Fig. 5c represents conflict with the RV. Collisions between the selected LVAD and critical neighboring anatomical structures are well detected and represented, with this preliminary solution. 3.2. Execution time and precision of collision detection
Fig. 4. Ventricle collision algorithm.
according to this value, whether the point is in conflict with the chest wall. Solid bones have an associated HU range of values from 200 to 1000. Our approach consists on using a thresholding method with a value chosen higher than 200 to avoid any confusion with soft tissues. However, for a better visualization of heart ventricles and the mitral valve, heart cavities are injected and therefore defined by HU values from 400 to 600. Confusion can therefore be made between bones and heart chambers. In order to differentiate collisions caused by the different organs, we eliminate from this set of points those that are superimposed with the RV. 3. Results Preoperative CT scans used in our experiments come from 3 different adult patients who have previously received a LVAD at the University Hospital of Rennes, France. They were resampled and the number of slices varied from 210 to 400. All images cover the complete heart but not the abdomen. 3.1. Interaction and collision detection A part of the implemented user interface is displayed in Fig. 5. The user can interact in real time with a 3D mesh representing the device, and find a correct position with the 3 slices shown of the CT scan (axial, sagittal and coronal slice). Fig. 5a represents a good positioning with the inflow cannula toward the mitral valve, and without collision.
The number of points composing the pump has an impact on the response time related to collisions detection. Fig. 6 represents the computation time for chest wall collision according to 3D mesh precision of HM II. The higher the number of points composing the pump is, the longer the computation time is. However, results show that the collision detection time is fully compatible with a real-time simulation, even for a 3D mesh composed of an important number of points (300 000). To evaluate the precision of the collision detection, we considered different CT images (voxels precision varying from 0.39 to 1.0), and a 3D mesh with different precisions in the case of HM II device. This study consists on identifying the position of the device where the first collision is detected. The device was placed at an initial point without collision and a translation along the direction of the chest wall was performed every millimeters. Fig. 7 illustrates results obtained for 1 CT scan, but they are similar with the others. When the precision of the 3D mesh describing the device is too small, collision is detected farther, which means that the detection is less precise. Nevertheless, when the 3D mesh is too much detailed, the precision of collision detection is not necessarily improved. Results show that a 3D mesh composed of about 20 000 points, provides a good compromise between the collision detection performance and the computation time. 4. Discussion–conclusion Nowadays, the incidence rate of heart failure is increasing and the development of VADs is prosperous. Although the use of LVAD is in expansion [12], currently there is no tool to assist surgeons to preoperatively plan VADs implantation or help them in the choice of the device. In the context of artificial heart, a work has been recently reported in the Journal of Heart and Lung Transplantation. It proposes a virtual implantation evaluation of the total artificial heart (Syncardia device, TAH-t) and compatibility [13]. This work is intended for the pediatric and small adult populations, in order to improve patient eligibility. Materialise Inc., developed a software suite (MIS—Mimics Innovation Suite) based on CT and MRI data that provide the visualization of patient anatomy. They mentioned a possible application related to the design of patient-specific medical devices such as TAH (AbioCor artificial heart) [14]. In both cases, works are based on segmentation of multiple organs from preoperative data and surface rendering of the TAH. Unlike the implantation of TAH
S. Collin et al. / IRBM 36 (2015) 317–323
321
Fig. 5. Representation of region of conflict with the HeartMate II: a. no region of conflict; b. conflict with the chest wall; c. conflict with the RV. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
322
S. Collin et al. / IRBM 36 (2015) 317–323
Fig. 6. Computation time for chest wall collision according to 3D mesh precision for the HM II.
Fig. 7. Precision for the thoracic collision detection for one CT scan.
where heart ventricles are removed, we focus in our work on LVAD which raises specific problems related to the possible collisions between the device and different kinds of anatomical structures (bones and right ventricle). In our approach, we do not provide just a visualization of the different studied structures but regions in conflict are detected and highlighted in real time. The development of planning tools in the context of deformable and moving anatomical structures is currently a challenge. Nowadays, no application can estimate if the device is able to be inserted in the thoracic cavity or not. The objective of our work was aimed at providing a first solution that allows the user to assess the compatibility between the device congestion and the anatomy of a specific patient. Collisions are detected with a configuration of the heart at a given temporal phase. This could be considered as a limitation of our work. However, image acquisition is made in the diastole phase, where the heart occupies the largest volume in the chest wall. Furthermore, when the device starts-up, it causes a depletion of the assisted ventricle. Therefore, this depletion can, on the one hand, displace the implanted device from its specified position but, on the other hand, it provides more intrathoracic space, which means a greater tolerance. In this first approach, the geometry tolerance has not been considered as critical. In further work, an automatic research of the optimal position of the device could be considered. However, it will be
necessary to analyze and integrate additional decision criteria based on, for instance, the positioning and orientation of the cannula relative to the septum, the influence of the cannula position on the fluid dynamics. These different analyses depend also on the device considered. In this paper we proposed a preliminary solution that provides a personalized simulation allowing the interactive placement of 2 different LVADs. It can be used to detect collisions between the selected device and critical neighboring anatomical structures. This patient-specific approach can be used to interactively simulate the positioning of a LVAD, in order to preoperatively help surgeons to select the most appropriate device. Acknowledgements This work was partially supported by the French National Research Agency (ANR) in the framework of the Investissement d’Avenir Program through Labex CAMI (ANR-11LABX-0004-01). References [1] Heidenreivh PA, Trogdon JG, Khavjou OA, Butler J, Dracup K, Ezekowitz MD, et al. Forecasting the future of cardiovascular disease in
S. Collin et al. / IRBM 36 (2015) 317–323
[2] [3]
[4]
[5]
[6]
[7] [8]
the United States: a policy statement from the American Heart Association. Circulation 2011;123(8):933–44. http://www.agence-biomedecine.fr (Bilan 2011). Flécher E. Assistance circulatoire mécanique de longue durée en 2014: une alternative à la transplantation cardiaque ? Presse Méd July 2014;43(7–8):809–12. Slaughter MS, Pagani FD, Rogers JG, Miller LW, Sun B, Russell SD, et al. Clinical management of continuous-flow left ventricular assist devices in advanced heart failure. J Heart Lung Transplant April 2010;29(4):S1–39. http://dx.doi.org/10.1016/j.healun.2010.01.011. Carrel T, Englberger L, Martinelli MV, Takala J, Boesch C, Sigurdadottir V, et al. Continuous flow left ventricular assist devices: a valid option for heart failure patients. Swiss Med Wkly October 18, 2012;142. http://dx.doi.org/10.4414/smw.2012.13701. Flecher E, Joudinaud T, Grinda J-M. Histoire de L’assistance Circulatoire Mécanique et Du Coeur Artificiel. Annales de Chirurgie, vol. 131. Elsevier; 2006. p. 473–8. HeartMate II LVAD: Instructions for use. Pleasanton, CA: Thoratec Corporation; 2008. Garcia-Alvarez A, Fernandez-Friera L, Lau JF, Sawit ST, Mirelis JG, Castillo JG, et al. Evaluation of right ventricular function and postoperative findings using cardiac computed tomography in patients with left ventricular assist devices. J Heart Lung Transplant May 2011;30(8):896–903. http://dx.doi.org/10.1016/j.healun.2011.03.009.
323
[9] Yushkevich PA, Piven J, Cody Hazlett H, Gimpel Smith R, Ho S, Gee JC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. NeuroImage 2006 Jul.;31(1):1116–28. [10] Peters J, Ecabert O, Meyer C, Kneser R, Weese J. Optimizing boundary detection via simulated search with applications to multimodal heart segmentation. Med Image Anal February 2010;14(1):70–84. http://dx.doi.org/10.1016/j.media.2009.10.004. [11] Hoogendoorn C, Duchateau N, Sanchez-Quintana D, Whitmarsh T, Sukno FM, De Craene M, et al. A high-resolution atlas and statistical model of the human heart from multislice CT. IEEE Trans Med Imaging January 2013;32(1):28–44. http://dx.doi.org/10.1109/TMI.2012.2230015. [12] Anselmi A, Collin S, Verhoye JP, Haigron P, Flecher H. Pilot study for preoperative simulation of patient-device interactions in left ventricular assist device implantation. In: Plenary Session at the ‘Society for Cardiothoracic Surgery (SCTS)/Association of Cardiothoracic Anaesthetists (ACTA)’ Annual meeting, Manchester, 2015 March. [13] Moore RA, et al. Virtual implantation evaluation of the total artificial heart and compatibility: beyond standard fit criteria. J Heart Lung Transplant November 2014;33(11):1180–3. [14] Wivell C. Mimics revolutionizes life-saving device and therapy development. In: D.E. Desktop Engineering, editors. 2009. http://www. deskeng.com/de/mimics-revolutionizes-life-saving-device-and-therapydevelopment/.
ScienceDirect www.sciencedirect.com IRBM 36 (2015) 317–323
RITS 2015
Virtual positioning of ventricular assist device for implantation planning S. Collin a,b,∗ , A. Anselmi a,b,c , J.P. Verhoye a,b,c , P. Haigron a,b , E. Flecher a,b,c a INSERM, U1099, Rennes, F-35000, France b Université de Rennes 1, LTSI, Rennes, F-35000, France c Pontchaillou University Hospital, Rennes, F-35000, France
Received 18 January 2015; received in revised form 30 June 2015; accepted 4 September 2015 Available online 9 October 2015
Abstract The use of Ventricular Assist Devices (VAD) is increasing in the context of refractory heart failure. Nevertheless, there is still a high rate of complications. This preliminary work analyzes more precisely the clinical needs and proposes a first solution for preoperative planning of device implantation. The proposed approach consists in representing within a common space the 3D mesh describing the device and the patient CT image, in order to interactively simulate the device positioning and detect collisions between the VAD and different kinds of surrounding anatomical structures (bones and right ventricle). CT scans from 3 adult patients who have previously received a VAD, were used for the experiments. We analyzed the influence of mesh precision on computation time and accuracy of collision detection. Results show that the proposed approach is compatible with fast and interactive simulation of virtual device positioning, in order to preoperatively plan its implantation. Such a solution could also facilitate the decision-making about the choice of the device taking into consideration the feasibility of implantation. © 2015 AGBM. Published by Elsevier Masson SAS. All rights reserved. Keywords: Image processing; Modeling; Simulation; Surgery planning; Implantable device
1. Introduction 1.1. Context Heart Failure (HF) occurs when the heart cannot pump enough blood to meet the body’s needs. This disease is the principal cause of death in Europe and the United States, and is also an economic burden due to the high medical costs (an estimated $32 billion each year for the USA nation [1]). Nowadays, over 23 million of people are living with a HF worldwide. Cardiac transplantation is the first solution for end stage HF but the lack of donors is significant and approximately 300,000 die annually. The number of people living with this disease is rising inexorably with the ageing population while there is stagnation in heart transplantation [2]. * Corresponding author.
E-mail address: [email protected] (S. Collin). http://dx.doi.org/10.1016/j.irbm.2015.09.005 1959-0318/© 2015 AGBM. Published by Elsevier Masson SAS. All rights reserved.
Over the last decades, implantable long-term mechanical Ventricular Assist Devices (VADs) have significantly evolved. These devices offer an alternative to delay or avoid the heart transplantation, in order to remedy the lack of donors, and allow patients a better mobility and quality of life [3]. However, although figures show an increase of their implantation, perand post-operative complications are still reported and are explained by the complexity of the clinical condition of each patient [4] and of the interventional procedure. To limit them patient selection and preoperative management are important [5]. Therefore, there is a need for dedicated preoperative software to help surgeon to plan the implantation of VADs and/or artificial hearts. The aim of this preliminary work was to analyze the clinical needs and to propose a first solution for preoperative planning of device implantation. In the following, we analyze more precisely the clinical needs. Then we present the solution proposed for the simulation of LVAD implantation, and report the results obtained from preoperative patient CT dataset.
318
S. Collin et al. / IRBM 36 (2015) 317–323
Fig. 1. Artificial hearts organization.
1.2. Clinical needs 1.2.1. Artificial hearts Cardiac pumps are composed of two external batteries, an external controller, an internal pump (second generation devices), and a percutaneous driveline that connects the controller to the pump. They are divided into two categories: Total Artificial Heart (TAH) that replaces both ventricles and partial artificial heart more commonly called Ventricle Assist Device, used to support heart function and blood flow. Among VAD, one can differentiate Left Ventricular Assist Devices (LVAD) from Right Ventricular Assist Devices (RVAD) and BiVentricular Assist Devices (Bi-VAD) [6]. An organization of these devices is given in Fig. 1. Our work focuses on two commercially available LVADs implanted at CHU—Rennes (HeartMate II (HM II) – Thoratec Inc., Pleasanton, CA; HeartWare (HW) VAD – HeartWare Inc., Framingham, MA). Their composition includes an inflow cannula placed through the apex of the LV, an outflow cannula anastomosed to the ascending aorta, and the pump with a continuous- or pulsatile-flow rotator, refer to Fig. 2. The appropriate device is chosen by the surgeon considering the patient needs, his metabolism and specificities of each device.
1.2.2. Clinical requirements During the surgical procedure, the pump has to be placed in the attended anatomic location [7]. The proper positioning has to be evaluated for a long-term prevention of infection and damage. Nowadays, surgeons plan it preoperatively by examining patients’ data but some difficulties are still subsisting. They are related to: (i) the positioning of the device. Surgeons have to place the inflow cannula pointing toward the mitral valve. A cannula oriented toward the septum or other free wall facilitates inflow obstruction with potential device malfunction and the creation of thrombus. In the same way, the outflow cannula must be positioned with an orientation that gives the maximum flow into the ascending aorta. (ii) The device collision with anatomical structures. Collisions between the device and specific neighboring organs can cause post-operative complications related to the device functioning or the patient health. On the one hand, during the chest closure a device collision with the thorax can occur and increases the risk of damage, bleeding and malfunctioning, as well as of improper inflow. On the other hand, the RV compression due to the device positioning may facilitate RV dysfunction, which is one relatively common and dreadful complication encountered after a LVAD implantation [8]. Patients in this case may require heart transplantation. (iii) The deformations. Certain parts of the device can be distorted during the implantation. Additionally, the device start-up causes a depletion of the assisted ventricle. These deformations could provoke the displacement of the device, deflecting it from its initial placement. The work presented in this paper is aimed at assessing the collision between the LVAD and two surrounding anatomical structures that are: the chest wall and the RV. 2. Materials and methods This section presents the method proposed for the virtual positioning of LVADs. It involves the devices and patient-specific
Fig. 2. LVAD composition (HeartMate II).
S. Collin et al. / IRBM 36 (2015) 317–323
319
2.1. Data, user interface and interaction
wall and the right ventricle. A collision between the device and these critical organs can increase the risk of damage and death. In the CT images, bone structures that are dense have a high Hounsfield Unit (HU) value and can easily be isolated from other organs. In the proposed solution, the evaluation of the LVAD collision with the thorax is not based on bones segmentation but on the analysis of HU from the CT scan. The anatomical structure of the RV entering in contact with the device is the right ventricular myocardium (muscles). In the neighboring of the heart, there are soft tissues such as liver or abdominal muscles, showing gray values similar to the myocardium ones. To distinguish the myocardium from surrounding tissues we use an approach based on the segmentation of the heart. In this study, the inflow cannula which is inside the LV is not considered.
The first question is related to the description of the patient anatomy. Many segmentation techniques and automatic segmentations of specific structures had been presented in the literature. Their generalization can be complex and can require an important part of interaction with the practitioner. According to the number of anatomical structures required for this study, and because a reconstruction of selected organs will delete much information provided by image data, our approach is to use a 3D slice view (coronal, transversal and sagittal planes) of 3D medical images as representation of the patient and to limit the number of segmented organs for collision detection. For artificial heart implantation, surgeons currently use a CT scan to choose the most appropriate device. Based on the previous analysis of the clinical requirements, the image modality used for this study is a thoracic CT scan, synchronized to the cardiac rhythm in order to minimize artifacts due to heart’s movements. The slice thickness was less than 1 mm and an intravenous injection was used to have a contrast flow in heart ventricles. In order to allow the practitioner to simulate the position of the cardiovascular pump inside the patient, the devices are represented by a 3D mesh and integrated within the 3D slice view of the patient. An ex-vivo rotational imaging of the LVAD (Cone Beam Computed Tomography – CBCT) was used to describe the device. The 3D mesh representing the device was created with the active contour method from ITK-SNAP software (Philadelphia, PA) (Fig. 3) [9]. CamiTK software (Computer Assisted Medical Intervention Toolkit, Grenoble, France) is an open-source modular framework developed in order to be used by clinicians and that can display simultaneously the two kinds of data mentioned previously (3D mesh and CT image). This software was used to implement the virtual positioning application. To allow the user to displace in real-time the device inside the CT scan and to launch the collision detection, a positioning window created with Qt was added. 3D rigid linear transformations were implemented with VTK library.
2.2.1. Right ventricle collision Numerous methods of heart segmentation exist such as Zheng et al. (2008), Ecabert et al. (2008) or Zhuang et al. (2010) [10,11]. However, the reported methods are dedicated to the left atrium, left cavity, left myocardium, right atrium and right cavity. For the right ventricle, myocardium is too slim, epicardium and endocardium are confused. This thinness and the low-contrast in CT images make the segmentation very challenging [12]. The region of the right myocardium entering in contact with the device is located at the apex, where the thickness is regular. Therefore, we decided to describe the RV by its cavities and a margin approximating the myocardium. For the RV segmentation, we chose a method with a simple approach in order to focus on the collision detection. We used ITK-SNAP to perform an interactive segmentation. During preoperative examinations surgeons request an echocardiography that provides information such as myocardium thickness. For more precision, the margin can be fixed by the surgeon according to this exam. The method used for the RV collision detection is represented in Fig. 4. It is based on a distance map implemented by an algorithm based on Danielsson’s Euclidean Distance and developed with the ITK library. Devices are described by a set of points that are characterized by their coordinates x, y and z. To detect parts of the LVAD that are in superimposition with the RV, the process evaluates for each point pi composing the cardiovascular pump, the distance to the right ventricular cavity c via the distance map D(c, pi ). If the point is located inside the region defined by the myocardium thickness parameter, its CT value is checked to confirm that the gray level corresponds actually to soft tissues. Because the heart segmentation is necessary for the assessment of the RV compression, we also use it for the creation of a 3D mesh representing heart chambers. Isosurfaces are extracted with the marching cube module of VTK library and the user can therefore visualize the 3D reconstruction.
2.2. Collision detection
2.2.2. Thorax collision To evaluate parts of the device that are superimposed with the thorax, at each cardiovascular pump’s point, the process evaluates the grayscale value of the CT scan and determines
Fig. 3. 3D mesh of the HeartWare created from a CBCT scan with an active contour method.
anatomy representation, the user interface implementation and the collision detection process.
In this preliminary solution, two surrounding anatomical structures are considered for the collision detection: the thoracic
320
S. Collin et al. / IRBM 36 (2015) 317–323
Several positions of the different devices (HM II and HW) were tested in order to validate the collision detection. Parts of the LVAD in conflict with the thorax are highlighted in red, while those in conflict with the right ventricle are highlighted in blue. The red region of the HM II in Fig. 5b represents the collision with the chest wall, and the blue region in Fig. 5c represents conflict with the RV. Collisions between the selected LVAD and critical neighboring anatomical structures are well detected and represented, with this preliminary solution. 3.2. Execution time and precision of collision detection
Fig. 4. Ventricle collision algorithm.
according to this value, whether the point is in conflict with the chest wall. Solid bones have an associated HU range of values from 200 to 1000. Our approach consists on using a thresholding method with a value chosen higher than 200 to avoid any confusion with soft tissues. However, for a better visualization of heart ventricles and the mitral valve, heart cavities are injected and therefore defined by HU values from 400 to 600. Confusion can therefore be made between bones and heart chambers. In order to differentiate collisions caused by the different organs, we eliminate from this set of points those that are superimposed with the RV. 3. Results Preoperative CT scans used in our experiments come from 3 different adult patients who have previously received a LVAD at the University Hospital of Rennes, France. They were resampled and the number of slices varied from 210 to 400. All images cover the complete heart but not the abdomen. 3.1. Interaction and collision detection A part of the implemented user interface is displayed in Fig. 5. The user can interact in real time with a 3D mesh representing the device, and find a correct position with the 3 slices shown of the CT scan (axial, sagittal and coronal slice). Fig. 5a represents a good positioning with the inflow cannula toward the mitral valve, and without collision.
The number of points composing the pump has an impact on the response time related to collisions detection. Fig. 6 represents the computation time for chest wall collision according to 3D mesh precision of HM II. The higher the number of points composing the pump is, the longer the computation time is. However, results show that the collision detection time is fully compatible with a real-time simulation, even for a 3D mesh composed of an important number of points (300 000). To evaluate the precision of the collision detection, we considered different CT images (voxels precision varying from 0.39 to 1.0), and a 3D mesh with different precisions in the case of HM II device. This study consists on identifying the position of the device where the first collision is detected. The device was placed at an initial point without collision and a translation along the direction of the chest wall was performed every millimeters. Fig. 7 illustrates results obtained for 1 CT scan, but they are similar with the others. When the precision of the 3D mesh describing the device is too small, collision is detected farther, which means that the detection is less precise. Nevertheless, when the 3D mesh is too much detailed, the precision of collision detection is not necessarily improved. Results show that a 3D mesh composed of about 20 000 points, provides a good compromise between the collision detection performance and the computation time. 4. Discussion–conclusion Nowadays, the incidence rate of heart failure is increasing and the development of VADs is prosperous. Although the use of LVAD is in expansion [12], currently there is no tool to assist surgeons to preoperatively plan VADs implantation or help them in the choice of the device. In the context of artificial heart, a work has been recently reported in the Journal of Heart and Lung Transplantation. It proposes a virtual implantation evaluation of the total artificial heart (Syncardia device, TAH-t) and compatibility [13]. This work is intended for the pediatric and small adult populations, in order to improve patient eligibility. Materialise Inc., developed a software suite (MIS—Mimics Innovation Suite) based on CT and MRI data that provide the visualization of patient anatomy. They mentioned a possible application related to the design of patient-specific medical devices such as TAH (AbioCor artificial heart) [14]. In both cases, works are based on segmentation of multiple organs from preoperative data and surface rendering of the TAH. Unlike the implantation of TAH
S. Collin et al. / IRBM 36 (2015) 317–323
321
Fig. 5. Representation of region of conflict with the HeartMate II: a. no region of conflict; b. conflict with the chest wall; c. conflict with the RV. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
322
S. Collin et al. / IRBM 36 (2015) 317–323
Fig. 6. Computation time for chest wall collision according to 3D mesh precision for the HM II.
Fig. 7. Precision for the thoracic collision detection for one CT scan.
where heart ventricles are removed, we focus in our work on LVAD which raises specific problems related to the possible collisions between the device and different kinds of anatomical structures (bones and right ventricle). In our approach, we do not provide just a visualization of the different studied structures but regions in conflict are detected and highlighted in real time. The development of planning tools in the context of deformable and moving anatomical structures is currently a challenge. Nowadays, no application can estimate if the device is able to be inserted in the thoracic cavity or not. The objective of our work was aimed at providing a first solution that allows the user to assess the compatibility between the device congestion and the anatomy of a specific patient. Collisions are detected with a configuration of the heart at a given temporal phase. This could be considered as a limitation of our work. However, image acquisition is made in the diastole phase, where the heart occupies the largest volume in the chest wall. Furthermore, when the device starts-up, it causes a depletion of the assisted ventricle. Therefore, this depletion can, on the one hand, displace the implanted device from its specified position but, on the other hand, it provides more intrathoracic space, which means a greater tolerance. In this first approach, the geometry tolerance has not been considered as critical. In further work, an automatic research of the optimal position of the device could be considered. However, it will be
necessary to analyze and integrate additional decision criteria based on, for instance, the positioning and orientation of the cannula relative to the septum, the influence of the cannula position on the fluid dynamics. These different analyses depend also on the device considered. In this paper we proposed a preliminary solution that provides a personalized simulation allowing the interactive placement of 2 different LVADs. It can be used to detect collisions between the selected device and critical neighboring anatomical structures. This patient-specific approach can be used to interactively simulate the positioning of a LVAD, in order to preoperatively help surgeons to select the most appropriate device. Acknowledgements This work was partially supported by the French National Research Agency (ANR) in the framework of the Investissement d’Avenir Program through Labex CAMI (ANR-11LABX-0004-01). References [1] Heidenreivh PA, Trogdon JG, Khavjou OA, Butler J, Dracup K, Ezekowitz MD, et al. Forecasting the future of cardiovascular disease in
S. Collin et al. / IRBM 36 (2015) 317–323
[2] [3]
[4]
[5]
[6]
[7] [8]
the United States: a policy statement from the American Heart Association. Circulation 2011;123(8):933–44. http://www.agence-biomedecine.fr (Bilan 2011). Flécher E. Assistance circulatoire mécanique de longue durée en 2014: une alternative à la transplantation cardiaque ? Presse Méd July 2014;43(7–8):809–12. Slaughter MS, Pagani FD, Rogers JG, Miller LW, Sun B, Russell SD, et al. Clinical management of continuous-flow left ventricular assist devices in advanced heart failure. J Heart Lung Transplant April 2010;29(4):S1–39. http://dx.doi.org/10.1016/j.healun.2010.01.011. Carrel T, Englberger L, Martinelli MV, Takala J, Boesch C, Sigurdadottir V, et al. Continuous flow left ventricular assist devices: a valid option for heart failure patients. Swiss Med Wkly October 18, 2012;142. http://dx.doi.org/10.4414/smw.2012.13701. Flecher E, Joudinaud T, Grinda J-M. Histoire de L’assistance Circulatoire Mécanique et Du Coeur Artificiel. Annales de Chirurgie, vol. 131. Elsevier; 2006. p. 473–8. HeartMate II LVAD: Instructions for use. Pleasanton, CA: Thoratec Corporation; 2008. Garcia-Alvarez A, Fernandez-Friera L, Lau JF, Sawit ST, Mirelis JG, Castillo JG, et al. Evaluation of right ventricular function and postoperative findings using cardiac computed tomography in patients with left ventricular assist devices. J Heart Lung Transplant May 2011;30(8):896–903. http://dx.doi.org/10.1016/j.healun.2011.03.009.
323
[9] Yushkevich PA, Piven J, Cody Hazlett H, Gimpel Smith R, Ho S, Gee JC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. NeuroImage 2006 Jul.;31(1):1116–28. [10] Peters J, Ecabert O, Meyer C, Kneser R, Weese J. Optimizing boundary detection via simulated search with applications to multimodal heart segmentation. Med Image Anal February 2010;14(1):70–84. http://dx.doi.org/10.1016/j.media.2009.10.004. [11] Hoogendoorn C, Duchateau N, Sanchez-Quintana D, Whitmarsh T, Sukno FM, De Craene M, et al. A high-resolution atlas and statistical model of the human heart from multislice CT. IEEE Trans Med Imaging January 2013;32(1):28–44. http://dx.doi.org/10.1109/TMI.2012.2230015. [12] Anselmi A, Collin S, Verhoye JP, Haigron P, Flecher H. Pilot study for preoperative simulation of patient-device interactions in left ventricular assist device implantation. In: Plenary Session at the ‘Society for Cardiothoracic Surgery (SCTS)/Association of Cardiothoracic Anaesthetists (ACTA)’ Annual meeting, Manchester, 2015 March. [13] Moore RA, et al. Virtual implantation evaluation of the total artificial heart and compatibility: beyond standard fit criteria. J Heart Lung Transplant November 2014;33(11):1180–3. [14] Wivell C. Mimics revolutionizes life-saving device and therapy development. In: D.E. Desktop Engineering, editors. 2009. http://www. deskeng.com/de/mimics-revolutionizes-life-saving-device-and-therapydevelopment/.