3D reconstruction for percutaneous interventions
Citation for published version (APA):
Schoonenberg, G. A. F. (2010). 3D reconstruction for percutaneous interventions. Technische Universiteit Eindhoven. https://doi.org/10.6100/IR684771
DOI:
10.6100/IR684771
Document status and date: Published: 01/01/2010 Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
3D Reconstruction for Percutaneous Interventions
Proefschrift
ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven, op gezag van de
rector magnificus, prof.dr.ir. C.J. van Duijn, voor een commissie aangewezen door het College voor
Promoties in het openbaar te verdedigen op woensdag 8 september 2010 om 16.00 uur
door
Gert Antonius Franciscus Schoonenberg
prof.dr.ir. B.M. ter Haar Romeny Copromotor:
To my mother Toos Schoonenberg and to the memory of my father Gerard Schoonenberg
prof.dr. J.D. Carroll University of Colorado, Aurora, CO, USA.
prof.dr.ir. M. Breeuwer Technische Universiteit Eindhoven, The Netherlands. prof.dr.ir. M.A. Viergever Universitair Medisch Centrum Utrecht, The
Netherlands.
pd.dr.rer.nat.habil. M. Grass Technical University of Hamburg-Harburg, Germany.
CIP-DATA LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN Gert Schoonenberg
3D Reconstruction for Percutaneous Interventions/ by Gert Schoonenberg. - Eind-hoven : Technische Universiteit EindEind-hoven, 2010.
A catalogue record is available from the Eindhoven University of Technology Library ISBN: 978-90-386-2298-9
NUR 954
Trefw.: beeld gestuurde interventies / minimaal invasieve behandeling / medische beeldverwerking / bewegingsgecorrigeerde reconstructie.
Subject headings: image guided interventions and therapy / minimal invasive treat-ment / medical image processing / motion compensated reconstruction.
Copyright © 2010 by G. Schoonenberg
All rights reserved. No part of this material may be reproduced or transmitted in any form or by any means, electronic, mechanical, including photocopying, recording or by any information storage and retrieval system, without the prior permission of the copyright owner.
Financial support by the Netherlands Heart Foundation for the publication of this thesis is gratefully acknowledged.
Support by the Technische Universiteit Eindhoven and Philips Healthcare for this research and the publication of this thesis is gratefully acknowledged.
L
IST OF ACRONYMS
2D two-dimensional 3D three-dimensional 4D four-dimensional
ART algebraic reconstruction technique CA contrast agent
CABG coronary artery bypass grafting CAD coronary artery disease
CT computed tomography CVD cardiovascular disease ECG electrocardiogram
FDK reconstruction method developed by Feldkamp, David, and Kress FN false negative
FP false positive IQ image quality
IVUS intravascular ultrasound LAD left-anterior descending LAO left anterior oblique
OCT optical coherence tomography PCI percutaneous coronary intervention
PTCA percutaneous transluminal coronary angioplasty QCA quantitative coronary analysis
RA rotational angiography RAO right anterior oblique ROI region of interest SA standard angiography SIC superior-inferior component
START simultaneous thresholding algebraic reconstruction technique TN true negative
List of acronyms i
Contents ii
1 General introduction 1
1.1 Medical background . . . 2
1.1.1 Coronary artery disease . . . 2
1.1.2 Diagnosis of coronary artery disease . . . 4
1.1.3 History of cardiac catheterization and coronary artery angioplasty . . . 4
1.1.4 History of coronary stenting . . . 5
1.1.5 Percutaneous coronary interventions . . . 5
1.1.6 Clinical needs . . . 6
1.1.7 Technical developments . . . 8
1.2 Approach and goals . . . 9
1.3 Outline . . . 9
2 Advanced visibility enhancement for stents and other devices: image processing aspects 11 2.1 Introduction . . . 12
2.2 Stent-visibility enhancement . . . 12
2.2.1 Processing steps . . . 13
2.2.2 Accuracy and clinical use . . . 14
2.3 StentBoost subtract . . . 17
2.4 Stent enhancement using rotational acquisition . . . 19
2.5 Three-dimensional stent reconstruction . . . 20
2.5.1 Clinical workflow and data processing . . . 21
2.5.2 Visualization tool . . . 25
2.6 Results of enhancement of other cardiovascular devices . . . . 25
2.7 Summary . . . 27
3 Projection-based motion compensation and reconstruction of coronary segments and cardiac implantable devices using ro-tational X-ray angiography 29 3.1 Abstract . . . 30
3.2 Introduction . . . 30
3.2.1 2D enhancement . . . 31
Contents
3.3 Method . . . 32
3.3.1 Data acquisition . . . 32
3.3.2 Marker couple detection . . . 33
3.3.3 Motion compensation and reconstruction . . . 34
3.3.4 Enhanced cross-sectional viewing: virtual IVUS . . . 34
3.4 Validation . . . 36
3.4.1 Marker detection . . . 36
3.4.2 X-ray angiography versus IVUS . . . 36
3.4.3 Measurements in reconstructed data sets . . . 37
3.5 Results . . . 37
3.5.1 Marker detection results . . . 37
3.5.2 Processing time . . . 38
3.5.3 Reconstructed volumes . . . 39
3.5.4 Virtual pullback visualization . . . 41
3.5.5 Validation with intravascular ultrasound . . . 41
3.5.6 Measurements in reconstructed data sets . . . 43
3.6 Discussion and conclusion . . . 44
4 3D motion compensation and reconstruction of rigid implantable devices using rotational angiography 49 4.1 Abstract . . . 50
4.2 Introduction . . . 50
4.2.1 Previous work on 2D motion compensation . . . 52
4.2.2 Previous work on virtual motion compensated acquisi-tion geometries and reconstrucacquisi-tion methods . . . 53
4.3 Method . . . 55
4.3.1 Data acquisition . . . 55
4.3.2 Feature point identification . . . 55
4.3.3 3D reference model generation . . . 55
4.3.4 3D device localization from a single projection . . . 56
4.3.5 Virtual geometry . . . 56 4.3.6 Reconstruction . . . 58 4.3.7 Phantom data . . . 59 4.3.8 In-vivo data . . . 60 4.4 Results . . . 61 4.4.1 Phantom results . . . 61 4.4.2 In-vivo results . . . 61
4.5 Conclusion and discussion . . . 61
5 Device reconstruction: various clinical cases 69 5.1 Introduction . . . 70
5.2 Clinical cases . . . 70
5.2.2 Evalve MitraClip . . . 71
5.2.3 Fractured coronary stent . . . 72
5.2.4 3D stent and vessel reconstruction . . . 72
5.3 Discussion . . . 73
6 Three-dimensional coronary visualization: 3D reconstruction 75 6.1 Introduction . . . 76
6.2 Data acquisition . . . 77
6.3 Cardiac gating and phase selection strategies . . . 79
6.3.1 Optimal cardiac phase determination . . . 80
6.3.2 Optimal gating window . . . 80
6.4 Three-dimensional coronary reconstruction methods . . . 80
6.4.1 Gated reconstruction methods . . . 82
6.4.2 Motion compensated reconstruction methods . . . 84
6.5 Clinical tools for 3D coronary reconstruction . . . 85
6.6 Clinical cases . . . 88
6.7 Conclusion and discussion . . . 90
7 Left coronary artery thrombus characterized by a fully auto-matic three-dimensional gated reconstruction 93 7.1 Introduction . . . 94
7.2 Case report . . . 94
7.3 Case discussion . . . 98
8 Reconstruction of extended XperSwing acquisitions 101 8.1 Introduction . . . 102
8.2 XperSwing acquisitions . . . 103
8.3 Extended XperSwing acquisitions . . . 103
8.4 Results . . . 104
8.4.1 Phantom data . . . 104
8.5 Discussion . . . 105
9 Discussion and outlook 107 9.1 3D device reconstruction . . . 108
9.1.1 Motion compensation in three dimensions . . . 108
9.1.2 Meaningful visualization of stents . . . 109
9.1.3 Clinical benefit . . . 109
9.2 Coronary reconstruction . . . 110
9.2.1 Optimal acquisition trajectory . . . 110
9.2.2 Meaningful visualization of coronaries . . . 111
9.2.3 Clinical benefit . . . 111
9.3 3D reconstruction for percutaneous interventions . . . 111
Contents
A Appendix 113
A.1 Supplementary material . . . 113
A.1.1 Chapter 2 . . . 113
A.1.2 Chapter 7 . . . 113
Bibliography 115 Summary 127 3D Reconstruction for Percutaneous Interventions . . . 127
Background . . . 127
Stent reconstruction and visualization . . . 127
Coronary reconstruction and visualization . . . 128
Discussion en conclusion . . . 129
Samenvatting 131 3D Reconstructie voor Percutane Interventies . . . 131
Achtergrond . . . 131
Stent reconstructie en visualisatie . . . 131
Kransslagader reconstructie en visualisatie . . . 132
Discussie en conclusie . . . 133
List of publications 135
Acknowledgements 139
C
H
A
P
T
E
R
1
G
ENERAL INTRODUCTION
“All human beings by nature desire to know. A sign of this is our liking for the senses; for even apart from their usefulness we like them for themselves - especially the sense of sight, since we choose seeing above practically all the others, not only as an aid to action, but also when we have no intention of acting. The reason is that sight, more than any of the other senses, gives us the knowledge of things and clarifies many differences among them.”
Aristotle.
In our daily life we use our senses all the time. Following Aristotle we choose sight above practically all the others. However our sight, our eyes, do not allow us to look inside objects such as the human body. This thesis focusses on interventional cardiologists and giving them ’sight’ into arteries of the heart, inside the human body.
1.1
Medical background
The main causes of death in the industrialized world are cardiovascular dis-ease (CVD) and cancer. CVD and cancer accounted for respectively 34.3% and 23.1% of all deaths in the United States in 2006 [1]. Coronary artery disease (CAD), a subgroup of CVD, caused 17.5% of all deaths in the US in the same year. A total of 652.000 percutaneous coronary interventions with stent placement were performed that year to treat CAD and the estimated direct costs associated with CAD for the US in 2010 are 96 million US dol-lars. The estimated direct costs for all cancer and benign neoplasms in 2008 are 93 million US dollars. CVD costs more than any other diagnostic group. Cancer and CVD are also the main two causes of death in the Netherlands [2], see Figure 1.1 for an overview of the last decades. Since the late 1990s the number of deaths by CVD declined by approximately 10.000 deaths. This is mainly caused by the reduction in deaths by coronary artery disease, see Figure 1.2.
1.1.1
Coronary artery disease
Atherosclerosis is the most common cause for CAD. Large quantities of cholesterol gradually become deposited beneath the endothelium at many
1.1. Medical background 10000 20000 30000 40000 50000 60000 N u m b e r o f d e a t h s
Main causes of death in the Netherlands
Cancer Cardiovascular disease 0 10000 20000 30000 40000 50000 60000 1 9 6 9 1 9 7 2 1 9 7 5 1 9 7 8 1 9 8 1 1 9 8 4 1 9 8 7 1 9 9 0 1 9 9 3 1 9 9 6 1 9 9 9 2 0 0 2 2 0 0 5 2 0 0 8 N u m b e r o f d e a t h s Year
Main causes of death in the Netherlands
Cancer
Cardiovascular disease
Figure 1.1: Overview of the number of deaths per year caused by the two
main causes of death in the Netherlands for the last 40 years [2]. Cancer is causing increasingly more deaths in the Netherlands, whereas cardiovascu-lar disease deaths have been declining the last decade.
0 5000 10000 15000 20000 25000 30000 N u m b e r o f d e a th s
Deaths in the Netherlands
Coronary artery disease
0 5000 10000 15000 20000 25000 30000 1 9 6 9 1 9 7 2 1 9 7 5 1 9 7 8 1 9 8 1 1 9 8 4 1 9 8 7 1 9 9 0 1 9 9 3 1 9 9 6 1 9 9 9 2 0 0 2 2 0 0 5 2 0 0 8 N u m b e r o f d e a th s Year
Deaths in the Netherlands
Coronary artery disease
Figure 1.2: Overview of the number of deaths per year caused by coronary
artery disease, a subgroup of cardiovascular disease, in the Netherlands for the last 40 years [2]. Notice the decline that started in the mid 1980s.
locations in arteries, due to genetic predisposition to atherosclerosis, due to eating excessive quantities of cholesterol and other fatty substances, or due to other risk factors. Gradually, these areas of deposit (plaques) are invaded by fibrous tissue and frequently become calcified. This process causes the atherosclerotic plaques to protrude into the vessel lumen and can potentially block or partially block blood flow through these arteries. These atheroscle-rotic plaques are common in the coronary arteries. An acute occlusion of a coronary artery can occur when the atherosclerotic plaque causes a clot (thrombus), which occludes the artery. Immediately after, blood flow ceases in the coronary vessels beyond the occlusion, except for small amounts of collateral flow from surrounding vessels. Irreversible myocardial cell damage and cell death will occur after a certain period in the area of muscle that has either zero flow or so little flow that it cannot sustain cardiac muscle function. This area is said to be infarcted. The overall process is called myocardial infarction. The main causes of death after acute myocardial infarction are decreased cardiac output, damming of blood in the pulmonary veins and pul-monary edema, fibrillation of the heart and rupture of the heart [3]. Coronary artery bypass grafting (CABG) and coronary artery angioplasty are surgical treatments of coronary artery disease. Currently most patients with coronary artery disease are treated with angioplasty [1].
1.1.2
Diagnosis of coronary artery disease
Coronary artery blockages or stenoses can cause insufficient blood flow to the cardiac muscle. A person who suffers from this can feel a certain cardiac pain, called angina pectoris. This pain typically occurs when people exercise; the load on the heart becomes too great in relation to the coronary blood flow. The electrocardiogram (recording of electrical potentials of the heart) can reveal this [3]. An exclusive diagnosis is often made after a cardiac catheterization. During such a procedure a contrast medium is selectively injected in the coronary arteries to identify blockages or stenoses of the left and right coronary artery trees.
1.1.3
History of cardiac catheterization and coronary artery
angioplasty
The history of cardiac catheterization, as described by Ryan [4], started more than five decades ago, when F. Mason Sones, Jr, on October 30, 1958 per-formed a selective injection of contrast media into a right coronary artery and created the first coronary angiogram. In 1967 he imaged the first angiogram of an aorto-coronary vein graft, which was inserted by R. Favaloro. Coro-nary artery bypass grafting became the most common surgical procedure performed in the US in the early 1970s. In 1977 A. Gruntzig et al. described
1.1. Medical background
a new method of achieving coronary revascularization by the endovascular dilation of an obstructing lesion and the method was referred to as percuta-neous transluminal coronary angioplasty (PTCA). This technique has been refined and is still used on a daily basis worldwide to treat coronary artery disease.
1.1.4
History of coronary stenting
During his Nobel lecture [5] in 1912, laureate Alexis Carrel described experi-ments with glass and metal tubes covered with paraffin that were introduced into the thoracic aorta of a dog. The blood did not coagulate (complex pro-cess by which blood forms clots) as long as the wall of the artery was not ulcerated. A dog whose thoracic aorta contained a piece of glass remained in perfect health for 90 days. These experiments showed that the presence of foreign bodies in the lumen of a vessel does not necessarily produce a thrombus, as was formerly believed.
Fifty years later research continued on these implantable foreign bodies, which we call stents nowadays. In 1964, Charles Dotter described the de-velopment of an expandable device suitable for percutaneous insertion [6]. In 1983, Charles Dotter et al. [7] and Andrew Cragg et al. [8] described expandable nitinol coil stents for the non-surgical treatment of several forms of vascular disease. More people started working on this topic and in 1985 various new stent types were introduced: Kenneth Wright et al. described a stainless steel wire bent in a zig-zag pattern [9] and Julio Palmaz et al. de-scribed a balloon-expandable stent system with continuous, woven, stainless steel wire as stent [10]. Not much later, the first coronary-artery stents were implanted in 19 patients by Jacques Puel and Ulrich Sigwart [11]. This latter study concluded that vascular endoprosthesis may offer a useful way to pre-vent occlusion and restenosis after transluminal angioplasty. Many studies followed and the balloon-expandable stent system was further refined. Fig-ure 1.3 shows a modern balloon-expandable stent system. By 1999, stent-ing comprised 84.2% of percutaneous coronary interventions [12]. In the late 1990s active stent coatings were developed to inhibit in-stent restenosis. These type of stents are still used on a daily basis and are called drug-eluting stents.
1.1.5
Percutaneous coronary interventions
Stenting during a percutaneous coronary intervention (PCI) is schematically shown in Figure 1.4. The groin is a typical location to get access to the vas-culature for a PCI. Under X-ray guidance a catheter is advanced to the ostia of the coronaries. Contrast media can be selectively injected to create an angiogram of either the right or left coronary artery. When a stenotic region
Figure 1.3: Boston Scientific Taxus Express22.75 x 28 mm stent on a guide wire with balloon and markers. The arrows indicate the markers on the wire. The already expanded stent is located between the two markers.
is identified to be treated, a balloon-tipped catheter is advanced and pushed through the partially blocked artery. The balloon is then inflated with high pressure to stretch the diseased artery. Afterwards a balloon-stent system is advanced to the location of the lesion, see Fig 1.4A and expanded (Fig-ure 1.4B). The balloon and wire are retracted and the stent remains in the widened artery, see Figure 1.4C.
1.1.6
Clinical needs
During coronary stenting procedures multiple decisions have to made by the interventional cardiologist. Assuming that a stenotic lesion needs to be treated and a stent needs to be placed, the following choices need to be made: where exactly does the stent need to be deployed; which stent is needed; and is the stent adequately deployed or does it need additional balloon-inflation? Each of these questions will be discussed next.
1.1.6.1 Stent positioning
The location of a severe coronary artery narrowing needs to be stented, how-ever there are multiple aspects to be taken into account when positioning a stent. Recent publications have shown that stents can fracture at locations that serve as hinges (bending points) during vessel movement in the car-diac contraction cycle. During a follow-up study of 256 patients with a total of 307 lesions, stent fractures were observed in eight (2.6%) lesions [14]. The stent fractures were all in locations that served as hinges during vessel movement. Seven of the eight fractured stents were adjacent to the edge of a previously implanted or overlapping stent. Another study concludes that
1.1. Medical background
Figure 1.4: Schematic overview of stenting during a percutaneous coronary
intervention. The stent is delivered to a stenotic artery caused by atheroscle-rotic plaque [13].
the exact mechanism for stent fracture is unknown, but that, among others, vessel tortuosity and overlapping stents play a negative role in stent fracture [15]. They also state that the increased incidence of stent fracture of over-lapping stented segments may be due to creation of rigid hinge points. So not only the shape of the lesion is important for stent positioning but also the dynamics of the coronaries.
1.1.6.2 Stent selection
Coronary angiograms are evaluated in order to select the right stent. This is often done by just looking at an angiogram and estimating the right dimen-sions (also called eyeballing). A quantitative coronary analysis (QCA) can be performed to get quantitative information on the vessels such as length and diameter information. Various QCA packages are available that calculate two-dimensional (2D) or three-dimensional (3D) measurements after some sort of user interaction [16]. If 2D QCA packages are used it is important to take vessel foreshortening into account. Vessel foreshortening occurs when a vessel is not viewed from a direction perpendicular to the vessel center-line. Vessel foreshortening in angiographic images may cause errors in the assessment of lesions or the selection of stents. Angiographic views used during stent deployment in the circumflex artery had the most foreshortening (mean, 14.1%±11.7%; range, 0-50%) in a study of 156 vessel segments [17]. If a view with 14% foreshortening is used to select the stent this might result in a stent that is 14% too short. It is possible that a second stent is needed to treat the lesion, which increases to the overal risk and cost of the procedure.
1.1.6.3 Stent deployment
Inadequate deployment of stents occurs due to various reasons: poor appo-sition of the stent struts against the arterial wall; an asymmetric deployment so that the stent lumen is not circular, which might predispose to turbulent flow; and underexpansion of stent struts [18]. Intravascular ultrasound (IVUS) can be used to visualize these inadequate deployment of stents, after which the stent can be expanded further with higher pressure or a larger balloon. This typically reduces restenosis of the stent. If IVUS is not used, angiograms can be used to verify stent deployment.
1.1.7
Technical developments
Various solutions have been made commercially available the recent years to provide the interventional cardiologist with solutions for stent positioning, stent selection and stent deployment. 3D modeling techniques combined
1.2. Approach and goals
with optimal view maps and 3D QCA provide a way to overcome the fore-shortening problem and calculate accurate 3D measurements for adequate stent selection and positioning [19, 20]. However these solutions typically require a significant amount of user interaction and therefore these solutions don’t fit well into the clinical workflow of a PCI procedure. Enhancement of deployed stents in angiograms can be achieved by motion compensation and temporal integration of image sequences [21]. This tool fits very well in the workflow of a PCI procedure, because the processing is done without user interaction. However, these enhanced angiograms show the deployed stent from one viewing direction and therefore do not reveal the 3D shape of the stent. A second acquisition is needed to image the stent from another view-ing direction to investigate underdeployment of the stent in the other viewview-ing direction. IVUS remains a useful imaging modality to acquire 3D shape in-formation of the coronary arteries, plaque characterization (with or without virtual histology tools) and to investigate stent deployment. However this so-lution is costly and adds time to the procedure [22].
1.2
Approach and goals
The workflow and clinical setting enforce requirements on the technical solu-tions for the clinical needs. User interaction with for example a workstation is cumbersome, because the interventional cardiologist works in a sterile envi-ronment and needs to be at the patient’s side. Even if a nurse would perform the necessary manual steps it probably would still lengthen the procedure time. The technical solution should be with very limited user interaction or should be fully automated. A meaningful 3D representation of the coronary arteries and deployed coronary stents should be visualized automatically to the interventionalist after an X-ray acquisition. This will greatly assist in stent positioning, stent selection and stent deployment. This could potentially lead to better patient outcome and reduced cost.
1.3
Outline
This thesis consists of two main parts. The first part deals with 3D recon-struction of coronary stents and other cardiovascular devices. It starts with an introduction on device enhancement for these devices (Chapter 2). It de-scribes the research performed to enhance devices in fixed view image se-quences and rotational image sese-quences by motion compensation and tem-poral integration. This chapter also describes the principles of 3D cardiovas-cular device reconstruction. Chapter 3 describes the feasibility of automatic marker detection in rotational acquisitions to enable automated 3D device reconstruction. This chapter shows various reconstructed stents, calcified
plaque and other devices using a 2D motion compensation method. This chapter also describes the virtual pullback, which is a new IVUS-like visual-ization method for X-ray based stent reconstructions to create an automated and meaningful visualization of a stent. Chapter 4 proposes a new method to compensate motion in 3D instead of in 2D. This method overcomes arti-facts of the previously described 2D motion compensation method caused by rotation of a device. Chapter 5 concludes the first part with some clinical cases of various devices, such as a vascular stent with calcified plaque and two MitraClips.
The second part, chapters 6 through 8, deals with 3D reconstruction of coronary arteries. This part starts with an introduction on 3D coronary re-construction (Chapter 6). In chapter 7 a clinical case is described where 3D reconstruction was able to visualize an important and potentially dangerous thrombus. Visualizing this to the interventional cardiologist in a meaningful way resulted in choosing not to place a stent, but to select another treatment based on medication. The second part concludes with chapter 8 on new dual-axis acquisitions as an ideal trajectory for both diagnostic imaging as well as high quality 3D reconstruction. Finally in chapter 9 we conclude both parts with a discussion and look into the future.
C
H
A
P
T
E
R
2
A
DVANCED VISIBILITY ENHANCEMENT FOR
STENTS AND OTHER DEVICES
:
IMAGE
PROCESSING ASPECTS
“Simplicity is the ultimate sophistication.”
Leonardo da Vinci.
2.1
Introduction
Cardiac devices, such as stents, are being implanted daily worldwide. To en-sure the proper outcome of these procedures, clinicians must be able to see how these devices are deployed. Reports have noted inadequate deploy-ment of stents, leading to incomplete apposition to the vessel wall [23, 24]. Late thrombosis and in-stent restenosis can also occur. It is therefore impor-tant that the deployment of devices can be analyzed at the time of deploy-ment and immediately improved if necessary. Visualization of these deployed stents is especially important now, in the era of the newer drug-eluting stents, because they are less radiopaque than bare metal stents and therefore more difficult to see with conventional X-ray coronary angiography. To evaluate full stent deployment, interventionalists sometimes make use of intravascular ultrasound (IVUS), although this practice remains relatively limited in most countries. An interesting alternative consists in relying on enhanced stent visualization tools for X-ray sequences, such as StentBoost (Philips Health-care, Best, The Netherlands), IC Stent (Siemens HealthHealth-care, Erlangen, Ger-many), or StentOptimizer (Paieon Inc., New York, New York). These latter tools enable the visualization and analysis of a deployed stent in one pro-jection direction. Three-dimensional (3D) stent reconstruction has been de-veloped to allow enhanced stent visualization and measurements in three instead of two dimensions. All those visibility-enhancement techniques us-ing X-ray based devices rely on unique features in the acquired images to allow for motion compensation of the object of interest, which, together with temporal integration or reconstruction, forms the basis of device visualiza-tion enhancement. This chapter gives an overview of the different methods used for enhanced stent visualization, describes the studies evaluating these methods, and summarizes results of these methods on other cardiac and noncardiac devices.
2.2
Stent-visibility enhancement
The general method for improving visibility of stents during percutaneous coronary intervention involves the creation of an enhanced exposure image that clearly outlines the geometry of the deployed stent from a fixed viewing direction. The enhancement is achieved by employing a motion
compen-2.2. Stent-visibility enhancement
sation process and temporal integration of the image sequence to increase contrast visibility.
2.2.1
Processing steps
The motion compensation process identifies the stent region in each projec-tion image to register the series of sequential images. This can be done in various ways. One of the common methods reported in the literature is the landmark-based technique to identify the stent region [25, 26, 27, 28, 29, 30, 31]. Those methods typically use balloon markers for identification of the stent region. An extension to this approach uses, in addition to the balloon markers, the wire between the balloon markers [32]. Of course, approaches based on markers and wires can only be effective if the balloon delivery sys-tem remains in place immediately after stent deployment. Also, one could argue that, once deployed, the stent might move somewhat independently of the deployment setup, which makes balloon catheter-based registration suboptimal. To circumvent those issues, one approach based on deformable boundary detection [33] has been proposed to localize the stent by detection of the outer contour of the stent in the projection images. Another technique based on a layer decomposition algorithm has been proposed. With this technique, the series of cine images is decomposed into different moving layers [34]. As a result of this decomposition, the stent is shown in one layer while background structures are contained in other layers. These methods all require, as a first step, the application of a segmentation process for identifi-cation of landmarks (e.g., balloon markers, the wire, or the stent contour) to accurately localize the stent region. In the next step, these landmarks need to be registered throughout the image sequence to compensate for com-bined cardiac and respiratory motion. When only the markers are used as landmarks, an affine transformation for each frame, consisting of translation, rotation, and stretching in two dimensions, suffices for the registration pro-cess, see Figure 2.1. If the curved wire or the contour of the stent is used as the landmark, an elastic transformation will be needed to take into account nonlinear movement (i.e., stent bending) of the object throughout the cardiac cycle and allow for more accurate registration. In the remainder of this chap-ter, the affine transformation for two-dimensional (2D) motion compensation is assumed. However, more advanced transformations can be used as well.
With the completion of the registration process, the transformed image is displayed and clinicians may adjust contrast and brightness to obtain the best view of the stent. Additionally, other typical image processing techniques can be applied, such as zoom-in on region of interest (ROI), adaptation of the gray-level dynamics to the selected stent-centric ROI, histogram manipula-tion in selective regions, and edge enhancement. As a result, the spatial processing methods in conjunction with the temporal integration process
en-Figure 2.1: Registration of two frames from the exposure sequence (Frame
1 and Frame 2) using an affine transformation consisting of translation, rota-tion, and stretching.
able a much clearer and more useful view of the stent. The detailed steps in the process of enhancing the image of the stent are summarized in Figure 2.2. Various cases are shown in Figure 2.3 to demonstrate the effective-ness and usefuleffective-ness of the process for enhanced stent visualization based on StentBoost software.
2.2.2
Accuracy and clinical use
Although stent visibility has been greatly improved in the enhanced images, the process of obtaining accurate and quantitative estimates of the stent’s geometry after it has been implanted (eg, 3D length and diameters) is tricky. This is because the projection image is not useful for recovering the 3D mor-phology and the relative position (i.e., the location and orientation) of an object (e.g., the implanted stent) with respect to the imaging system. Ad-ditionally, because of the characteristics of cone-beam imaging projection, an object farther away from the detector will appear bigger in the projection image than an object near the detector. Therefore a calibration process is still needed to perform any traditional 2D quantitative measurements based on the enhanced images. Calibration is a burden to the operator because it requires manual selection of a calibration method and, depending on the method, additional user interaction. Calibration is not needed in the workflow of generating the enhanced stent images, but it is necessary to validate the accuracy of this technique. The pixel size of the enhanced images can be calibrated in different ways, namely catheter calibration, marker calibration, and automatic pixel-based calibration. The automatic pixel-based calibration assumes that the stent is in the isocenter of the C-arm system, where the calibration factor (mm/pixel) is known, because the distance from the
gen-2.2. Stent-visibility enhancement
Figure 2.2: Processing steps for 2D enhancement of stents in X-ray
expo-sure sequences: acquiring projection images; setting an optional region of interest (ROI); marker couple detection based on marker detection; boost-ing of the image by registerboost-ing all frames usboost-ing an affine transformation and temporal integration of these registered images; and finally displaying the end result to the user with optimal visualization settings.
erator to the isocenter and the distance from the isocenter to the detector are known. With catheter and marker calibration, a catheter or marker needs to be segmented and their dimensions need to be specified to calculate the calibration factor. To get an accurate calibration, the catheter must be in the same plane as the stent (in a plane parallel to the detector) and the stent must not be foreshortened. The accuracy can fluctuate depending on the calibration method, segmentation accuracy, and operator skills. Measure-ment errors can occur when the stent is foreshortened, the stent is not in the isocenter of the system, and the catheter used for calibration is not in the same plane. Various studies have examined the clinical usefulness of this en-hancement technique. The stent visibility of those enhanced images is better
Figure 2.3: Various StentBoost cases. (A) An image from an exposure
se-quence with a stent and (B) the corresponding enhanced stent image. (C) An image from an exposure sequence with a longer and more curved stent and (D) the corresponding enhanced stent image. (E-H) Various images from a stent deployment where calcifications were present. (E) Image from an exposure sequence with a deployed 3.5 x 20-mm Taxus stent (Boston Scientific, Maple Grove, Minnesota) dilated at 14 atm. (F) An enhanced stent image showing a calcification (white arrow) and stent underdeployment (black arrow). (G) Positioning of a noncompliant 10/ 4.0-mm Extensor bal-loon (Medtronic, Santa Rosa, California) dilated at 22 atm. (H) Result shows improved stent expansion.
2.3. StentBoost subtract
than that of the standard images, as demonstrated in a study of 27 consecu-tive patients and 35 stents [25, 35]. On a scale from 1 (not visible) to 5 (crystal clear), the conventional images scored 1.7 ±0.6 and the enhanced images scored 3.5±0.8. Quantitative stent measurements from the enhanced im-ages and IVUS were compared in three studies [25, 36, 37]. One study also made a comparison with quantitative coronary analysis (Table 2.1). Various measurements where compared between StentBoost and IVUS: minimum diameter, mean diameter, and area ratios for IVUS with diameter ratios for 2D StentBoost. All studies show a medium to large correlation between the two modalities. Besides accuracy and validation of this technique, the clini-cal use of stent visibility enhancement tools has been reported for bifurcation stenting [38].
2.3
StentBoost subtract
An extension to the aforementioned enhanced angiograms is a tool called StentBoost Subtract (Philips Healthcare, Best, The Netherlands). By show-ing the stent in relation to the vessel wall, this tool, in addition to enhancshow-ing stent visibility, enables operators to assess the apposition of the stent to the vessel wall. This is achieved by modifying the protocol as originally used for StentBoost image acquisition. Contrast agent is injected halfway during the acquisition so that the coronary segment of interest is opacified and included in the subsequently acquired image sequence. The frames with the same cardiac phase (e.g., end diastole) in the images of coronary arterial segment and the enhanced StentBoost images are identified so that they can be com-bined. Various rendering approaches can be used to examine these two images. The simplest approach is to display the images side by side. How-ever, with this approach, interpreting the relation between the stent and the vessel wall is still difficult. Therefore, dynamic approaches were evaluated that combine the two images in movie loops. Figure 2.4 shows four differ-ent approaches of these animations. Figure 2.4A shows a curtain sliding up and down. Figure 2.4B shows blinds sliding to the left. Figure 2.4C shows a fade-in and fade-out of the enhanced stent image on the angiogram im-age. Figure 2.4D shows results in the similar way with a gray-scale inverted enhanced stent image. This latter visualization method became the default visualization method for the Philips StentBoost Subtract tool. No other sup-pliers seem to offer this functionality. Clinical use of this tool has typically been reported in bifurcation stenting cases [38, 39, 40] and stent fracture cases [41, 42]. An example of StentBoost Subtract in bifurcation stenting is illustrated in five frames from the fade-in and fade-out animation as shown in Figure 2.5.
T ab le 2.1: V alidation studies of StentBoost. Study Hanekanp , K oolen, et al. [25, 35] Conw a y et al. [36] Conw a y et al. [36] Mishell et al. [37] Mishell et al. [37] Number of patients 27 33 33 30 30 Number of coronar y stents 35 43 43 48 48 Gold standard IVUS IVUS IVUS IVUS QCA Measurement Mean diameter for IVUS and StentBoost Ratio minim um/ mean stent cross-sectional area for IVUS and ra-tio minim um/ mean di-ameter for StentBoost Ratio mini-m um/ maxim um stent crosssectional area for IVUS and ratio minim um/ maxim um diameter for S tent-Boost Minim um diameter for IVUS and StentBoost Minim um diameter for QCA and StentBoost Correlation coefficient method -T w o-tailed Spear man T w o-tailed Spear man P earson product-moment P earson product-moment Correlation coefficient 0.819 ± 0.122 0.44 0.57 0.75 0.49 P v alue <0.05 0.003 <0.001 <0.0001 0.0004
2.4. Stent enhancement using rotational acquisition
Figure 2.4: Overview of different animated visualization modes for displaying
an enhanced stent image with an angiogram. (A) One frame of curtain sliding up and down. (B) One frame of blinds sliding to the left. (C) Four frames of fade-in and fade-out animation. (D) Four frames of fade-in and fade-out animation, with grayscale-inverted enhanced stent image.
2.4
Stent enhancement using rotational acquisition
Stent visibility enhancement is often done using two viewing directions [25]. If the stent is underdeployed in a certain viewing direction, another expo-sure from a viewing direction rotated by approximately 90 degrees around the axis of the stent is needed to correctly visualize this underdeployment. A minimum of two viewing directions is therefore needed to adequately evalu-ate stent deployment. Instead of setting up two views, one can also make a rotational acquisition and enhance this run. An apparent problem is the changing projected shape of the device throughout the run, which makes automatic detection of the stent region and registration more difficult. Au-tomated marker-based segmentation of the stent region is feasible for rota-tional acquisitions [31]. The enhancement of rotarota-tional runs is possible, but due to changing geometry between successive images only a limited num-ber of frames can be integrated for each individual enhanced image [43].
Figure 2.5: StentBoost Subtract (Philips Healthcare, Best, The Netherlands)
in a bifurcation stenting case. (Top to bottom) Five images from a fade-in, fade-out sequence of the enhanced stent image and an angiogram.
Because using more frames in the enhancement phase theoretically results in better images, this limitation on the number of frames could be a shortcom-ing. A rotational acquisition over 90 degrees will obtain similar information as two fixed-view acquisitions at a similar dose. Figure 2.6 shows an enhanced image from a rotational acquisition of a coronary stent.
2.5
Three-dimensional stent reconstruction
3D motion compensated volumetric stent reconstruction has been developed to give insight into the 3D geometry of the deployed stent. These projection images, instead of being used to enhance the image of the stent, are used to reconstruct the stent in a volumetric dataset. In principle, the exposure images can be motion compensated in the same way as previously shown in Figure 2.1 for 2D stent-visualization enhancement. After motion
compen-2.5. Three-dimensional stent reconstruction
Figure 2.6: Application of StentBoost in a rotational run. (Left) Single
unen-hanced image of the Taxus Express2 3.0 x 16-mm stent (Boston Scientific,
Maple Grove, Minnesota) of the rotational acquisition. (Right) Enhanced im-age of this stent placed in the left anterior descending artery.
sation, a standard cone-beam based reconstruction method can be used to generate the volumetric dataset. Phantom and animal studies have indeed shown that stents can be reconstructed this way assuming that the motion of the markers on the balloon catheter is equal to the motion of the stent [28, 44]. Pilot studies of 3D stent reconstruction in humans using an image intensifier system [45, 46] and a flat panel detector [47] have shown the fea-sibility of this technology in a clinical setting. A recent publication [31] shows that the whole process of 3D stent reconstruction can be automated. How-ever, 3D stent reconstruction is clinically and technically more challenging than enhancing 2D projection images. Table 2.2 shows an overview of these enhancement and reconstruction techniques and summarizes some key el-ements. The following section describes in more detail the clinical workflow, the image processing steps, and clinical tools for 3D stent reconstruction.
2.5.1
Clinical workflow and data processing
A rotational acquisition starts with isocentering the device in the C-arm sys-tem to ensure that the device will be visible throughout the acquisition. Typ-ically, patients are instructed to hold their breath during the rotational run to minimize respiratory motion. A calibrated monoplane C-arm system will then rotate 180 degree, generating exposure images at 30 Hz for a duration of 5 to 7 seconds [28, 31]. The estimated effective dose for a typical acquisition of 7 seconds at 30 Hz using a torso phantom (posterior-anterior diameter 24 cm, left-right diameter 30 cm) is 0.7 millisievert (mSv) based on an esti-mated dose area product with an effective dose conversion factor of 0.183 mSv/(Gy x cm2) [48]. Using the same conversion factor the average mean
T ab le 2.2: Ov er vie w of 2D a nd 3D de vice enhancement techniques . 2D De vice Enhancement 2D Rotational De vice En-hancement 3D De vice Enhancement Commercially a v ailab le Y es No No Acquisition Static vie wing direction Shor t rotation: 90 deg rees Long rotation: 180 deg rees Projection images 45 90 157-211 Adv antages V er y fast processing (order of magnitude seconds) Stent deplo yment from v ar i-ous angles using dose sim-ilar to that for 2 static-vie w acquisitions Full 3D quantification of the stent Disadv antages Enhanced images from 1 vie wing direction at a time; only measurements in 2D Rotational acquisition lim-its enhancement; only mea-surements in 2D Dose; Slo w processing (or-der of magnitude min utes) Example image
2.5. Three-dimensional stent reconstruction
patient dose during coronary angiography is 7.2 mSv [49] and dose resulting from CT coronary angiography ranges from 4.0 to 21.4 mSv [50]. After the acquisition, the projection images are available for review and the processing of the data can start.
Schoonenberg and colleagues [31] report fully automatic processing in 9 out of 10 coronary stents in humans. Figure 2.7 gives an overview of this fully automatic process. First the operator can set an optional ROI to indi-cate the stent region and thus reduce the search space for the subsequent automated processing steps. The stent regions are automatically identified in each projection image. The only a priori information for detecting the stent region is the need to identify two periodically moving markers in each projec-tion image. Due to foreshortening, the distance between these markers can vary in the projection images. Also, due to the rotation of the C-arm system, the orientation of these projected markers can vary. First the algorithm tries to find as many marker-like structures as possible and then it tries to select the best marker couple for each frame. The marker couple then defines the stent region in an image. Because images are acquired at high frame speed (30 Hz), small angular difference between frames (less than 1 degree), and relative low heart rate (approximately 1 Hz), the displacement of the mark-ers between successive frames of the rotational run is limited. This tempo-ral constraint helps the algorithm finding the correct solution throughout the acquisition. The typical percentage of correct marker-couple detection (true positives) in the images of a rotational acquisition is about 42% without a ROI and 79% with a ROI with respectively 15% and 5% false positive detections [31].
To improve these results, a second marker-couple detection step is intro-duced. This uses a 3D model of the markers generated from the first step. Two frames with the highest likelihood of correct marker-couple detection are used to model the two markers in 3D using epipolar geometry [51]. By forward-projecting the 3D model of markers on the acquired image, a loca-tion in the projecloca-tion image is calculated. This localoca-tion serves to determine an ROI for an additional stent region detection step. The steps of marker-like point detection, marker-couple selection, and 3D marker model generation are repeated to improve the results (see optional second marker-couple de-tection pass in Figure 2.7). The 3D model of markers generated from the second step is used as the reference state to which the projection images will be motion compensated. The 3D model is forward-projected on the ac-quired image and the difference between the forward-projected markers and the detected markers is a measure of the motion of the stent. By warping the images from the detected positions to the forward-projected reference posi-tion, the motion can be corrected (see Figure 2.1). A standard cone-beam reconstruction method can be applied to the motion compensated images, resulting in a motion compensated reconstruction of the stent.
Figure 2.7: Processing steps of 3D stent enhancement: data acquisition,
marker couple detection, motion compensation and reconstruction. For a more detailed description of all the steps see Schoonenberg et al. [52] or chapter 3.
2.6. Results of enhancement of other cardiovascular devices
Figure 2.8: Volume-rendered reconstructed Taxus Express2 3.0 x 16-mm stent (Boston Scientific, Maple Grove, Minnesota) with five cross-sectional views (virtual pullback). Arrows indicate how views A through E correspond to locations. In image A, thick structure in the middle is the marker on the balloon wire. In images B through E, the wire is visible. The other dots in images A through E are stent struts. The cross sections of the stent are circular and no underexpansion is observed.
2.5.2
Visualization tool
The 3D reconstructed volume can be displayed with standard visualization techniques, such as volume rendering. The operator can freely zoom, pan, and rotate the reconstructed stent. Additionally, the stent reconstruction can also be used in a more tailored visualization technique called virtual pullback [52]. This unique tool allows an automatic moving cut plane per-pendicular to the guide wire to move and generate planar reformats similar to those obtained with IVUS. Figure 2.8 shows a volume-rendered recon-structed stent and five automatically generated cross-sectional images. Both the 3D volume-rendered reconstructed stent and the automatically generated cross-sectional images can be used for taking measurements.
2.6
Results of enhancement of other cardiovascular
devices
The techniques of 2D stent enhancement, 2D rotational stent enhancement, and 3D stent reconstruction can also be applied to other devices that have markers or marker-like features. Table 2.2 shows an overview of these
de-Figure 2.9: (Left) Exposure image from a rotational acquisition of an atrial
septal defect closure device. (Middle) Corresponding enhanced stent image. (Right) Enhanced image from a different viewing direction.
vice enhancement techniques and summarizes some key elements. These techniques can of course be applied to cardiac devices that exhibit cyclic motion, but they are also potentially beneficial for other non-cardiac, vascular devices. We will show the results of these techniques applied to two other devices in cardiac and non-cardiac applications. First an atrial septal defect closure device is imaged using the rotational acquisition protocol, see Figure 2.9. Atrial septal defect closure devices typically do not have markers. How-ever, their wire frame, made of Nitinol, an alloy of nickel and titanium, features two inherent marker-like parts: the two connection points of all wires. These marker-like parts are automatically identified throughout the rotational run as if they were balloon markers. Rotational StentBoost and 3D stent recon-struction are then applied (Figures 2.9 and 2.10). Secondly, a non-cardiac 6.0 x 18-mm ParaMount Mini GPS (ev3 Inc., Plymouth, Minnesota) stent was placed after pre-dilation within the right renal artery ostium. Then a rotational acquisition was performed. This particular stent has four radiopaque tanta-lum markers built into each end of the stent. Our automated algorithm [31] was unable to coherently select the same two markers throughout the run. Manual interaction was needed to identify one marker on each end of the stent. After this manual step, all frames were motion compensated automat-ically and the whole sequence was reconstructed. Even though motion in the renal artery is significantly less than in coronaries, it is enough to de-crease image quality to unacceptable levels. Figure 2.11 shows the motion compensated reconstructed stent.
2.7. Summary
Figure 2.10: Two motion compensated reconstructions of closure devices.
Figure 2.11: Image of 3D reconstructed vascular stent (6.0 x 18 mm) using
motion compensation.
2.7
Summary
A percutaneous intervention involving vascular devices cannot be successful unless those devices are precisely implanted and properly deployed. Tradi-tionally stent expansion can be assessed using IVUS. Routine use of IVUS may result in improved stent expansion [53]; however, it is not routinely per-formed in all patients undergoing percutaneous interventions. Disadvantages of IVUS are extra cost, additional procedure time, and increased risk of me-chanical complications. Other implanted vascular devices are also often imaged with different modalities, such as 3D transesophageal echocardio-graphy and intracardiac echo to guide interventional procedures and verify proper device placement. Bringing other modalities to the catheterization lab-oratory typically lengthens the procedural time, especially when it is a mobile system, but also when it is integrated within the X-ray suite. It also requires
additional capital investments and training of the catheterization laboratory staff.
Over the last decade, researchers have developed technology that en-ables better visualization and analysis tools based on the X-ray system itself. With the introduction of X-ray based 2D stent enhancement tools, the visibil-ity of the deployed stent has been improved in the exposure sequences [25]. This technique is an alternative to IVUS and commercially available from the main manufacturers of cardiovascular X-ray suite equipment. Philips, Paieon and Siemens all offer a similar enhanced stent visibility tool, but only Philips offers the subtract functionality to visualize the lumen together with the stent. Disadvantages of this technique are a small additional amount of radiation. However, this technique requires less additional procedural time and no ex-tra cost per patient, except for the initial costs of the software itself. A clear limitation of this technique remains the nature of 2D projection, which makes it impossible to generate cross-sectional images like those generated with IVUS. A way to tackle this problem is to make 3D reconstructions in a man-ner similar to that used in neurovascular imaging. Fully automated, motion compensated, 3D device reconstruction based on rotational X-ray angiogra-phy has been developed for cardiovascular implantable devices that exhibit motion. One apparent advantage of this angiography over the other X-ray based 2D enhancement tools is that it offers the possibility of obtaining ac-curate measurements in 3D. However, this advantage has yet to be proven in a large-scale comparison study with IVUS. Device reconstruction does not provide any information regarding surrounding tissue during device implan-tation; it only has the potential to show calcifications. This is a shortcoming compared with IVUS.
Another limitation of the 3D motion compensated reconstruction technol-ogy is foreshortening of the device in parts of the rotational run. This is reported for cardiac stents [31] and typically occurs in the proximal and distal portions of all coronary vessels. This limitation is clinically pertinent given the amount of myocardium at risk if a proximal stent were to have acute stent thrombosis. Hence, for stents in these locations, the operator should rely on 2D enhancement. This limitation might be addressed by changing the image acquisition trajectory to reduce foreshortening of the device or by improved motion compensation techniques.
A third limitation of this technique is that, because markers are required, it can only be performed during stent implantation and not during follow-up procedures. However, if in the future stents are provided with markers, or au-tomated detection using the stent contour is possible in combination with mo-tion compensamo-tion, this technique might be useful for follow-up procedures. Because 3D device reconstruction is a low-risk, cost-effective, time-efficient method, it might become an alternative for 2D enhanced device visualization and the ultrasound-based modalities to verify proper device placement.
C
H
A
P
T
E
R
3
P
ROJECTION
-
BASED MOTION
COMPENSATION AND RECONSTRUCTION OF
CORONARY SEGMENTS AND CARDIAC
IMPLANTABLE DEVICES USING ROTATIONAL
X-
RAY ANGIOGRAPHY
“The beginning of knowledge is the discovery of something we do not understand.”
Frank Herbert.
3.1
Abstract
Cardiologists use two-dimensional projection images in conventional X-ray coronary angiography for the assessment of three-dimensional structures. During minimally invasive interventions there is a need to clearly visualize and analyze contrast filled coronary arteries, surrounding tissue, and im-planted devices. Three-dimensional reconstruction of these structures is challenging due to the cardiac and respiratory motion. In this chapter we describe a method to automatically generate motion compensated recon-structions of various structures using rotational X-ray angiography.
The method uses markers on a device or guide wire to identify and esti-mate the motion of an object or region of interest in order to register and mo-tion compensate the projecmo-tion images to generate a momo-tion compensated reconstruction. The method is evaluated on 20 rotational acquisitions and the average marker couple detection rate is 84% for cardiac stents, 90% for closure devices and 20% for contrast filled coronaries. The projection images are motion compensated based on the semi-automatically detected markers and subsequently used for reconstruction. We conclude that it is feasible to reconstruct cardiac stents, closure devices, contrast filled coronaries, and calcified plaques using rotational X-ray angiography.
3.2
Introduction
Visualization of three-dimensional (3D) structures in the heart during per-cutaneous interventions is typically achieved by acquiring two-dimensional (2D) projection images with a C-arm based X-ray system. In standard an-giography (SA) multiple fixed view image sequences are used for diagnosis and treatment, whereas in rotational angiography (RA) image sequences are generated with a continuously rotating C-arm system. Recently published studies demonstrate the safety of prolonged coronary injections for rotational acquisitions [54] and that RA exposes patients to acceptable levels of con-trast and radiation while at the same time providing adequate angiographic data to complement or replace SA in the evaluation of coronary artery dis-ease [55]. Due to the nature of the acquisitions, different image processing techniques are used to enhance the structures of interest in these projection image sequences.
3.2. Introduction
3.2.1
2D enhancement
Multiple sets of images are acquired from a static gantry position in standard angiography. Due to cardiac and respiratory motion the object of interest moves in a set of X-ray projection images. Image enhancement by straight-forward averaging of multiple subsequent images will blur the moving objects. Various solutions to this problem have been proposed in literature to enhance the moving structures of interest. Spatial filtering techniques can be applied to individual images. In [56], a fairly broad review of a number of noise re-duction / image enhancement algorithms is given. This includes anisotropic filtering, total variation methods, neighborhood filtering, frequency domain fil-ters with the extensive wavelet filtering and image enhancement technique family, the more recent statistical neighborhood approaches, and finally the non-local means family, which can all be applied to SA. Combinations of spatial and temporal filtering with limited a priori information of the objects of interest has been proposed [57].
Various less general methods tailored towards enhancement of stents and other devices have also been developed. In these methods motion com-pensation is achieved by identifying the device region in all projection images in order to register these subsequent images before temporal integration. A common method reported in literature uses the balloon markers on the deliv-ery catheter to identify the stent region [26, 30, 45]. An extension to this ap-proach also uses the wire in between the balloon markers [32]. Other meth-ods do not use markers, but instead use deformable boundary detection [33] or layer decomposition based on motion [34]. The use of stent enhancement in a clinical setting has been reported [25, 36, 37]. In SA it is very common to acquire multiple contrast enhanced and none contrast enhanced image sequences from different C-arm positions to visualize the coronary tree or implanted devices from multiple angles. Based upon two images acquired from different gantry positions a 3D model can be generated [58], which al-lows 3D viewing and analysis of the object of interest.
3.2.2
3D modeling and reconstruction
Rotational angiography is very suitable for 3D modeling and tomographic reconstruction. Some of the aforementioned noise reduction and object en-hancement methods can be applied on rotational acquisitions as well. En-hancement of rotational runs by temporal filtering after marker-based regis-tration has been proven to be feasible [43]. Modeling of the coronary tree us-ing two view modelus-ing results in a geometrical description of centerlines and corresponding diameter information [58] and multiple view modeling results in a full mesh of the coronary tree [59]. Various tomographic reconstruc-tion methods have been proposed for rotareconstruc-tional X-ray angiography. These
methods typically use more projection images than modeling techniques and are often linked to a particular cardiac phase. 3D motion compensation ap-proaches used in combination with iterative reconstruction methods [60] or filtered back projection reconstruction methods [61], projection based motion compensation [62] and iterative sparse object reconstruction [63] have been proposed for contrast enhanced rotational coronary angiography. Balloon marker-based registration to motion compensate all projection images has been proposed for coronary stents [47], however this method was not fully automatic due to suboptimal marker detection.
In this chapter we will use automated, marker-based motion compensa-tion of projeccompensa-tion images and FDK reconstruccompensa-tion [64] to reconstruct a variety of implantable devices, coronaries and surrounding tissue.
3.3
Method
Our method consists of the following steps: data acquisition, marker couple detection, motion compensation, reconstruction and enhanced visualization. Each of these steps will be explained in more detail in this section.
3.3.1
Data acquisition
An acquisition consisting of 211 projection images (up to 1024x1024 pixels of 154 micron) of the isocentered object of interest is made during 180 degrees of continuous rotation (7.1 seconds) using a calibrated monoplane cardio-vascular C-arm X-ray system (Allura Xper FD20, Philips Healthcare, Best, The Netherlands). The patient is instructed to take a breath hold during the rotational run in order to minimize respiratory motion. Optionally a contrast agent can be injected to visualize the coronaries.
The study was approved by the University of Colorado Institutional Review Board. Inclusion criteria included age>= 18years, ability to provide informed consent, and an appropriate clinical indication for any vascular device, car-diac and non-carcar-diac. Exclusion criteria were known allergies to iodinated contrast, a creatinine of 1.5 mg/dL or higher, patients with ST elevation my-ocardial infarction, pregnant and breast feeding women.
In total, 20 rotational runs of cardiac stents, closure devices and con-trast filled coronaries have been acquired. The 10 imaged cardiac stents are Taxus Express2stents (Boston Scientific, Natick, MA 01760-1537, USA) with
various sizes, the closure devices are four Amplatzer septal occluder devices (AGA Medical Corporation, Plymouth, MN 55442, USA) and one CardioSeal septal repair implant (NMT Medical, Inc., Boston, MA 02210, USA), which we will refer to as closure device type A and B respectively.
3.3. Method
3.3.2
Marker couple detection
The only a priori information for the marker couple detection is the assump-tion that in each frame of the rotaassump-tional acquisiassump-tion two periodically moving markers will exist. Assuming there is no bending of the wire, these mark-ers have a constant distance in 3D space, but due to foreshortening they can have a varying distance in the projection images. The first step of the marker couple detection is to detect blob like structures. An example of blob detection with automatic scale selection is described by [65]. A marker cou-ple selection algorithm is needed, because blob detection methods identify many potential markers. Potential candidate marker couple pairs are iden-tified based upon these found markers. Parameters, such as distance be-tween the two markers, orientation and location of these markers, can be calculated for each marker pair. Adding a priori information results in tempo-ral constraints for successive frames: the marker positions will have relatively small displacements between subsequent frames due to the high framerate (30 Hz), small angular difference between successive frames (less than 1 degree) and relatively low heart rate (approximately 1 Hz).
This marker couple detection algorithm will identify markers but also other blob like features in the rotational run. If the blob detector did not identify one of the two markers in a frame, the marker couple might not be able to identify a pair that fulfills the temporal, location and orientation constraints for that frame and the frame is simply skipped.
To improve the marker couple detection results, a second pass mecha-nism is implemented. The idea is that in a first pass only two frames with correctly detected markers are necessary to generate a 3D model that can serve as an aid for marker couple selection in the second pass. A 3D model of the two markers is generated based upon detection results of two frames from the first marker detection pass. Generation of the 3D model can fail when one or both of the selected images for generating the model have in-correct marker positions. However, the user can in-correct the marker positions or select frames in which the markers were identified correctly in the unlikely event that this may indeed occur. This 3D model is forward projected to all the projection images and serves as a clue for the marker couple detection in the second pass. The distance between the markers and the angle of the imaginary line between the markers are reliable features for the marker cou-ple selection. Due to the cardiac and possibly respiratory motion the location is a less reliable feature. The location of the model can still be used to serve as a region of interest (ROI) for the blob detection, whereas the orientation information is used for the final couple selection.
