AngioLab—A software tool for morphological analysis and endovascular treatment planning of intracranial aneurysms

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

journal homepage: www.intl.elsevierhealth.com/journals/cmpb

AngioLab—A software tool for morphological analysis and endovascular treatment planning of intracranial aneurysms Ignacio Larrabide a,b,∗ , Maria-Cruz Villa-Uriol b,a , Rubén Cárdenes b,a , Valeria Barbarito b,a , Luigi Carotenuto b,a , Arjan J. Geers b,a , Hernán G. Morales b,a , José M. Pozo b,a , Marco D. Mazzeo b,a , Hrvoje Bogunovi´c b,a , Pedro Omedas b,a , Chiara Riccobene b,a , Juan M. Macho c , Alejandro F. Frangi b,a a b c

Networking Biomedical Research Center on Bioengineering, Biomaterials and Nanomedicine, Barcelona, Spain Center for Computational Imaging & Simulation Technologies in Biomedicine (CISTIB), Universitat Pompeu Fabra, Barcelona, Spain Department of Vascular Radiology, Hospital Clinic i Provincial de Barcelona, Barcelona, Spain

a r t i c l e

i n f o

a b s t r a c t

Article history:

Determining whether and how an intracranial aneurysm should be treated is a tough

Received 8 June 2011

decision that clinicians face everyday. Emerging computational tools could help clinicians

Received in revised form

analyze clinical data and make these decisions. AngioLab is a single graphical user interface,

18 April 2012

developed on top of the open source framework GIMIAS, that integrates some of the latest

Accepted 4 May 2012

image analysis and computational modeling tools for intracranial aneurysms. Two work-

Keywords:

Planning (ETP). AngioLab has been evaluated by a total of 62 clinicians, who considered the

flows are available: Advanced Morphological Analysis (AMA) and Endovascular Treatment Intracranial aneurysm

information provided by AngioLab relevant and meaningful. They acknowledged the emerg-

Patient management

ing need of these type of tools and the potential impact they might have on the clinical

Advanced morphological analysis

decision-making process.

Endovascular treatment planning

© 2012 Elsevier Ireland Ltd. All rights reserved.

Integrative platform Clinical evaluation

1.

Introduction

Intracranial aneurysms (IAs) are abnormal focal dilations of cerebral arteries that may rupture and eventually cause a subarachnoid hemorrhage (SAH) [1]. Stroke is among the leading causes of death in the western world. SAHs correspond to 20% of all strokes with IAs accounting for 85% of all SAHs [2]. Over the last decade, great advances in interventional imaging and a new generation of therapeutic devices have provided major improvements in IA diagnosis and treatment. Currently, clinicians take into account patient age, presence

of symptoms, and aneurysm size and shape when assessing the best treatment option [3]. There are two treatment strategies: Surgery (clipping or by-pass) and endovascular therapies (coil embolization, stent-assisted coil embolization, and flow diverter deployment). Both strategies aim to exclude the aneurysm from the blood circulation. However, particularly with endovascular devices, the outcome of IA treatment is difficult to predict. Clinicians are becoming more interested in image quantification and personalized computational modeling; they demand tools to improve patient assessment and treatment planning. Therefore, the development and assessment of such

∗ Corresponding author at: Networking Biomedical Research Center on Bioengineering, Biomaterials and Nanomedicine, Barcelona, Spain. Tel.: +34 935422966. E-mail address: [email protected] (I. Larrabide). 0169-2607/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.cmpb.2012.05.006

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

tools is receiving increasingly more attention in recent years [4]. We can find general purpose libraries (e.g. Visualization Toolkit (VTK) [5] and Insight Toolkit (ITK) [6]), others specialized for vascular analysis (e.g. VMTK [7], TubeTK [8]) and commercial and non-commercial end-user applications which make use of the previous ones (e.g. 3D Slicer [9], Amira [10], Osirix [11]). This tool offers a set of advanced capabilities for segmenting and analyzing geometrical characteristics of patient-specific vasculature. In Table 1, we summarize the main features of these vascular analysis tools and those of the software we are presenting here. One important software initiative called the Vascular Modeling Toolkit (VMTK) [7] has been designed to perform 3D reconstruction, geometric analysis, mesh generation and surface data analysis of vascular models. However, the tool is not focused on a specific vascular disease and only part of its features have been integrated as a plugin into 3D Slicer [9] to provide a graphical user interface (GUI). Another important software package is OsiriX [11]. This package offers one plugin specifically designed for coronary vasculature segmentation from CTA images, called CMIV-CTA [12] that also allows multiple visualization capabilities. Other software, currently under development, is TubeTK [8], a software package without GUI, with capabilities to segment and extract vessel geometrical properties of vessels from medical images that can also be used for vascular atlas construction. Another similar tool is @neuFuse [13], which is an end-user application for the creation of 3D vascular models, morphological analysis, set-up and post-processing of hemodynamics simulations developed in the context of the @neurIST project [14]. Some of the algorithms and methods integrated in @neuFuse are available in the tool described in this work (e.g. vascular segmentation, morphological analysis and virtual stenting). For this reason, @neuFuse is the tool most similar to the one presented here. Finally, we mention Mimics [15], which is a medical image processing tool that forms part of a more general application called Mimics Innovation Suite. This tool offers a set of advanced capabilities for segmenting and analyzing geometrical characteristics of patient-specific vasculature. The objective of this work is to present AngioLab, a software tool for the morphological analysis and endovascular treatment planning of IAs. Two workflows have been created and implemented in AngioLab for image-based management of IAs: 1. Advanced Morphological Analysis, and 2. Endovascular Treatment Planning, which are schematically represented in Fig. 1. These two workflows have been implemented in four plugins within the GIMIAS framework [20] described later. The included methods are a subset of those described by Villa-Uriol et al. [4]. AngioLab has also been brought into the hands of clinicians for their assessment during two hands-on workshops. Participants were mainly neuroradiologists and neurointerventionists who used the software for the first time and provided feedback through an anonymous survey. The survey inquired the participants about the usability and the clinical value of the methods presented in a normal use case. We would like to point out that the validation of the algorithms included in AngioLab is not in the scope of this study and has been addressed in the references cited herein.

807

The goal of AngioLab is to provide information to clinicians, engineers and computer scientists that help them to better evaluate IA patients and understand the effect of treatment. AngioLab is an end-user application, based on the mentioned libraries, that includes a complete set of features for a specific clinical application. So far, there is no distribution strategy for this software, because it is still in a prototype stage.

2.

AngioLab description

AngioLab has been developed on top of GIMIAS (Graphical Interface for Medical Image Analysis and Simulation), a workflow-oriented software framework for medical image computing and computational physiology [19,20]. This framework allows building medical prototypes for clinical evaluation and simplifies the integration of tools for building new clinical workflows, by using open-source libraries to accomplish various tasks, including graphical editing, user interfaces, image analysis, anatomical modeling, computational physiology and visualization. Prototypes can be verified by research users, thus reducing the effort required to translate new concepts to the clinical environment. In this section, we briefly describe the GIMIAS architecture, how AngioLab was designed on top of it, and AngioLab’s GUI.

2.1.

GIMIAS framework

GIMIAS relies on a set of open-source libraries that are standards in biomedical imaging, such as VTK, ITK, the DICOM Toolkit (DCMTK), the Medical Imaging Interaction Toolkit (MITK), Boost, wxWidgets and CGNS. The integration of these libraries is obtained by leveraging a software infrastructure including routines to read and write different file formats, to support and optimize data handling, to provide consistent user interaction and visualize different data types (2D, 2D+t, 3D and 3D+t images, surface and volumetric meshes, signals, landmarks, contours, vector fields, etc.). The wide number of features provided by standard libraries are available through an Application Programming Interface (API), which permits to be easily extended and reused. The GIMIAS extensibility stems from a modular architecture, which permits users such as scientific developers and bioengineering researchers to extend the framework with their own algorithms and methods, as depicted in Fig. 2. The architecture is implemented as a set of layers featuring different levels of abstraction. The architecture layers can be summarized as follows.

• The library layer contains widely used open-source libraries. The user can add new libraries, which can contain new algorithms and methods developed by her/him or can be third party ones. • The framework layer provides an API through which the plugins interact with the library layer and with each other. It includes GUI components available to all plugins (widgets), facilities for user interaction with rendered data (interactors), readers/writers of various file formats, data handling functions.

808

Funtionality Vessel segment Geom. analysis Z.M.I. analysis Centerline Virt. stent. Virt. coiling Advanced Meshing GUI Visualiz.

Based on Platform support License Clinical application References Group/company

VMTK

3D Slicer VMTK

OsiriX + CMIV-CTA

TubeTK

Mimics

@neuFuse

AngioLab

Yes Yes No Yes No No Meshing

Yes Yes No Yes No No Meshing

Yes Yes No Yes No No –

Yes Yes No Yes Yes No Meshing

Yes 2D / 3D

Yes 2D/3D

VTK ITK python Lin, Win, MacOS BSD Vascular [7] Orobix

VTK, ITK CTK, KWW Lin, Win, MacOS BSD Vascular [9,16,7] BWH, HMS + Orobix

Yes 2D / 3D MPR, CPR VTK, ITK DCMTK, Papyrus MacOS GNU-GPL Coronaries [11,17,18] OsiriX F. + CMIV

Yes Yes Yes Yes Yes No Mesh Edit Yes 2D/3D

Yes Yes Yes Yes Yes Yes Mesh edit.

No No

Yes Yes No Yes No No Atlas const. No No ITK, CTK

–

VTK, ITK KWWidgets

Lin, Win, MacOS Apache 2.0 Vascular [8] Kitware

Win Commercial Vascular [15] Materialise

Win – Cerebral/vascular [13] B3C

Yes 2D/3D, MPR, Virt. X-ray VTK, ITK, MITK, VMTK wxWidgets Lin, Win Academic Cerebral/vascular [19] CISTIB, UPF

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

Table 1 – Information and features of seven different vascular applications, including AngioLab. Abbreviations: CMIV: Center for Medical Image Science and Visualization; MPR: multi-planar reformatting; CPR: Curved Plane Reformatting; ZMI: Zernike moments invariants; CTA: Computer Tomography Angiography; CTK: Common Toolkit; DCMTK: DICOM ToolKit; MITK: Medical Imaging Interaction ToolKit; KWW: KW Widgets. BSD: Berkeley Software Distribution; GNU-GPL: GNU General Public License; BWH, HMS: Brigham & Women’s Hospital, Harvard Medical School; B3C: Biocomputing Competence Center.

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

809

Fig. 1 – High-level schematic view of the Advanced Morphological Analysis and Endovascular Treatment Planning workflows. These workflows go from patient data and device descriptions to concrete descriptors for the diagnosis of IA and their treatment selection. The clinical assessment was focused on the three components highlighted in dark red (namely anatomical modeling, morphological characterization and classification, and virtual stenting). Adapted from Villa-Uriol et al. [4]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

• The plugins layer extends GIMIAS by adding the GUI elements and provide access to specific algorithms or filters of a given library. Generally, a new plugin includes a main window, toolbars, working areas and widgets. • The application layer allows arranging a set of functions provided by the plugins layer in the form of a workflow, so that those functions are organized in the final application as a sequence of operations within a monolithic and coherent GUI. The GIMIAS architecture allows decoupling the design of the user interaction from the underlying algorithms, and the integration with third party libraries (such as image and geometry filtering, data rendering and I/O), which are managed by the underlying framework.

these tools to patient data, derive patient-specific descriptors and visualize this data. The plugins have been organized in the application in the form of workflows. All modules needed to build AngioLab, including those belonging to GIMIAS and those specific to AngioLab, are represented in Fig. 2. Therefore, a set of C++libraries comprising algorithms for the management of IAs has been developed: The Anatomical Modeling Library can be used for the extraction of vascular anatomy from medical images; the Morphological Analysis Library can be used for the morphological analysis and geometrical characterization of IAs; the Virtual Treatment Library can be used for the simulation of patient treatment with different endovascular devices. The methods and the computational models of these libraries are described in more detail below.

2.2.

2.3.

AngioLab design on top of GIMIAS

The tools depicted in Fig. 1 have been integrated in GIMIAS. A set of plugins has been developed to allow the user to apply

Graphical user interface

The AngioLab interface is composed of five parts (Fig. 3): menu bar, data tree, plugin selector, operations panel, visualization

Fig. 2 – Schematic representation of the implementation of the application AngioLab (on top of GIMIAS). The modules needed to build AngioLab include libraries and plugins belonging to GIMIAS (white) and specific libraries and plugins that are specific for the workflow (grey).

810

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

Fig. 3 – AngioLab working area description for the segmentation plugin showing the segmentation results for a 3D rotational angiography (3DRA) image and the main areas in the graphical user interface. The main components of the interface are also highlighted (i.e., menu bar, data entity list, plugin selector, operations panel, visualization panel and toolbar).

panel and toolbar. The menu bar provides access to the standard file, edit, view, tools and help menus. It allows the user to load medical images in DICOM and VTK formats, connect to a PACS to retrieve images and save data on a local hard drive. The view menu provides access to visualization modalities, including the possibility of opening floating windows and tool-bars with processing and visualization functionalities. Basic capabilities permit to change the visual appearance of entities, to explore the properties of data, to record movies and to capture screenshots. The plugin selector allows the user to choose one of available plugins to perform different operations. The data tree shows a hierarchical list of loaded data in which the current selection is highlighted. The supported entity types are those provided by the GIMIAS framework and include image, surface and volume mesh, skeleton, points and measurements. The operations panel lists the operations available for each plugin, together with a sub-panel with the corresponding options for the current operation. Finally, depending on the plugin and selected operation, the configuration of the working area changes.

3.

AngioLab workflows

Two workflows, namely Advanced Morphological Analysis and Endovascular Treatment Planning, have been built into AngioLab. The steps that build these workflows are represented in Fig. 4 and summarized below:

• Morphological characterization and classification: The extracted geometry representing the cerebral vasculature, including the IAs, is quantified by extracting the morphological indices and by retrieving a list of similar aneurysms together with their related clinical information. This is obtained through the Anatomical Modeling plugin and the Morphological Analysis plugin. • Endovascular treatment planning: In this workflow, a larger portion of the vessel is considered where a virtual device can be deployed. Then, a virtual X-ray visualization can be generated from the patient image. Finally, the results of the hemodynamics simulations are visualized for the extraction of hemodynamics descriptors. For this workflow, the Anatomical Modeling plugin, the Virtual Treatment plugin and the Advanced Visualization plugin are used. AngioLab plugins provide all user interaction functionalities required to access the algorithms contained in the corresponding libraries while exploiting the facilities provided by the framework. In addition to AngioLab’s specific plugins, GIMIAS DICOM plugin has been considered to build the final application. This plugin allows DICOM data search and retrieval from the local hard drive or from a PACS.

3.1.

Anatomical modeling

An accurate representation of the vascular region of interest (ROI) is required for both the aforementioned workflows. To

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

811

Fig. 4 – Representation of the workflows built in AngioLab.

obtain this, a Geodesic Active Regions (GAR) [21] segmentation method is used. The GAR method is based on a geometric deformable model within a level set framework. The internal force of the deformable model depends on the local curvature of the evolving surface, while the external force combines region-based descriptors with gradient-based ones to drive the evolution of the model towards the vascular boundaries (Fig. 5(a) and (b)). Given that a fixed training dataset is needed for a given image modality, an image intensity standardization (IIS) filter [22] is used to ensure that intensity ranges corresponding to the same tissues are similar between the data to segment and the training data, eliminating the dependence on the specificities of the imaging equipment employed [22]. After the segmentation, a 3D surface representation of the vascular wall is obtained, which requires some further postprocessing preceding the morphological analysis. In this step, the mesh manipulation operations are fundamental for creating anatomically plausible models for the hemodynamic characterization of IAs (Fig. 5(c)). The Anatomical Modeling plugin provides tools for the selection of a region of interest, for the generation of the anatomical models from the segmented medical images, and for the visualization of the segmentation results within the working area. Surface correction and edition through the elimination of

triangles, hole filling and smoothing, amongst others, is also possible with this plugin.

3.2.

Morphological analysis

Morphology is among the primary factors evaluated to recommend or discard a treatment. If an endovascular intervention is to be selected, an estimation of the aneurysm and parent vessel dimensions are taken into account to select the endovascular devices [23]. Usually, this selection relies on the clinician’s prior experience. AngioLab offers a tool for measuring geometrical features relevant for IA treatment selection. First, basic measurements are performed on the surrounding vasculature (vessel diameters) and on the aneurysm (dome dimensions [4]), which require user intervention. For calculating the vessel diameters, a probe is placed where the measurement needs to be made, and for determining the aneurysm dome dimensions, the aneurysm neck is delineated to isolate it from the rest of the vasculature (Fig. 6(a)). Measurements are automatically computed and graphically visualized on top of the geometric model and in a table. In standard clinical practice, these measurements are manually performed on a volume rendering visualization, which makes them highly dependent on the view angle and visualization parameters

Fig. 5 – Results of the anatomical modeling showing (a) the 3DRA image being segmented and the segmentation result overlayed, (b) the segmentation result and (c) the vessel cutting tool.

812

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

Fig. 6 – IA morphology characterization. (a) Manual delineation of the aneurysm neck and automatically computation of the basic measurements. (b) Schematic representation of the aneurysm morphological feature space (ZMI of order 1–3). (c) Morphological and clinical data of the most similar aneurysms are shown to the user after having queried a database for similar aneurysms. The user has the possibility of populating the database with the aneurysm and the patient information of the analyzed case.

chosen by the user. In AngioLab, despite the necessity to perform editing operations, the computed measurements provide an accurate observer independent estimation of the critical parameters used to select the dimensions of the endovascular device to be employed for the intervention. Second, more complex descriptors of the shape of the aneurysm and surrounding vasculature are Zernike moment invariants (ZMIs) [24], which are extracted in real time (Fig. 6(b)) [25]. These descriptors have been previously correlated to aneurysm rupture risk in a population of cerebral aneurysms located at the Middle Cerebral Artery. Moreover, they can be used to automatically search for similar IAs in an existing database (Fig. 6(c)). In this database, each aneurysm is stored along with relevant information on the patient’s clinical history, including gender and age, aneurysm rupture status, and information about treatment and treatment outcome. The user of AngioLab can consult this information to make a more informed decision on the aneurysm of interest. The Morphological Analysis plugin can measure the vessel diameter at any location by placing a probe on the centerline. It allows the user to delineate the aneurysm neck and define perpendicular cuts of surrounding vessels to select regions of interest (ROI), which are used as input to automatically compute dome dimensions and ZMIs. Based on the ZMIs, the plugin identifies similar aneurysms in an external database, which are then visualized alongside the aneurysm that is being processed. The user can browse through the similar aneurysms to review their clinical background in more detail.

3.3.

Virtual treatment

The ability to virtually implant different devices into the patients’ own vasculature, permits the user to assess different treatment alternatives a priori and evaluates the most appropriate intervention for the current patient. With this tool, the different treatment alternatives (associated with different devices, device models, geometries, sizes and positions) can be virtually explored to assist treatment planning.

The Fast Virtual Stenting (FVS) method, developed by Larrabide et al. [26], has been integrated in the Endovascular Treatment Planning workflow. The FVS method is based on constrained deformable models, where a simplified description of the stent geometry is used to constrain the expansion of the stent (Fig. 7(a)). Different stent models, which represent real commercial stents, can be approached by the method (Fig. 7(b)) allowing clinicians to implant different stents before treating the patient. This method allows for a fast virtual implantation of stents in anatomically realistic vascular models (Fig. 7(c)), providing accurate results in a few seconds [27,28]. These characteristics help the user evaluate the appropriate stent design and size (Fig. 7(d)). Remarkably, the sole use of the visualization of the medical image merged with the device provides useful insights to the clinician about the potential treatment outcome (coverage of the aneurysm neck, stent apposition on the vessel wall, stent coverage of branching vessels, etc.). This information is provided to the clinician in approximately 10–20 s. Endovascular coiling is another minimally invasive therapeutic option for IAs. Simulating coil deployment is challenging because of the geometrical complexity of both coil and aneurysm. In the work of Morales et al. [29], a methodology was first presented to simulate IA coiling in realistic geometries extracted from medical images. In this algorithm, the coil tip advances while minimizing the potential energy associated with the vessel wall boundaries and previously inserted coils. Further details on the implementation of the method are provided by Morales et al. [30]. With this algorithm, clinically reported coil packing rates ranging from 8 to 40% can be achieved in realistic IA geometries [31]. Moreover, it opens up the way for further studies concerning pre- and post-treatment outcome assessment in patients with diverse morphologies (aneurysm shape, size and parent vessel structure, etc.). Additionally, the use of stent-assisted coiling has become very popular [32]. This treatment strategy is mostly adopted

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

813

Fig. 7 – In (a) is visualized a detail of the constrained 2-simplex mesh used for the representation of the stent. In (b) a stent mesh used by the FVS method is shown together with the corresponding micro-CT of the original stent. Different vascular geometries of IAs after the virtual treatment with stents are presented in (c). In (d) is presented a detail of the coiled aneurysm and the released stent position.

for wide-necked aneurysm, where a stent is used to give mechanical support to the coils, keeping them inside the aneurysm and avoiding coil migration. This treatment can also be modeled by AngioLab. The Virtual Treatment plugin computes the centerline of the vasculature and virtually deploys one or more stents (selected from a database) within the parent vessel. The plugin provides functionalities to select different stent models, the position of the stent (by choosing the stent’s central point on the vessel centerline), move the stent to a different position and visualize the deployed stent within the patient vessel geometry. It also allows for the virtual deployment of coils by specifying the number of coils, their length and width after selecting them from a coil database.

3.4.

Advanced visualization

An appropriate visualization of the treatment planning results is mandatory to ensure an adequate interpretation of the simulations and models created. In particular, clinicians routinely base their work on X-ray images. An X-ray rendering technique for the visualization of volume data (see Fig. 8(a)) has been developed and integrated in the workflow. In particular, the 3D angiography image and the implanted device are fused and presented to the user in a single X-ray image. This has been specifically designed to assess the treatment in an interactive fashion. The outcome is presented in a similar way as the clinician normally visualizes the treatment for evaluation after the intervention. This visualization allows for the a priori assessment of the stent position along the vessel. Endovascular devices are designed to modify the local hemodynamics in the region of the aneurysm by occluding and isolating it from the main blood stream. These changes can be further evaluated by the use of computational

fluid dynamics (CFD) techniques [33]. In the case of elective patients, where the aneurysm is detected incidentally and the treatment is scheduled several weeks in advance, further CFD simulations with and without treatment can be performed to understand and assess the effect of treatment on the local hemodynamics [34]. This information is intended to support the clinician in selecting the most suitable device(s) for a particular patient. The visualization of information using advanced rendering capabilities such as streamlines and wall shear stress (WSS) is valuable (Fig. 8(b) and (c)). In particular, the visualization of CFD results in a clinician-friendly manner is achievable by the use of virtual angiographies [35]. Such data, which is typically very large (in the order of gigabytes), demands for fast algorithmic solutions that allow an effective exploration and interpretation. Such computationally efficient implementations are available in AngioLab. The Advanced Visualization plugin can render the results of virtual treatment by activating the X-ray rendering. In addition, the user can load results of CFD simulations performed externally and activate the desired visualization to compare the results of the treated and untreated models. The user can inspect the different time frames and choose among different visualizations: dye contrast by means of virtual angiographies, WSS by means of color maps, velocity by means of streamlines.

4.

Results

AngioLab was evaluated during the 2nd and 3rd Teaching Courses organized by the European Society for Minimally Invasive Neurovascular Treatment (ESMINT) (http://www.esmint.eu/). Both events were held in Barcelona, Spain, respectively on December 13–18, 2009 and December

814

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

Fig. 8 – Visualization of flow in IAs. Fig. (a) presents the visualization of the stenting output with a surface rendering, the volume rendering of the vasculature with the embedded stent and the virtual X-ray visualization. In (b) are presented the streamlines at systole after the flow simulation of the untreated and treated cases. Fig. (c) presents the wall shear stress (WSS) for the untreated and treated cases at the systole and diastole. Finally, in (d) the virtual angiography visualization for selected time-steps in the untreated and treated cases during the contrast fill-in and wash-out in the aneurysm region is provided.

5–10, 2010. The attendees were neurologists, neuroradiologists and neurointerventionists. In each of these events, a 1-h hands-on workshop on AngioLab was held. In both occasions, the attendance was optional for the participants of the ESMINT Teaching Course main track.

4.1.

Workshop objectives and structure

The main objective of both workshops was to evaluate the usability and potential impact in clinical practice of the AngioLab workflows presented in Section 2. The first workshop focused on exposing the users to the different plugins composing it. This workshop did not consider the virtual coiling operation, which was only shown in videos. Participants were asked to load a DICOM image, select the vascular ROI, perform the segmentation, clean the extracted mesh and compute the mesh centerline. As part of the Advanced Morphological Analysis workflow, they had to delineate the aneurysm neck, isolate the geometry to perform the aneurysm quantification, and subsequently obtained the k most similar aneurysms from a provided database. As part of the Endovascular Treatment Planning workflow, users had to select the position of the stent along the vessel centerline, the particular device to be used and its size (radius and length), to virtually place it within the selected geometry. Surface and volume rendering, and virtual X-ray visualizations were used to present the virtual treatment results. The second workshop was extended by allowing the user to compare and evaluate the flow alterations induced by several therapeutic alternatives, including coil embolization (with and

without stent-assistance). Because of time constraints, users were freed from doing the mesh cleaning, set-up of CFD models and computation of CFD simulations, which were given as input. Flow simulations in the absence (pre-treatment) and presence (post-treatment) of the simulated devices were visualized using virtual angiography. It was the first-time exposure of all participants to AngioLab. After a general introduction to the software and the workshop objectives, participants were asked to use and explore AngioLab following a tutorial specifically written for each occasion. During that time, each participant had access to a workstation where AngioLab was installed, and received the assistance of instructors. In the first workshop, the provided workstation had an Intel® CoreTM 2 Duo E8500 3.16 GHz processor, 4 GB of RAM memory, an NVIDIA® GeForce® 9600 GT graphics card, a 32-bit Windows XP operating system, and AngioLab release 1.0.0 (GIMIAS 0.8.0, 32-bit Windows). In the second workshop, the workstations had the same hardware, a 64-bit Windows XP operating system, and AngioLab release 1.3.0 (GIMIAS 1.1.1, 64-bit Windows). After the completion of the tutorial, each participant filled out an anonymous questionnaire about the workflow and its activities. A summary of the survey results is presented below.

4.2.

Workshop statistics

The first and second workshops were attended by 30 and 32 participants of which 23 and 28 were clinicians. When comparing the two workshops, the population could only be considered different for one age group (marked with * ) due to

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

the margin of error for the sample size: 8.7% vs. 10.7% were in the age range of 20–29 years old, a 65.2% vs. 28.6% between 30 and 39*, a 21.7% vs. 32.1% between 40 and 49, a 4.3% vs. 10.7% between 50 and 59, and a 0% vs. 14.3% were 60 or above. In the second workshop the attendees with more than 5 years of experience represented a larger proportion, a 52.2% vs. 71.4%, and below or equal, a 47.8% vs. 28.6%. Attendees were residents in 10 vs. 13 different countries. They were mainly located in Europe (82.6% vs. 78.6%), although there were a few representatives from America (13% vs. 0%), Africa (4.3% vs. 3.6%) and Asia (0% vs. 3.6%). Fig. 9 summarizes the user satisfaction ratings for the morphological analysis, the virtual stenting and the virtual coiling. For the first two, we observed similar levels of satisfaction for the morphological analysis (69.6% and 78.6%) and for the virtual stenting (60.9% and 64.3%) considering the margin of error due to sample size. If we add the “partially satisfied” answers, a similar result is observed (78.3% and 92.9% for the morphological analysis, 78.3% and 85.7% for the virtual stenting). Virtual coiling was only evaluated during the second workshop, in which 57.1% of the users were satisfied with its results. Adding the group of partially satisfied users, the total becomes 89.2%. Two types of flow visualizations were made available during the second workshop: virtual angiography and WSS. Both visualizations could be used to compare the CFD results between the untreated and virtually treated cases for one aneurysm. Fig. 10(a) compares the level of user satisfaction achieved by both types of visualizations. In general, we observe that the WSS visualization was more widely considered as a radical improvement with respect to usual treatment assessment (42.9% for the WSS vs. 32.1% for the virtual angiography), but a smaller number of users considered virtual angiography as not providing any improvement in the understanding of the changes induced after virtually treating an aneurysm (7.1% for the WSS vs. 3.6% for the virtual angiography). Participants were also enquired about the impact of this tool in the daily clinical practice and they were asked who could be its ideal user for planning the endovascular treatment (Fig. 10(b)). Most of the clinicians would like to use the tools themselves (73.9% vs. 78.6%), whereas others would prefer to give this task to a dedicated clinical scientist (21.7% vs. 10.7%). When participants were asked if the information provided by this software will support them in their therapeutic decisions (Fig. 11(a)), they responded that it will have at least some influence in 82.6% vs. 75.0% of the cases in the first and second workshops, respectively. Notably, no response indicated that this tool would not have any impact in their decision making process. The participants were also asked if the software is ready to be introduced into clinical practice (Fig. 11(b)). In both events, most of the responses indicated that, according to their opinion, further work is needed (78.3% vs. 57.1%) but in the latest edition of the workshop more participants perceived that the software is ready to be introduced in the clinical practice (13.0% vs. 32.1%). No response indicated that this software will never reach the clinical practice. The participants where also enquired about the urgency and demand for this software (Fig. 11(c)). The responses were balanced in both workshops: significant (39.1% vs. 39.3%), emerging (39.1% vs. 35.7%) and low (13.0 vs. 10.7). In total, more than 75% of the participants see an emerging or already existing demand for this software.

5.

815

Discussion

None of the libraries and applications presented in Section 1 implements a workflow solution for managing the intracranial aneurysms, as we do in AngioLab. There are existing tools specifically designed for processing and analyzing vascular structures. However, many of these tools are supporting libraries without GUIs (e.g. TubeTK, VMTK), or they are designed for a general purpose (e.g. 3D Slicer). There are also applications that include solutions to more specific and challenging vascular problems, as the case of Osirix and its plugin, CMIV, to model coronary arteries. Other commercial applications such as MIMICS allow the user to perform advanced analysis and measurements of vessels, and also simulations of device implantation. Finally, the most similar application, @neuFuse, integrates many useful algorithms for aneurysm quantification, but its strongest point is the vascular mesh editing. AngioLab is equipped with basic algorithms commonly found in the aforementioned packages such as visualization, segmentation and centerline computation techniques, but it includes several features for vessel and aneurysm analysis not available elsewhere, such as the advanced morphological analysis of aneurysm shapes with ZMIs or the rapid search method for finding similar aneurysms. Another feature that distinguishes AngioLab is the possibility to simulate the implantation of intravascular devices (stents and coils) in an interactive and efficient way. Although aneurysm stenting is also possible with MIMICS and @neuFuse, for instance, this feature is not integrated with coiling. The integration of the various algorithms into a single pipeline is finally what makes AngioLab a useful and attractive tool for the clinicians, and also differentiates it from similar solutions dealing with vascular problems. AngioLab has been designed on top of GIMIAS. This makes AngioLab easy to adapt, refine and extend by incorporating new algorithms [36,37]. Although the GUI is not yet tailored for clinical use (see responses in Fig. 11(b)), it is particularly useful for research. In the near future, once the methods included in AngioLab are fully validated, a version suitable for clinical practice will be available. The two workflows analyzed in this study are based on algorithms and methods described elsewhere [4]. The Morphological Characterization and Classification workflow relies on the quantification of geometrical features [38] and on the Zernike Moments Invariants decomposition for the selection of similar aneurysms on the basis of the shape [24,25]. The aneurysm database needed for this workflow can be of any size and can be progressively populated by the clinician as she/he treats IA patients. All additional data, such as followup images, treatment outcomes and evolution, can be linked to the database thus providing a simple and fast way for the clinician to access prior information. The Endovascular Treatment Planning workflow mainly relies on virtual stenting and virtual coiling techniques, for the creation of different treatment scenarios. Each one of these scenarios can be studied by the clinician to assess which option is most appropriate for the patient based on qualitative (visualization of the stent in the vessel and of the coils in the aneurysm, etc.) and quantitative (neck

816

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

Fig. 9 – (a) Were you satisfied with the results provided by the software when searching for similar aneurysms and performing automatic measurements of aneurysm dimensions?, (b) Were you satisfied with the results provided by the software when presenting the final location and shape of the stents? and (c) Were you satisfied with the results provided by the software when presenting the final location and shape of the coils?

coverage by the stent, % of laterals vessels occlusion, coil packing rate, etc.) descriptors. The combination of virtual treatment and CFD simulations provides meaningful insights for optimizing the treatment outcome [33,34]. Although the CFD analysis takes considerable time and is not applicable in urgent cases, its use in elective cases is feasible. By simulating different treatment options, the clinician can assess each one of them separately to select the most appropriate for the patient. The clinical value of CFD derived biomarkers is currently under investigation and there is still no concluding evidence indicating their predictive value. Still, such biomarkers are showing potential for the early evaluation of treatment outcome [39–41]. Most participants of the workshop agreed that there is an emerging or even significant clinical need for tools such as AngioLab and the majority indicated that the provided information would at least occasionally be used to support therapeutic decisions. Still, more work is required before this

software can be introduced into daily clinical practice. AngioLab will continue to aim at narrowing the gap between engineering tools and clinical practice. One illustration of this is the possibility to visualize the flow field using virtual angiography. The use of this advanced visualization technique that mimics X-ray imaging provides a familiar interface for qualitatively assessing hemodynamic simulations [35]. The efficient implementation of virtual angiography in AngioLab allows clinicians to interact with transient blood flow simulations from different view points in real time. AngioLab has been evaluated at two major clinical events in consecutive years. The user satisfaction was high for the results presented in each of the workflows (70% for the morphological analysis, 60% for the virtual stenting and 55% for the virtual coiling). From our point of view, as endovascular treatments become more complex, the need for predictive tools capable of assessing its effect accurately bear more importance. The workshop results also provide evidence that

Fig. 10 – (a) How much the visualization of wall shear stress (i.e. comparing the WSS and VA between the untreated and the virtually treated case) improved your understanding of the possible treatment outcomes? (b) Ideally, Who should use the tool for planning the endovascular treatment?

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

817

Fig. 11 – (a) The information provided by this software will support me in my therapeutic decisions (i.e. selection of treatment). (b) Is this software ready to be introduced into your daily clinical practice? (c) How urgent/demanding is the clinical need for this software?

clinicians are interested in using the morphological quantification and treatment planning features by themselves, rather than delegating this task on a third party (indicated by more than 70% of the responses). Therefore, the improvement of the usability and adaptation of the user interfaces for nonengineers are recognized to be crucial. Nowadays, several methods are being developed for analyzing vascular geometry and topology. Among them, the automated classification of cerebral vasculature segments and the isolation of the aneurysm [36] are promising tools for the clinician. Such tools would provide unambiguous criteria for image-based geometrical aneurysm characterization and for treatment selection, removing inter- and intra-observer variability. Also, the development of more computationally efficient CFD methods is required to bring these methods closer to the clinical practice. The use of steady-state vs. transient simulations when appropriate would drastically reduce the computational time required for CFD simulations [42]. Similarly, the use of porous medium models to reproduce the effect of stents and/or coils may provide an accurate representation at a much lower computational cost [43]. This software cannot be made available to the general public. Nevertheless, it can be made available for research purposes upon request.

6.

When the data of a new aneurysm is loaded, AngioLab can search for similarly shaped aneurysms in a pre-existing database. By evaluating the virtual treatments results, the clinician can assess different options before the intervention. By combining this with CFD analysis, it is possible to characterize aneurysm hemodynamics and quantify the effect of the treatment on these hemodynamics. Virtual angiographies and WSS visualizations improve the interpretation of the results by the clinician. The combination of morphological analysis, treatment planning and treatment evaluation based on CFD makes AngioLab a unique tool for the integrated management of IAs. AngioLab has been evaluated by clinicians in two international events during two consecutive years. Clinicians recognized an emerging need for the capabilities provided by AngioLab and indicated that they might influence their opinion in the daily clinical decision-making process. Nevertheless, these tools are still under development and further numerical and clinical validation is needed. With respect to existing tools, AngioLab provides a complete environment tailored for the management of IAs in one application, improving turnaround time. Results obtained during the clinical evaluation support the future extensions of AngioLab and the utility of GIMIAS as an efficient development framework for research tools and prototypes.

Conclusions

We have described the workflows currently implemented in AngioLab, an image analysis and computational modeling tool developed within the GIMIAS framework [20]. These workflows are framed within a more ambitious integrated pipeline for intracranial aneurysm management [4]. The current implementation of these workflows comprises several methods and algorithms: anatomical modeling, morphological characterization and classification, virtual treatment and virtual X-ray visualization. It also includes advanced visualization capabilities for performing the real time virtual angiographies based on CFD simulations.

Acknowledgments The authors would like to thank Martin Bianculli, Jakub Lyko, Xavier Planes, David Lucena, Albert Sanchez and Yves Martelli for the support on the development, and Annarita Bernardini, Carolina Valencia, Chong Zhang, Martha Aguilar and Minsuok Kim affiliated to the Center for Computational Imaging & Simulation Technologies in Biomedicine (CISTIB), Universitat Pompeu Fabra, Barcelona, Spain for their support during the workshop. This work was partially supported within the CENIT-CDTEAM and CENIT-cvREMOD projects funded by

818

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

the Spanish CDTI and partly within the framework of the @neurIST Project (IST-2005-027703), which is co-financed by the European Commission within the Sixth Framework Program. R.C. is partially funded by the Beatriu de Pinos program from AGAUR (Generalitat de Catalunya, Spain), A.F.F. is partially funded by the ICREA-Academia program.

references

[1] J. Brisman, J. Song, D. Newell, Medical progress: cerebral aneurysms, The New England Journal of Medicine 355 (9) (2006) 928–939. [2] H.R. Winn, J.A. Jane, J.A. Taylor, D.R. Kaiser, G.W. Britz, Prevalence of asymptomatic incidental aneurysms: review of 4568 arteriograms, Journal of Neurosurgery 96 (2002) 43–49. [3] D. Wiebers, The International Study of Unruptured Intracranial Aneurysms Investigators, Unruptured intracranial aneurysms: natural history clinical outcome and risks of surgical and endovascular treatment, Lancet 362 (9378) (2003) 103–110. [4] M.-C. Villa-Uriol, I. Larrabide, J.M. Pozo, M. Kim, O. Camara, ´ M. de Craene, C. Zhang, A.J. Geers, H. Morales, H. Bogunovic, R. Cárdenes, A.F. Frangi, Toward an integrated management of cerebral aneurysms, Proceedings of the Royal Society A 368 (1921) (2010) 2961–2982. [5] W. Schroeder, K. Martin, W. Lorensen, The Visualization Toolkit: An Object Oriented Approach to Computer Graphics, 3rd edition, Kitware Inc., 2004. [6] L. Iba nez, W. Schroeder, L. Ng, J. Cates, The ITK Software Guide, 2nd edition, Kitware Inc., 2005, http://www.itk.org/ItkSoftwareGuide.pdf. [7] Vascular Modeling ToolKit, VMTK, Orobix, Bergamo, Italy, http://www.vmtk.org (last access: June 2011). [8] TubeTK, Kitware, Inc., http://www.vtk.org/Wiki/TubeTK (last access: June 2011). [9] 3D Slicer software, MIT Artificial Intelligence Lab and the Surgical Planning Lab at Brigham & Women’s Hospital, Harvard Medical School, http://www.slicer.org (last access: June 2011). [10] Amira, Mercury Computer Systems Inc., http://www.amiravis.com (last access: June 2011). [11] OsiriX Imaging Software, OsiriX Foundation, http://www.osirix-viewer.com (last access: June 2011). [12] CMIV CTA Plug-in for OsiriX, Center for Medical Image Science and Visualization (CMIV), Linköping University, http://www.osirix-viewer.com/CTA Plugin/index.htm (last access: June 2011). [13] @neuFuse prototype, B3C, Bologna, Italy, http://www. biomedtown.org/biomed town/AneuFuse/reception (last access: June 2011). [14] A.F. Frangi, D.R. Hose, D.A. Ruefenacht, The @neurIST project: towards understanding cerebral aneurysms, SPIE Newsroom, 2007. [15] MIMICS Innovation Suite, Materialise, Leuven, Belgium, http://www.materialise.com/mimics (last access: June 2011). [16] D.T. Gering, A. Nabavi, R. Kikinis, W.E.L. Grimson, N. Hata, P. Everett, F.A. Jolesz, W.M. Wells III, An integrated visualization system for surgical planning and guidance using image fusion and interventional imaging, in: C.A. Taylor (Ed.), Medical Image Computing and Computer-assisted Interventions, MICCAI, vol. 1679, Heidelberg, Berlin, 1999, pp. 809–919. [17] A. Rosset, L. Spadola, O. Ratib, OsiriX: an open-source software for navigating in multidimensional DICOM images, Journal of Digital Imaging 17 (3) (2004) 205–216.

[18] C. Wang, H. Frimmel, A. Persson, O. Smedby, An interactive software module for visualizing coronary arteries in CT angiography, International Journal of Computer Assisted Radiology And Surgery 3 (1–2) (2008) 11–18. [19] GIMIAS, Center for Computational Imaging and Simulation Technologies in Biomedicine, Universitat Pompeu Fabra, Barcelona, Spain, http://www.gimias.org (last access: June 2011). [20] I. Larrabide, P. Omedas, Y. Martelli, X. Planes, M. Nieber, J.A. Moya, C. Butakoff, R. Sebastián, O. Camara, M. de Craene, B.H. Bijnens, A.F. Frangi, GIMIAS: an open source framework for efficient development of research tools and clinical prototypes, in: Functional Imaging and Modeling of the Heart, vol. 5528/2009 of Lecture Notes in Computer Science, 2009, pp. 417–426. [21] M. Hernandez, A.F. Frangi, Non-parametric geodesic active regions: method and evaluation for cerebral aneurysms segmentation in 3DRA and CTA, Medical Image Analysis 11 (3) (2007) 224–241. [22] H. Bogunovic, J.M. Pozo, M.-C. Villa-Uriol, C.B.L.M. Majoie, R. Berg, H.A.F.G. van Andel, J.M. Macho, J. Blasco, L.S. Román, A.F. Frangi, Automated segmentation of cerebral vasculature with aneurysms in 3DRA and TOF-MRA using geodesic active regions: an evaluation study, Medical Physics 38 (210) (2011), http://dx.doi.org/10.1118/1.3515749. [23] H. Ujiie, H. Tachibana, O. Hiramatsu, A.L. Hazel, T. Matsumoto, Y. Ogasawara, H. Nakajima, T. Hori, K. Takakura, F. Kajiya, Effects of size and shape (aspect ratio) on the hemodynamics of saccular aneurysms: a possible index for surgical treatment of intracranial aneurysms, Neurosurgery 45 (1) (1999) 119–130. [24] R. Millan, L. Dempere-Marco, J.M. Pozo, J.R. Cebral, A.F. Frangi, Morphological characterization of intracranial aneurysms using 3-D moment invariants, IEEE Transactions on Medical Imaging 26 (9) (2007) 1270–1282. [25] J.M. Pozo, M. Villa-Uriol, A.F. Frangi, Efficient 3D Geometric and Zernike moments computation from unstructured surface meshes, IEEE Transactions on Pattern Analysis and Machine Intelligence 367 (1896) (2010) 2293–2310. [26] I. Larrabide, M. Kim, L. Augsburger, M.C. Villa-Uriol, D.A. Rüfenacht, A.F. Frangi, Fast virtual deployment of self-expandable stents: method and in-vitro validation for intracranial aneurysmal stenting, Medical Image Analysis 16 (3) (2012) 721–730. [27] A. Bernardini, I. Larrabide, L. Petrini, G. Pennati, E. Flore, M. Kim, A.F. Frangi, Deployment of self-expandable stents in aneurysmatic cerebral vessels: comparison of different computational approaches for interventional planning, Computer Methods in Biomechanics and Biomedical Engineering, in press. [28] A. Bernardini, I. Larrabide, H.G. Morales, G. Pennati, L. Petrini, S. Cito, A.F. Frangi, Influence of different computational approaches for stent deployment on cerebral aneurysm haemodynamics, Interface Focus Journal 1 (2011) 338–348. [29] M. Morales, E.E. Kim, M.C. Vivas, I. Villa-Uriol, T. Larrabide, L. Sola, A.F. Guimaraens, Frangi, How do coil configuration and packing density influence intra-aneurysmal hemodynamics? American Journal of Neuroradiology 32 (10) (2011) 1935–1941. [30] H. Morales, I. Larrabide, M. Kim, M. Villa-Uriol, J. Macho, J. Blasco, L.S. Roman, A. Frangi, Virtual coiling of intracranial aneurysms based on dynamic path planning, in: MICCAI (1), vol. 1, 2011, pp. 355–362. [31] M. Sluzewski, W.J. van Rooij, M.J. Slob, J.O. Bescós, C.H. Slump, D. Wijnalda, Relation between aneurysm volume, packing, and compaction in 145 cerebral aneurysms treated with coils, Radiology 231 (2004) 653–658. [32] I. Wanke, A. Doerfler, B. Schoch, D. Stolke, M. Forsting, Treatment of wide-necked intracranial aneurysms with a

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 806–819

[33]

[34]

[35]

[36]

[37]

self-expanding stent system: initial clinical experience, American Journal of Neuroradiology 24 (2003) 1192–1199. A.G. Radaelli, L. Augsburger, J.R. Cebral, M. Ohta, D.A. Rüfenacht, R. Balossino, G. Benndorf, D.R. Hose, A. Marzo, R. Metcalfe, P. Mortier, F. Mut, P. Reymond, L. Socci, B. Verhegghe, A.F. Frangi, Reproducibility of haemodynamical simulations in a subject-specific stented aneurysm model—a report on the virtual intracranial stenting challenge 2007, Journal of Biomechanics 41 (10) (2008) 2069–2081. J.R. Cebral, R. Löhner, Efficient simulation of blood flow past complex endovascular devices using an adaptive embedding technique, IEEE Transactions on Medical Imaging 24 (4) (2005) 468–476. M. Ford, G. Stuhne, H. Nikolov, D. Habets, S. Lownie, D. Holdsworth, D. Steinman, Virtual angiography for visualization and validation of computational models of aneurysm hemodynamics, IEEE Transactions on Medical Imaging 24 (12) (2005) 1586–1592. I. Larrabide, M.C. Villa-Uriol, R. Cárdenes, J.M. Pozo, J. Macho, L.S. Roman, J. Blasco, E.E. Vivas, A. Marzo, D.R. Hose, A.F. Frangi, Three-dimensional morphological analysis of intracranial aneurysms: a fully automated method for aneurysm sac isolation and quantification, Medical Physics 38 (5) (2011) 2439–2449. ´ I. Larrabide, A.F. Frangi, R. Cárdenes, J.M. Pozo, H. Bogunovic, Automatic aneurysm neck detection using surface Voronoi

[38]

[39]

[40]

[41]

[42]

[43]

819

diagrams, IEEE Transactions on Medical Imaging 30 (10) (2011) 1863–1876. M.L. Raghavan, B. Ma, R.E. Harbaugh, Quantified aneurysm shape and aneurysm rupture, Journal of Neurosurgery 102 (2) (2005) 355–362. J.R. Cebral, F. Mut, J. Weir, C.M. Putman, Association of hemodynamic characteristics and cerebral aneurysm rupture, American Journal of Neuroradiology 32 (2) (2011) 264–270. J. Xiang, S.K. Natarajan, M. Tremmel, D. Ma, J. Mocco, L.N. Hopkins, A.H. Siddiqui, E.I. Levy, H. Meng, Hemodynamic–morphologic discriminants for intracranial aneurysm rupture, Stroke 42 (1) (2011) 144–152. J.R. Cebral, F. Mut, J. Weir, C. Putman, Quantitative characterization of the hemodynamic environment in ruptured and unruptured brain aneurysms, American Journal of Neuroradiology 32 (1) (2011) 145–151. A.J. Geers, I. Larrabide, H.G. Morales, A.F. Frangi, Comparison of steady-state and transient blood flow simulations of intracranial aneurysms, in: Proceedings of the IEEE Engineering in Medicine and Biology Conference 2010, Buenos Aires, Argentina, 2010. N.M. Kakalis, A.P. Mitsos, J.V. Byrne, Y. Ventikos, The haemodynamics of endovascular aneurysm treatment: a computational modelling approach for estimating the influence of multiple coil deployment, IEEE Transactions on Medical Imaging 27 (6) (2008) 814–824.