Content uploaded by Bernhard Preim
Author content
All content in this area was uploaded by Bernhard Preim on Jul 16, 2020
Content may be subject to copyright.
A2822
VICTORIA
An interactive online tool for the VIrtual neck Curve and
True Ostium Reconstruction of Intracranial Aneurysms
Benjamin Behrendt1, Samuel Voss2,3, Oliver Beuing4, Bernhard Preim1,
Philipp Berg2,3, Sylvia Saalfeld1,3
1Department of Simulation and Graphics, University of Magdeburg, Germany
2Department of Fluid & Technical Flows, University of Magdeburg, Germany
3Forschungscampus STIMULATE, University of Magdeburg, Germany
4Institute of Neuroradiology, University Hospital of Magdeburg, Germany
sylvia.saalfeld@ovgu.de
Abstract. For the characterization of intracranial aneurysms, morpho-
logical and hemodynamic parameters provide valuable information. To
evaluate these quantities, the separation of the aneurysm from its parent
vessel is required by defining a neck curve and the corresponding ostium.
A fundamental problem of this concept is the missing ground truth. Re-
cent studies report strong variations for this procedure between medical
experts yielding increased interobserver variability for subsequent eval-
uations. To make further steps towards consensus, we present a web
application solution, combining a client based on HTML and JavaScript
and a server part utilizing PHP and the Matlab Runtime environment.
Within this study, participants are requested to identify the neck curve of
five virtual aneurysm models. Furthermore, they can manipulate the os-
tium surface to model the original parent artery. Our application is now
available online and easily accessible for medical experts just requiring
an internet browser.
1 Introduction
The assessment of intracranial aneurysm (IA) rupture risk increasingly depends
on morphological as well as hemodynamic parameters that are calculated based
on the aneurysm sac [1,2]. Furthermore, therapy planning requires a detailed
knowledge of the individual IA neck size to select an appropriate treatment
strategy and device, respectively [3]. In addition, a virtual separation of the
IA from the parent vessel allows for the extraction of parameters from a 3D
model rather than 2D projected images in the clinical routine that suffer from
increased user- as well as image (i.e. the viewing angle of the 2D projections)
dependency [4].
However, the separation between parent vessel and aneurysm sac (i.e. the
ostium) strongly varies, which might lead to insufficient analyses and in conse-
2 Behrendt et al.
quence to unreliable conclusions. The separation of the IA from the healthy par-
ent vessel is often realized using a (planar) cut-plane [5,6], which might be error-
prone for complex IA shapes. In previous work, we developed a semi-automatic
extraction of an anatomical, bent neck curve [7], but the question about the
correct neck curve remains unanswered.
Therefore, VICTORIA (VIrtual neck Curve and True Ostium Reconstruc-
tion of Intracranial Aneurysms) aims at a standardization of IA neck curve and
ostium reconstruction. For the definition of these neck curves and ostia, highly
experienced physicians are required. Since their availability for user studies is
limited due to the clinical work load, we developed an easy accessable frame-
work solely requiring a web browser. Thus, we can gather expert-knowledge
from physicians as well as biomedical engineers. The study has been successfully
launched at https://VICTORIA.cs.ovgu.de/.
2 Materials and methods
Within the VICTORIA study, participants are requested to identify the neck
curve of patient-specific IA models extracted from 3D digital subtraction data.
The extraction of surface models is described in our previous work [8].
2.1 VICTORIA web application
The study is conducted using a specialized web application, consisting of two
parts: a client part (written in HTML and JavaScript), and a server part (written
in PHP and using Matlab runtime environment). The VICTORIA study can be
accessed via the internet using a web browser. Between client and server, the
data is exchanged in the JSON format. The server stores user-submitted data
in a relational database, thus allowing for easy sorting and filtering of the data,
see Figure 1.
In order to keep the motivation to participate high, the user is directly pre-
sented with five datasets and two tasks, respectively. For both tasks, a surface
Fig. 1. Illustration of the different parts of the VICTORIA web application
VICTORIA 3
mesh of an IA and its parent vessel is shown. The rendering is performed using
the WebGL2 API, thus allowing it to run in most current desktop browsers. The
mesh is illuminated using the Phong lighting model to increase shape percep-
tion. The user has limited control over the camera (rotation, zoom and panning),
ensuring that the neck region is always visible and centered in the image.
2.2 Neck curve definition
First, the user draws a neck curve onto the vessel mesh by selecting surface
vertices, see Figure 2. To connect the selected vertices into a circular path, the
surface triangle mesh is interpreted as a bidirectional graph. The shortest paths
between the selected points are then calculated using the A* algorithm by Hart
et al. [9]. However, the A* algorithm is only designed to find the shortest path
between two points, not the shortest path connecting a list of unordered points.
To prevent having to force users to select the points in order and thus sacrificing
usability, the list of points has to be sorted automatically. Here, a distance matrix
stores the length of the shortest path between all points. Whenever a new point
is added to the matrix, the distances to all other points are determined using
A*. To sort the list of points, we begin with a list containing only one randomly
chosen point. As long as there are points that have not been added to the list,
we take the last entry, look up the closest point (based on the distance matrix)
Fig. 2. The different steps of the ostium definition. (A) 3D visualization of the
surface model with WebGL. (B) The user can interactively select points on the
aneurysm surface. The points are automatically selected (see also inlay). (C) If
the points are close to each other, the neck curve is automatically closed. (D) In
the second step, an automatic ostium triangulation is provided. (E) The user can
hover over the points and the active point is highlighted. (F) the point can be
moved (including a reduced movement of its neighbors) until the user is satisfied
with the ostium shape.
4 Behrendt et al.
that has not been added yet, and add it to the back of the list. After all points
have been processed, the resulting list contains the shortest path connecting all
neck points selected by the user. The last and first point in the list will only
be connected if their distance is shorter than the longest path between any of
the other adjacent points in the list. This prevents the neck curve from closing
before the user has finished adding all desired points. In case the resulting neck
curve does not match the users’ expectation additional points can be added.
The sorting and path finding algorithm are implemented in JavaScript and
are performed entirely on the client side. Even on less powerful devices, they run
interactively without any noticeable delay.
2.3 Ostium creation and manipulation
After submitting the neck curve, the server calculates the associated ostium sur-
face mesh. This step is implemented in Matlab, motivated by previous work [7] as
well as to provide additional geometric functionalities and to spare the client’s
PC performance. To ensure the general applicability, we use the Matlab run-
time environment and provide it at our linux server. The Matlab script receives
the previously selected surface meshes and downsamples them to a predefined
threshold. The process is illustrated in Figure 3. For the illustration, the neck
curve points are resampled to 32 points. The ostium triangulation is analytically
defined and centered around the origin (z-coordinates equal 0). Next, the ana-
lytically defined border points are replaced by the neck curve points and simple
Laplacian smoothing (50 iterations, σ= 1.0) is applied to all vertices except the
border points, resulting in the smooth ostium surface (Fig. 3).
The second task consists of the adjustment of the previously calculated os-
tium surface. This is important to better approximate the inflow area for further
postprocessing steps in order to better approximate the parent vessel’s original
geometry. During this task, only the backfaces of the vessel are rendered to reveal
the ostium. The user can grab and drag any vertex of the ostium surface that is
not part of its border. To keep the interaction as simple and straight-forward as
possible, the vertices can only be moved in the direction of the average surface
Fig. 3. Illustration of the ostium triangulation. The neck points are resampled
to 32, p1-p32 (left). The pre-defined triangulation is centered around 0 with
border points q1-q32 (center). Next, q1-q32 are replaced by p1-p32 and all
points undergo Laplacian smoothing yielding the triangulated ostium surface.
VICTORIA 5
normal of the ostium. As moving one vertex at a time would be tedious, dragging
one vertex by a distance of valso moves the surrounding vertices by a distance of
vnbased on their normalized Euclidean distance dnto the original vertex. The
parameter acontrols the size of the affected area. We empirically determined a
value of 1
3to result in an interaction that feels natural.
vn=v·(max(0, dn−(1 −a)) ∗a−1)0.75
The second task can be considered optional, as it is possible for the user to
directly accept and submit the ostium surface generated by the server without
performing any modifications.
Finally, we implemented a registration form combined with a questionnaire.
Users enter their name, e-mail address, occupation and employer, as well as
answer questions about their experience with IAs. Thus, an examination of dif-
ferences between user groups (e.g., physicians vs. engineers) can be conducted
in the future.
3 Results
The VICTORIA study was implemented at the Otto-von-Guericke University,
Magdeburg, Germany, and can be easily assessed at https://VICTORIA.cs.
ovgu.de/. We tested this interactive survey sucessfully with three medical coop-
eration partners. They requested more hints and icons regarding the controlling
of the application which we included. The neck curve definition requires ap-
proximately two minutes per case and less than ten minutes overall. The study
requires the use of Mozilla Firefox or Google Chrome. Other, less commonly used
browsers, such as Apple Safari or Microsoft Edge do not support the required
WebGL technologies. The evaluation of the submitted ostia segmentations is not
part of this work but ongoing research. In this work, the implementation of the
web-based study is presented and the prototype does fulfill its purpose.
4 Discussion
Currently no ostium ground truth is available, yet it is required for many post-
processing steps of IA models. We presented an approach that allows for an easy
participation of international field experts and physicians by utilizing a combi-
nation of JavaScript, PHP and Matlab-programs, including 3D visualizations,
editing and modification options.
Based on a ground truth ostium definition, a precise morphological evalu-
ation of the 3D IA shape is further promoted, and it is highly beneficial for
the quantification of hemodynamic flow simulations [10]. Furthermore, a 3D
neck curve determination and subsequent parameter evaluation is superior to 2D
analysis [4]. Particularly, since relevant blood flow parameters that are associ-
ated with rupture (e.g., normalized wall shear stress, shear concentration index,
oscillatory shear index [11]) need to be calculated with high accuracy, wrong
6 Behrendt et al.
aneurysm-vessel-separation or high user-dependency can lead to clear variations
regarding the analysis.
Furthermore, we expect our survey results to be utilized as ground truth
data for deep learning-based ostium extraction approaches, e.g. similar to the
MeshCNN network [12]. Finally, our architecture can be easily adapted to other
medical image processing questions that require 3D models and user interaction.
Acknowledgement. This study was funded by the Federal Ministry of Ed-
ucation and Research in Germany within the Forschungscampus STIMULATE
(grant number 13GW0095A) and the German Research Foundation (grant num-
ber SA 3461/2-1, BE 6230/2-1).
References
1. Detmer FJ, Chung BJ, Jimenez C, et al. Associations of hemodynamics, mor-
phology, and patient characteristics with aneurysm rupture stratified by aneurysm
location. Neuroradiology. 2019;61(3):275–284.
2. Niemann U, Berg P, Niemann A, et al. Rupture Status Classification of Intracranial
Aneurysms Using Morphological Parameters. In: Proc. of IEEE Symposium on
Computer-Based Medical Systems (CBMS); 2018. p. 48–53.
3. Paliwal N, Tutino V, Shallwani H, et al. Ostium ratio and neck ratio could pre-
dict the outcome of sidewall intracranial aneurysms treated with flow diverters.
American Journal of Neuroradiology. 2019;40(2):288–294.
4. Wong SC, Nawawi O, Ramli N, et al. Benefits of 3D rotational DSA compared with
2D DSA in the evaluation of intracranial aneurysm. Acad Radiol. 2012;19(6):701–
707.
5. Lauric A, Baharoglu MI, Malek AM. Ruptured status discrimination perfor-
mance of aspect ratio, height/width, and bottleneck factor is highly dependent
on aneurysm sizing methodology. Neurosurgery. 2012;71(1):38–46.
6. Xiang J, Natarajan SK, Tremmel M, et al. Hemodynamic–morphologic discrimi-
nants for intracranial aneurysm rupture. Stroke. 2011;42(1):144–152.
7. Saalfeld S, Berg P, Niemann A, et al. Semiautomatic neck curve reconstruction for
intracranial aneurysm rupture risk assessment based on morphological parameters.
Int J Comput Assist Radiol Surg. 2018;13(11):1781–1793.
8. Glaßer S, Berg P, Neugebauer M, et al. Reconstruction of 3D Surface Meshes for
Bood Flow Simulations of Intracranial Aneurysms. In: Proc. of the Annual Meeting
of the German Society of Computer- and Robot-Assisted Surgery (CURAC); 2015.
p. 163–168.
9. Hart PE, Nilsson NJ, Raphael B. A formal basis for the heuristic determination
of minimum cost paths. IEEE transactions on Systems Science and Cybernetics.
1968;4(2):100–107.
10. Berg P, Beuing O. Multiple intracranial aneurysms: a direct hemodynamic com-
parison between ruptured and unruptured vessel malformations. Int J Comput
Assist Radiol Surg. 2018;13(1):83–93.
11. Cebral JR, Mut F, Weir J, et al. Quantitative characterization of the hemody-
namic environment in ruptured and unruptured brain aneurysms. AJNR Am J
Neuroradiol. 2011;32(1):145–151.
12. Hanocka R, Hertz A, Fish N, et al. MeshCNN: A Network with an Edge. ACM
Trans Graph. 2019;38(4):90:1–90:12.
E2822
