EVALUATING VISUALISATION AND NAVIGATION TECHNIQUES

FOR INTERPRETATION OF MRA DATA

B. W. van Schooten

, E. M. A. G. van Dijk

E. V. Zudilova-Seinstra

, P. J. H. de Koning

and J. H. C. Reiber

Human Media Interaction, University of Twente, Enschede, The Netherlands

Section Computational Science, University of Amsterdam, The Netherlands

LKEB, Leiden University Medical Center, The Netherlands

Keywords:

Volume visualization, 3D navigation, Radiology, MRA.

Abstract:

We argue that a more systematic treatment of the many existing options for medical volume visualisation is

desirable. We show that combining the most common medical visualisation and navigation techniques in a

systematic way leads to a meaningful set of interesting and sometimes novel UI techniques. We also propose

a technique for using generated data and tasks suited to non-medical users for conducting user experiments.

We evaluate the UI techniques qualitatively to arrive at a set of promising techniques for future research.

1 INTRODUCTION

The domain of magnetic resonance angiography

(MRA) involves computer-aided interpretation of 3D

(greyscale) volumes, using specialised visualisation

techniques. In this domain, user interfaces (UIs) are

still designed in a rather ad hoc way. There are few

comparative usability evaluations between different

visualisations, or evaluations taking a broader look at

for example navigation or editing techniques. We pro-

pose an experimental framework enabling more sys-

tematic experiments, including a taxonomy of medi-

cal interpretation tasks, and combinations of visual-

isation and navigation techniques, based on known

medical practice. The tasks we will focus are veriﬁca-

tion of automatic analyses of blood vessels (Zudilova

and Sloot, 2005), in particular the vessel centerline (a

curved line going through a particular vessel), and the

vessel segmentation (a mesh describing the surface of

a particular vessel).

Medical experts and expert-annotated medical

data are hard to get, so we argue that using computer-

generated data and non-medical users greatly facili-

tates usability experiments. This approach is, how-

ever, only rarely found (Moise et al., 2005). We

designed an algorithm for generating vascular-like

structures and mock automatic analyses with artiﬁ-

cally introduced errors. Tasks and data are simpliﬁed

so they can be performed by non-medical users.

2 THE FRAMEWORK

The framework can be subdivided into task, visuali-

sation method, and navigation method. We consider

the following subtasks:

• Interpreting the overall 3D structure and locating

a particular vessel in 3D space (overview).

• Viewing ﬁne greyscale details at relevant lo-

cations around the centerline and segmentation

(greyscale).

• Seeing deviations of the centerline (deviate C)

and segmentation (deviate S) from the greyscale

data, indicating problem areas.

2.1 Visualisation

The most commonly used MRA visualisations are de-

scribed below; see also ﬁgure 1. First we have the

slice visualisations, only showing a slice of the data:

• Cross-section - shows the volume data that in-

tersects an arbitrary plane. One particular cross-

section that is often used is a close-up cross-

405

W. van Schooten B., M. A. G. van Dijk E., V. Zudilova-Seinstra E., J. H. de Koning P. and H. C. Reiber J.

EVALUATING VISUALISATION AND NAVIGATION TECHNIQUES FOR INTERPRETATION OF MRA DATA.

DOI: 10.5220/0001797804050408

In Proceedings of the Fourth International Conference on Computer Graphics Theory and Applications (VISIGRAPP 2009), page

ISBN: 978-989-8111-67-8

section of the vessel at a particular point on the

centerline.

• CPR (Curved Planar Reformatting) (Kanitsar,

2004) - a curved cross-section following the cur-

vature of the vessel.

Secondly we have the volume visualisations

showing the entire volume:

• DVR (Direct Volume Rendering) (Mueller et al.,

2005). This involves rendering the volume as

a semi-transparent 3D object, representing the

greyscale values as densities. Various enhance-

ments exist, such as shading and colour mapping.

• Isosurface (Preim and Oeltze, 2007). This in-

volves converting the greyscale volume to a plain

mesh surface by connecting the points which have

a particular greyscale value. A lot of the data is

lost in this visualisation, however.

• MIP (Maximum Intensity Projection) (Preim and

Oeltze, 2007). This involves creating a 2D pro-

jection in which each pixel value is the bright-

est value found in a ray cast into the scene from

that point. Unlike DVR or isosurface, there is

no visual cue indicating which parts of the visi-

ble structure are in front or behind others.

Figure 1: Top left: CPR with centerline. Top right: DVR.

Bottom left: isosurface with segmentation and centerline.

Bottom right: MIP with centerline.

In cross-section and CPR, the centerline can be

shown by projecting it onto the slice plane, and the

segmentation is typically shown by only showing the

lines that intersect the slice plane. In the volume

views, both centerline and segmentation can be shown

in an obvious manner. Additionally, occlusion cues

can be used for displaying centerline and segmenta-

tion in some cases. Isosurface enables the easiest oc-

clusion cues, but they are also possible with properly

thresholded DVR.

It is also possible to combine multiple visualisa-

tions by integrating them into a combined visualisa-

tion. We came up with the combinations below (see

also ﬁgure 2). As far as we know, these combinations

form novel medical visualisation techniques, except

the volume-volume combination.

• Isosurface-slice - isosurface with any slice visu-

alisation as integrated 3D ﬁgure. To compensate

for the incompleteness of isosurface data, isosur-

face is integrated with a slice visualisation, draw-

ing the slice as an oriented plane in the 3D view.

• CPR-volume - CPR with any volume visualisa-

tion, with the volume view simply drawn over the

CPR. A particular centerline point is chosen for

which both views are made to coincide graphi-

cally. Additionally, the orientation of the 3D view

around the centerpoint is equal to that of the CPR.

• Volume-volume - any volume visualisation with

any volume visualisation. One visualisation is

drawn regularly, one or more others are drawn

as oriented planes inside the main visualisation.

For example, there may be three separate oriented

planes on the back faces of the cube surrounding

the rendered volume. This type of visualisation

is used by the QMRAVWI application (de Kon-

ing et al., 2003) developed at Leiden University

medical center (LUMC).

• Isosurface-DVR - isosurface with DVR as inte-

grated 3D ﬁgure.

Figure 2: Top left: isosurface-slice (using CPR for slice).

Top right: CPR-isosurface. Bottom left: volume-volume

(DVR-MIP). Bottom right: isosurface-DVR.

2.2 Navigation

Designing appropriate navigation techniques can be

difﬁcult and needs to be studied separately. For 2D

representations, navigation consists of panning and

zooming. For 3D representations, this includes 6 de-

grees of freedom (6 DOF) camera manipulation, plus

optional zooming. The 6 DOF can be separated into

3 DOF positioning and 3 DOF rotation.

We distinguish several common types of (posi-

tion) navigation:

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

406

Table 1: Table summarising all navigation types and their decomposition into different degrees of freedom. The valid types

are given in boldface. Legend: 1D, 2D, 3D: positioning in resp. 1,2,3 dimensions; rot: 3DOF rotation; axis: 1DOF rotation

around vessel axis. Note that zooming is possible and useful in all cases, and adds one extra DOF to the total.

visualisation → 3D cross-section CPR

position↓ rotation→ Free Centerline Flyby Flythrough (none) (axis)

Free 3D + rot not useful not useful not useful 2D 2D + axis

Centerline 1D + rot 1D + rot 1D + axis 1D 1D 1D + axis

Pickray 2D + rot pick confusing cannot pick cannot pick N/A N/A

1. centerline-based navigation: the user can cycle

through the points of the centerline forwards and

backwards. The view is centered around the cur-

rently selected centerline point (the focus point).

2. free navigation: The position can be determined

freely rather than being ﬁxed to a centerline point.

3. pickray navigation: A coordinate is selected by

clicking on the centerline or vessel wall. The cam-

era will navigate to the selected point.

Rotation in 3D can be treated separately. There

are several obvious options:

1. free rotation angle

2. rotation angle always follows angle of centerline.

This is a relatively novel navigation technique for

MRA. We distinguish several variants:

• centerline: the user can freely specify a relative

rotation angle.

• ﬂyby: the camera is perpendicular to the center-

line, looking at the vessel from “above”. This

orientation is less useful for isosurface-DVR

or isosurface-slice because the isosurface ob-

scures what is inside the vessel.

• ﬂythrough: the camera is oriented parallel to

the centerline, and we zoom in close, like vir-

tual angioscopy (Giachetti et al., 2001). This

orientation is less useful for DVR or MIP.

We summarise all possible combinations of rota-

tion and position navigation in table 1. The table sum-

marises which ones are valid and potentially useful,

and how many DOF each requires.

3 RESULTS

Most combinations of the given visualisation and nav-

igation techniques form meaningful and sometimes

novel interaction techniques. We have made a qual-

itative assessment about the suitability of each com-

bination for each task, see table 2. For each visual-

isation, a rating was given in the range not possible

to excellent, and we summarise what kind of navi-

gation was necessary to perform the task. With help

of this assessment, we can select the most promising

techniques for future user experiments. Based on the

table, we conclude that the most promising visualisa-

tion techniques overall are isosurface-slice, volume-

volume, and CPR-volume.

ACKNOWLEDGEMENTS

This research is funded by the NWO/VIEW project

“A Multi-modal Visualization Environment for Inter-

active Analysis of Medical Data” (N 643.100.602).

REFERENCES

de Koning, P. J. H., Schaap, J. A., Janssen, J. P., van der

Geest, R. J., and Reiber, J. H. C. (2003). Automated

segmentation and analysis of vascular structures in

magnetic resonance angiographic images. Magnetic

Resonance in Medicine, 50(6):1189.

Giachetti, A., Tuveri, M., and Zanetti, G. (2001). Measur-

able models of abdominal aortic aneurysm on the web.

Stud. Health Technol. Inform., 81:158–160.

Kanitsar, A. (2004). Curved Planar Reformation for Ves-

sel Visualization. PhD thesis, Institute of Computer

Graphics and Algorithms, Vienna University of Tech-

nology, Favoritenstrasse 9-11/186, A-1040 Vienna,

Austria.

Moise, A., Atkins, M. S., and Rohling, R. (2005). Evalu-

ating different radiology workstation interaction tech-

niques with radiologists and laypersons. Journal of

Digital Imaging, 18(2):116–130.

Mueller, D. C., Maeder, A. J., and O’Shea, P. J. (2005). En-

hancing direct volume visualisation using perceptual

properties. In Proc. SPIE, Vol. 5744, 446, pages 446–

454.

Preim, B. and Oeltze, S. (2007). Visualization in Medicine

and Life Sciences, chapter 3D Visualization of Vascu-

lature: An Overview, pages 39–60. Springer Verlag.

Zudilova, E. V. and Sloot, P. M. A. (2005). Bringing com-

bined interaction to a problem solving environment

for vascular reconstruction. Future Gener. Comput.

Syst., 21(7):1167–1176.

EVALUATING VISUALISATION AND NAVIGATION TECHNIQUES FOR INTERPRETATION OF MRA DATA

407

Table 2: Table summarising all visualisations. Legend: Task type: C=centerline; S=segmentation. 3D cues: C=centerline

occlusion or S=segmentation occlusion. Navigation type: F=free, C=centerline-based, P=pickray. F=C means that F and C

are very similar. Rotation type: F=free, C=follow-centerline, B=ﬂyby, T=ﬂythrough. The indications (b) and (t) mean that

resp. the ﬂyby and ﬂythrough orientations are less useful. Note that good and excellent ratings are printed in boldface.

visualisation 3D cues navigation rotation

cross-section C – F=C – – – – –

overview: not possible

greyscale: fair; well visible but extensive centerline-based navigation required.

deviate: good; excellent for both C and S but extensive centerline-based navigation required.

CPR – – F=C – – – – –

overview: poor, rotation around vessel axis required

greyscale: good, rotation around vessel axis required

deviate C: excellent, error can be seen immediately as a veering away or gap in the vessel

deviate S: good, only for too-wide or too-narrow type errors, rotation around vessel axis required

DVR C S F C P F C B (t)

overview: fair, little navigation required

greyscale: fair, rotating and zooming required

deviate: fair, rotating and zooming required

isosurface C S F C P F C B T

overview: good, little navigation required

greyscale: not possible

deviate C: excellent, occlusion cues help see if the centerline is in front of or behind the vessel,

so little navigation required

deviate S: good if isosurface is representative of the segmentation location, rotating and zooming

required

MIP – – F C P F C B (t)

overview: fair, rotation required, sometimes hard to see what’s in front or behind

greyscale: fair, rotating and zooming required

deviate: fair, rotating and zooming required

isosurface-slice C S F C P F C (b) T Note: “slice” may be cross-section or CPR

overview: fair, little navigation required. For CPR, the slice plane may make visual scene harder

to interpret

greyscale: good. Requires close-up view and centerline-based navigation.

deviate C: excellent, occlusion cues help see if the centerline is in front of or behind the vessel,

so little navigation required

deviate S: good if isosurface is representative of the segmentation location, navigation required

CPR-volume C S – C – – – B – Note: “volume” is DVR, MIP, or isosurface

overview: fair to good, little navigation required

greyscale: good, navigation same as particular volume visualisation method

deviate: good to excellent, similar to CPR

volume-volume C S F C P F C B T Note: “volume” is DVR, MIP, or isosurface

overview: excellent, the cube backface projections help locate a point precisely, except when

zooming in closely.

greyscale: fair, rotating and zooming required

deviate: fair to excellent, inherits properties from both visualisation methods used

isosurface-DVR C S F C P F C (b) (t)

overview: good, little navigation required

greyscale: fair, though not as good as DVR by itself, rotating and zooming required

deviate: good, similar to isosurface, though not as good.

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

408