DAAPMed: A Data-aware Anchor Point Selection Tool for Medical

Models in VR Environments

Eva Moncl

us Lahoya, Pere-Pau V

azquez Alcocer and Isabel Navazo

Alvaro

ViRVIG-LSI, Universitat Polit

ecnica de Catalunya, c. Jordi Girona 1–3, Edif. Omega, 08034 Barcelona, Spain

Keywords:

3D interaction, 3D selection, Medical Visualization, Virtual Reality.

Abstract:

There is a number of problems where the analysis of medical datasets requires the selection of anchoring points

in 3D space, such as the measurement of anatomical structures (i.e. lengths of bones), pathological structures

(i.e. tumors), and the measurement of other elements such as the air contents in the lungs or the gut. Previous

research indicates that measurement tasks can be usually carried out more efﬁciently in VR environments than

in desktop-based systems. However, there is a lack of tools for measurement support for medical models in

VR environments. This paper presents a new VR-based interaction technique, Data-Aware Anchor Points for

Medical models (DAAPMed), specially focused on the efﬁcient selection of 3D points in datasets rendered

using methods with semi-transparency such as Direct Volume Rendering. We will show that our method is

effective, precise, and reduces the amount of movements required to set the anchor points as compared with

other classical techniques based on clipping planes.

1 INTRODUCTION

In medical applications, the quantitative analysis of

spatial relations between structures is crucial for

many tasks such as diagnosis, treatment and surgical

planning, and documentation. These measurements

include, among others, the extension of pathological

structures or the distance between pathological struc-

tures and structures at risk (blood vessels). In the ﬁeld

of Neurosurgery, for instance, distance between the

brain surface and the ventricles is an important pa-

rameter that may determine the surgical trajectory. In

a different ﬁeld, digestologists are interested in per-

forming a morpho-volumetric analysis of the amount

of air in different organs such as the gut or the lungs,

for diagnostic purposes. In order to carry out these

quantitative analyses, researchers set several anchor

points using medical images as a support (Accarino

et al., 2009). However, the use of these 2D images

does not facilitate the perception of the relative posi-

tion of the structures, and as a consequence, it is often

rough for accurately locating anchor points and thus

obtaining precise 3D magnitudes.

Advances in volume visualization allow for the

3D reconstruction and analysis of anatomical struc-

tures from a stack of intensity-based images acquired

from, usually, CT or MRI modalities. Initial al-

gorithms identiﬁed and extracted the isosurfaces of

the anatomical structures as triangle meshes. This

process is time-consuming and loses contextual in-

formation. Later methods directly render the vol-

ume (Direct Volume Rendering, or DVR) by assign-

ing color and opacity to the samples as a function

of its density by using pre-deﬁned transfer functions.

Semi-transparencies provide a means to increase the

amount of information visible to the users, and fa-

cilitate establishing spatial relationships between el-

ements such as the skin and the bones. This led the

development of new techniques for anchor point deﬁ-

nition in desktop-based applications. However, occlu-

sions still remain as a problem for the selection. This

is often addressed with the introduction of clipping

planes showing the volume cut mapped on them.

Reitinger et al. (Reitinger et al., 2006) found that

measurement tasks can be carried out more efﬁciently

in a Virtual Reality environment than in a desktop

setup. The cost reduction of VR systems and GPUs

is helping the introduction of such systems in surgi-

cal planning and diagnose. Stereo vision facilitates

the perception of the relative position of anatomical

structures, although occlusion remains as a research

problem.

The purpose of this paper is to provide an easy-

to-use tool for the fast and accurate selection of 3D

points on the implicitly deﬁned surfaces of anatomi-

cal structures present in a volume dataset in a virtual

308

Monclús Lahoya E., Vázquez P. and Navazo Álvaro I..

DAAPMed: A Data-aware Anchor Point Selection Tool for Medical Models in VR Environments.

DOI: 10.5220/0004287603080317

In Proceedings of the International Conference on Computer Graphics Theory and Applications and International Conference on Information

Visualization Theory and Applications (GRAPP-2013), pages 308-317

ISBN: 978-989-8565-46-4

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

environment. It is important to note that we are not

interested in selecting a concrete structure, but a point

on it, without any previous surface extraction nor seg-

mentation process. Additionally, the occlusion prob-

lem remains, as well as the fact that a VR selection

environment is not familiar to medical experts.

The contributions of our paper are threefold:

• DAAPMed: An anchor point selection tool suit-

able for medical models in VR environments that

enriches the selection ray trajectory with the infor-

mation of candidate anchor points. These are au-

tomatically computed by performing an on-the-ﬂy

isosurface detection along the ray (see Figure 1).

• A series of visual cues that provide feedback on

the ray position through the use of mirror views

and supporting planes.

• A user study that analyzes the accuracy and per-

formance of the selection, and compares with the

classical selection using clipping planes.

The implementation of all these components (both

GPU and CPU) guarantees real-time feedback and in-

teraction. This is an important issue in VR environ-

ments which require rendering the model twice.

The user study showed that our technique is effec-

tive, with an accuracy compared to a selection tool

in a desktop-based application with a mouse, and

more efﬁcient. Furthermore, it also reduces the ef-

forts (hand displacements) and time required for the

selection as compared with a classical clipping plane

technique in a Virtual Reality environment.

Figure 1: User interacting with a model using the

DAAPMed metaphor.

2 RELATED WORK

In a pioneer work, Hinckley et al. (Hinckley et al.,

1994a) proposed a 3D user interface for pre-operative

neurosurgical planning based on the physical manip-

ulation of familiar real-world objects (head, cutting-

plane and stylus-shaped props) in free space to access

and manipulate a virtual model. This approach offers

the possibility of selecting anchor points in a brain

model. They use a clipping plane to access occluded

or interior points in the brain and then select anchor

points on it as the intersection of the linear trajectory

deﬁned by the stylus and the cutting-plane. Qi and

Martens (Qi and Martens, 2005) also focused in de-

signing a 3D (tangible) user interface for the manip-

ulation of a clipping plane into a volume dataset in a

small size VR system (based on a 14” display), but

their system does not provide any selection mecha-

nism. More recently, Song et al. (Song et al., 2011)

propose the use of a touch mobile for manipulating

(positioning and orienting) a slicing plane. Since their

objective is exploration, they do not address the prob-

lem we are concerned with: anchor point selection.

Preim et al. (Preim et al., 2002) introduced a set of

applicable tools for the computation of distances, an-

gles, and volumes in 3D visualizations. The tools

are 3D virtual objects such as a distance line, a

ruler and angular measurements that are manipulated

using the mouse in a desktop-platform. They al-

low to determine anchor points on the surface of

the pre-segmented anatomical structures. Rossling et

al. (R

ossling et al., 2010) proposed a method for the

automatic determination of different distance−based

measures (shortest distance, diameters and wall thick-

ness) also on segmented anatomic structures. The

necessity of this kind of tool is justiﬁed by the fact

that manual distance calculation is tedious and impre-

cise in single 2D slices, and although it is possible

to achieve an accurate result in 3D, it would also be

tiresome. However, completely automatic measure-

ments are difﬁcult to generalize due to the great va-

riety of problems and anatomical structures. Notice

that both previous approaches (Preim et al., 2002;

ossling et al., 2010) work on triangle mesh repre-

sentations, so a surface extraction process is needed

previously to use them. Moreover, they always se-

lect the nearest visible point and they do not deal with

semi-transparent models.

Reitinger et al. (Reitinger et al., 2006) presented a 3D

measurement toolkit developed for liver surgery espe-

cially tailored for a VR platform. Their measurements

include distance, volume, and angles. Their evalua-

tion indicated that VR-based measurement tools have

a sufﬁcient beneﬁt compared to 2D desktop-based

systems in terms of task completion time. In terms

of accuracy, slightly better results in most of the tasks

were achieved. The anatomical structures models

(liver, vessels,...) are computed through segmentation

DAAPMed:AData-awareAnchorPointSelectionToolforMedicalModelsinVREnvironments

309

from CT scans and they are represented by opaque tri-

angle meshes where the user may select points by us-

ing a virtual pencil. Hagerdorn et al. (Hagedorn et al.,

2007) proposed a set of tools for performing measure-

ments in a virtual reality visualization environment.

A 3D Rubberbanding line for selecting free points in

the scene is proposed. They use clipping planes for

accessing interior parts of the volume dataset. Their

scene is also composed by triangle meshes.

Segmentation and surface extraction are time con-

suming operations. To overcome this problem, Has-

treiter et al. (Hastreiter et al., 1998) suggest direct

volume rendering of the entire data volume, giving

insights to interior and superimposed information. In

order to inspect interior structures, independent clip-

ping planes provide an intuitive way to virtually cut

off parts of the volume data set. Then, anchor points

can be interactively placed on the clipping planes.

Gallo et al. (Gallo et al., 2008) present a Virtual Real-

ity system for the exploration of volume datasets us-

ing a Wiimote. Apart from the basic interaction tech-

niques for navigating they propose a mechanism of

selection of points based on the classical ray-casting

technique adding the mechanism of ﬁshing reel in

which the users can move the cursor closer or farther

away by using two buttons in order to accurately lo-

cate a mark. Unfortunately, points’positions are not

aware of the isosurfaces and no visual cue is used to

reveal the cursor when it is moved into an occluded

region.

Many researchers have investigated 3D object se-

lection techniques for general -non medical- VR ap-

plications where the objects are represented by polyg-

onal meshes; they focused on improving the user ex-

perience in this kind of tasks (Bowman et al., 2004).

Ray-based techniques (Mine, 1995) have shown a bet-

ter performance than point-based techniques. These

former approaches are usually based on a cone or a

ray. Since our interest is on accurate anchor point

selection, we only consider ray-based tools. In or-

der to solve the inherent problem of multiple inter-

section candidates, several disambiguation techniques

have been proposed. Olwal et al. (Olwal and Feiner,

2003) describe the ﬂexible pointer, a ray cursor tech-

nique that allows users to point around objects with

a curved arrow, to select fully or partially obscured

objects. It is important to note that most of these VR

selection metaphors are focused on selection and ma-

nipulation of objects (not points) in populated scenar-

ios, and thus they were not specially concerned about

accuracy in point selection.

Grossman et al. (Grossman and Balakrishnan,

2006) explored 3D selection techniques for volumet-

ric displays and proposed new ray cursor techniques

which provide disambiguation mechanisms for multi-

ple intersected targets. The Depth Ray tool augments

the ray cursor with a depth marker. The position of

this marker is changed dynamically moving the hand

forwards and backwards. As the hand also controls

the placement of the ray cursor, the two phases could

potentially interfere with each other. To solve it, they

propose the Lock Ray, a similar technique, where se-

lection and disambiguation phases are carried out se-

quentially, in a two-step process. First the user selects

the ray. Once it is locked, the depth marker appears.

Then, forward and backward hand movements ﬁx the

depth marker and the intersected target closest to it is

highlighted in red indicating that it can be selected by

releasing the button.

Our approach also decouples the selection and dis-

ambiguation phases in two sequential steps (Gross-

man and Balakrishnan, 2006), though using a cy-

cling method for candidate selection (Hinckley et al.,

1994b). In contrast to these previous works, we

work directly with the captured volume dataset (using

DVR) without any kind of costly preprocess to extract

the isosurfaces.

3 DAAPMED METAPHOR

Our objective is a user-friendly, efﬁcient, accurate

anchor point selection technique that facilitates get-

ting measurements in VR environments with medical

models. We also require ease of use and limiting the

amount of effort the user has to perform. In contrast

to normal desktop environment, where working with

a mouse allows the user to rest the arm, in 3D en-

vironments the user usually does ample arm move-

ments and has no surface to rest. With the objec-

tive of facilitating the integration with the special-

ists’ clinical work, we directly use the captured vol-

ume dataset rendered using DVR with a transfer func-

tion that shows semi-transparent and opaque struc-

tures (see Figures 1 and 2).

As a ﬁrst approach, we extended and adapted to

3D the classical desktop point selection using a clip-

ping plane (see Figure 3). However, as shown in our

user study, this metaphor requires quite a long time

and large movements from the user, because the cor-

rect deﬁnition of a clipping plane suitable for poste-

rior point selection is difﬁcult. In order to overcome

these limitations we have developed a ray-based ap-

proach that uses the data information to facilitate the

ray setting and point selection, reducing time and dis-

placements.

GRAPP2013-InternationalConferenceonComputerGraphicsTheoryandApplications

310

Figure 2: DAAPMed metaphor: A ray is used for selection,

and a couple of supporting planes help the user to locate

it in relation to the 3D structures. Potential anchor points,

represented by colored small spheres, are computed as the

intersections of the ray with the isosurfaces. Finally, Helper

Views provide a better perception of the ray position as well

as aids disocluding interior candidate points. Notice that the

bottom view shows that the large orange sphere is hollow.

3.1 Data aware 3D Selection Metaphor

Studies have shown that ray-based selection tech-

niques often result in faster selection times than hand

extension techniques in VR environments (Bowman

et al., 1999). Unfortunately, as commented previ-

ously, ray cursor techniques have an inherent prob-

lem: the ray may intersect multiple objects. A na

ıve

approach simply selects the ﬁrst target which is in-

tersected; however, it becomes very difﬁcult or even

impossible to select occluded points. Thus a more

sophisticated method is required. The DAAPMed

metaphor has three main components (see Figure 2):

• Ray Cursor Tool. It casts a pointing ray through

the volume. The ray path visualization is enriched

with the candidate selection points and its sup-

porting planes, which provide a better insight of

its position and orientation.

• Helper Views. We provide two views that help

to understand the position of the ray inside the

volume. This extra-visualization is inspired by

the Magic Mirrors View (K

onig et al., 1999), but,

instead of showing the whole model, our view

shows the model clipped by a plane that enables

the possibility of showing the ray trajectory with-

out any occlusion.

• Disambiguation Mechanism. Once the ray is

locked, we may select among the different in-

tersections of the ray with the isosurfaces in the

model. We adopt the same solution as Hinck-

ley (Hinckley et al., 1994b) cycling from one tar-

get to the next.

The key difference with previous selection meth-

ods is the ability to work with volumetric models by

automatically generating candidate points through a

rapid isosurface detection. Moreover, we also add vi-

sual cues that facilitate the understanding of the ray

position and orientation, and disocclude inner inter-

section points.

Figure 2 shows all the components involved in the

metaphor. In this example, the dataset consists of four

spheres of different materials. The metaphor works as

follows: when the user presses the back button of the

input device, the selection task starts and the ray is

painted with a gradient color from red to yellow (in

this way we provide users with a visual cue of the

depth of the ray). Throughout this process, the sys-

tem continuously computes the proper set of candi-

date points. This set is composed by all the intersec-

tions of the ray with the implicitly deﬁned isosurfaces.

Upon button release, the last ray shown is locked,

meaning that the selection phase has ﬁnished and the

disambiguation task begins. The nearest candidate

point is marked in orange (default selection) and the

rest of the points are in white. The joystick provided

by the input device allows the user to cycle between

all the candidate points. This is convenient because

it reduces movements. Since candidate points may

have a random distribution, tracking the user’s move-

ment to reach all the candidate points without a large

arm movement (as proposed in (Grossman and Bal-

akrishnan, 2006)) would be difﬁcult and might result

in large varying patterns for different rays of the same

volume.

As the 3D ray is painted over the volume, it is

sometimes difﬁcult to interpret how the volume is tra-

versed. In order to give the user a second cue on the

intersection of the ray with the volume, we provide

the Helper Views. These showed to be of great utility,

since some candidate points are usually occluded by

other parts of the volume (Figure 1 shows a snapshot

of our technique and the accompanied video details

the interaction process). We augment the visualiza-

tion of the volume models with a wireframe represen-

tation of the cutting planes used in the Helper Views

in order to provide the users with a visual feedback of

the placement of such planes.

3.2 Implementation Details

In this section we give some details on how the isosur-

faces are detected in real time as well as on how the

DAAPMed:AData-awareAnchorPointSelectionToolforMedicalModelsinVREnvironments

311

Helper Views are created. One key difference with

other anchor point selection methods is the automatic

detection of isosurfaces on-the-ﬂy along the pointing

ray. Since we have a non-segmented model, this iso-

surfaces must be determined in real-time, as they de-

pend on the transfer function. Throughout all the pro-

cess we use a DVR method using a GPU-based ray

casting.

3.2.1 Ray - Isosurface Intersection Detection

Volumetric models can be seen as a 3D scalar func-

tion f : V ⊆ ℜ

→ ℜ (e.g. density value of a mate-

rial). Let T F : ℜ → ℜ

be the transfer function used

in the volume rendering algorithm, that assigns color

and opacity to a scalar property. First of all, we have

to deﬁne the conditions that a point p of the volume

dataset V must fulﬁll to be considered a boundary-

surface candidate point. These are:

1. p must belong to a visible material. This

condition can be expressed formally as

opacity(T F( f (p))) > 0.0

2. p must belong to the boundary of a well-deﬁned

isosurface. This condition is satisﬁed if:

(a) The gradient at point p, ∇ f (p), has to be well

deﬁned. This means that k∇ f (p)k is larger than

a certain threshold. This threshold is automati-

cally set by a previous analysis of the range of

the magnitudes of the gradient.

(b) There exists a change in the sign of the direc-

tion of the gradient at p at the neighborhood

of p. This property expresses the fact that the

boundary passes through p.

Since the detection of the boundary condition

(2.b) may not be real-time in a VR environment, the

information necessary to test this condition is precom-

puted. This is carried out by applying a 3D edge de-

tection process (Monga et al., 1990) to the volume V

and storing the result in a 3D texture which consists

of a value per voxel that indicates the possibility of

being crossed by the boundary of a surface. The sec-

ond condition is tested by checking whether p belongs

to a boundary voxel. Our system guarantees testing

at least a point for each voxel intersected by the ray,

thus the accuracy of our approach is related to voxel’s

size. As shown in Section 4, we obtain an accuracy

comparable to that of both a clipping plane selection

approach and to a desktop application which works

with a triangle mesh models (not a volume model) for

the anatomical structures. This is due to the fact that

surface extraction methods also have an accuracy pro-

portional to the voxel size. Its computation is compa-

rable to the model loading time.

3.2.2 Helper Views

The goal of Helper Views is to provide additional in-

formation on the exact position of the ray inside the

volume. These views are drawn on two ﬁxed planes,

located to the left (YZ) and bottom (XZ) of the vol-

ume dataset (see Figure 2). Images displayed on these

image planes are generated with the same algorithm

used for rendering the volume dataset but clipping it

by the plane that contains the ray and is the most par-

allel to each of the image planes YZ and XZ, respec-

tively. This has a main advantage: it shows the candi-

date points that lie inside the volume, therefore facili-

tating ray selection without previous manipulation of

the volume (i. e. clipping).

These views can also be adjusted by the user: they

can be rotated, moved, and the image projected onto

them may be zoomed. We provide a default orien-

tation with the planes slightly oriented towards the

viewer and at a relatively small distance from the ob-

ject that seems usable for several models. This default

orientation has been decided after a previous infor-

mal experiment with users that did not participate in

the user study. To facilitate its manipulation, the pro-

jection onto the Helper View plane is stored in a tex-

ture; therefore, the different operations on these views

come at no cost. Figure 2 shows the placement and the

visualization of the Helper Views design. The visual-

ization of the cut volume dataset is enhanced with an

illustrative motif:view-dependent contours computed

by using a Sobel kernel in order to enhance the sil-

houette of the clipped region.

4 EVALUATION AND RESULTS

We have conducted a formal user study to evaluate

Figure 3: Adaptation to VR of the clipping plane technique

for selecting points located on it.

GRAPP2013-InternationalConferenceonComputerGraphicsTheoryandApplications

312

the accuracy, efﬁciency and ease of use of our ap-

proach. We take as a reference an implementation of

the Clipping Plane (CP) selection method, since it is a

technique that has been widely used in medical appli-

cations (see Section 4.1 for the details of our imple-

mentation). Furthermore, we have also compared the

precision of our technique with a typical desktop ap-

plication based on triangle meshes (see Section 4.3).

The results show that our technique is as accurate as a

desktop-based method and exhibits good timings. We

also found that the users required far less movement

with our system than with CP.

We performed the user study in an immersive vir-

tual reality setup composed of a 2.7 × 2 meters pas-

sive stereo PowerWall. Users were tracked using

an Intersense IS-900 Motion Tracking System device

consisting on a Head Tracker and a MiniTrax Wanda

with a joystick and ﬁve programmable buttons.

4.1 Design Details of the Clipping Plane

Technique in Virtual Reality

In order to compare our selection technique with the

classical approach for anchor point selection using

clipping planes (CP), we ported this metaphor the fol-

lowing way. Two buttons of the input device allows

the user to set the action to be performed: rotate or

translate the clipping plane. While the user is press-

ing the selected button, the clipping plane is rotated or

translated accordingly to the user’s hand movement.

The rotation is based in the paradigm of the Rolling

Ball (Hanson, 1992). The translation is always done

in the direction of the plane’s normal. Once the plane

is ﬁxed, the user can select a point on it using the ray-

cursor paradigm. By pressing another button, the user

indicates the desired point, so every point inside the

volume, belonging to the plane, could be a candidate

point to be selected. However, due to changes in the

holding forces done by users when pressing or releas-

ing a button, called Heisenberg effect (Bowman et al.,

2002), the accuracy of the selection may be affected.

In order to solve this problem, we enhanced the visu-

alization of the ray with a freezing timer.

4.2 Test 1: Synthetic Points

Medical doctors often address two different point

selection problems: selection of well-established

anatomical points and distance measurement. As a

consequence, we decided to test two different tasks:

one regarding the selection of some speciﬁc points

and the other focused on performing the calculation

of certain measure.

In this ﬁrst test, the points and distances shown to

he users are synthetic in the sense that they do not be-

long to relevant anatomical points common in medical

environments. In Section 4.3 we will address a case

more related to practical medicine.

In these experiments, we want to evaluate the efﬁ-

ciency and the accuracy of our proposal with respect

the CP technique. Throughout the tests, we measured

several magnitudes that will provide information on

the amount of displacement (and thus, effort) each

technique requires. We recorded the following indi-

cators for each of the experiments in the test:

• Task Completion Time. We measured the amount

of time devoted to complete each experiment.

• Input Device Footprint. We measured the length

of the total path followed by the device to com-

plete each task.

• User Footprint. It measures the user displacement

inside the VR environment.

• Accuracy: This value measures the error in the se-

lection with respect to the reference points, taking

into account the dimension of the voxel dataset as

a metric of the error made.

• Hit Rate. This variable tracks the number of hits

the user has to do. Since each introduced point

may be changed if it is not satisfactory, we also

count the number of times a point is selected be-

fore its validation.

Data Preparation. We prepared two different

datasets. The ﬁrst one was used for training, while

the other was used for the test. The training model

consisted of a set of four spheres of different material

(see Figure 4-left). The second model consisted of

a typical model in volume visualization, a CT tooth

dataset using a transfer function that enhanced the

outside and inner shape of it (see Figure 4-right).

Figure 4: The training (left) and testing (right) datasets. The

ﬁgures show the anchor points to be selected in the ﬁrst ex-

periment.

Subjects and Procedure. 17 subjects participated in

the evaluation; 13 male and 4 female, ranging be-

tween 23 and 63 years old. Subjects were asked to

classify (as Low, Medium or High) their experience

in a VR setup, experience with input devices and ex-

pertise in 3D application. All of the participants were

DAAPMed:AData-awareAnchorPointSelectionToolforMedicalModelsinVREnvironments

313

Figure 5: Images that show the second experiment descrip-

tion, as presented to the participants in the test.

people from our department. Computer scientists at

different levels of studies (master and PhD students)

and faculty staff.

Every user performed both experiments once. Be-

fore the ﬁrst experiment started, a complete training

(using the spheres dataset) was performed to get fa-

miliar with the two interaction techniques to evaluate:

DAAPMed and CP techniques. The test was divided

into two blocks, one for each technique. The order

of the blocks was chosen randomly in order to avoid

skewing one of the techniques with a learning effect.

As said before, the test consisted of two kind of

experiments: selecting two predeﬁned points, and

measuring a certain magnitude deﬁned by two points.

For the ﬁrst experiment, we asked the users to intro-

duce two anchor points (P

and P

) at positions that

were marked in the model with the use of a cone rep-

resentation (see Figure 4). Once completed, we stop

tracking the movements of the user until she is ready

for the next experiment. The second experiment con-

sisted of taking a measure (calculated as a distance

between two anchor points). We refer to this calcu-

lation as Thickness. The speciﬁcation of this exper-

iment was accompanied with different descriptions

and pictures of the goal (see Figure 5). None of the

users involved in the experiment had any problem un-

derstanding the objective of the experiment. Users

can repeat the selection of a point as many times they

need. But once the point was validated by the user, he

or she could not repeat its placement.

4.2.1 Results

A repeated measures within subjects design was used.

The independent variable was the technique and the

dependent variables were the set of tracked variables.

A one-way analysis of variance (ANOVA) comparing

both techniques was use.

Table 1 summarizes the statistical analysis of the

relevant variables. For each variable the mean and the

standard deviation are shown. The ﬁrst experiment is

tagged as P

and P

, corresponding to the two anchor

points. Second experiment is tagged as T hickness.

Regarding Completion Time, there is signiﬁcant

evidence in all the experiments that DAAPMed per-

formed better than CP. For P

(p = 0.028, F = 5.83),

for P

(p = 0.008, F = 9.35) and for Thickness (p =

0.044, F = 4.79). Figure 6 shows the total time for

each technique.

Figure 6: Results of the completion task timings. The boxes

show the interquartile range with the median as the horizon-

tal bar. The whiskers extend to the minimum and maximum

of the data. CP exhibits longer selection times.

Regarding Input Device Footprint, we measured

the length of the total path which the device took to

complete the experiment. We have found a signiﬁcant

effect on the Input Device Footprint variable for P

(p = 0.036, F = 5.24) and for P

(p = 0.004, F =

11.70). Figure 7 illustrates the effect. The reduction

of footprint is especially important since a handheld

6-DOF device is being used, which can lead to fatigue

with extended use (Ware and Slipp, 1991).

Figure 7: Input device footprints. The boxes show the in-

terquartile range with the median as the horizontal bar. The

whiskers extend to the minimum and maximum of the data.

For point selection it is clear that DAAPMed method per-

formed signiﬁcantly better than CP.

We also split the movement done by the device

taking into account whether the movement was due

to an exploration phase (rotating or translating the

model) or due to the selection phase. We have only

found signiﬁcative statistical difference between the

two techniques for P

(p = 0.007, F = 9.44), for the

rest of the experiments DAAPMed performed bet-

ter comparing means and standard deviations. With

CP technique, the user performs similar amount of

moves during navigation and during selection. On

GRAPP2013-InternationalConferenceonComputerGraphicsTheoryandApplications

314

Table 1: The overall statistical results of the evaluation shown as means and standard deviations of the variables measured for

the tooth model. Regarding the mean and the standard deviation, DAAPMed is superior to CP. The one-way ANOVA analysis

showed which differences were statistically signiﬁcative.

CP DAAPMed

Thickness P

Thickness

Accuracy 0.76 ± 0.23 0.93 ± 1.37 1.15 ± 0.81 0.562 ± 0.229 1.37 ± 3.13 1.081 ± 0.79

Completion Time 62.42 ± 34.08 73.8 ± 47.1 119.8 ± 65.5 43.07 ± 36.54 41.1 ± 25.7 84.1 ± 43.4

Movement Tool 3.711 ± 2.75 4.86 ± 4.81 7.92 ± 5.57 2.335 ± 3.26 1.88 ± 1.46 5.42 ± 3.31

Movement User 1.942 ±1.53 2.41 ±2.33 4.281 ± 3.23 1.331 ± 1.87 1.261 ± 1.04 2.835 ± 1.79

the other hand, when selecting using DAAPMed tech-

nique, users devoted a larger effort to the navigation

phase than to the selection one (see Figure 8).

Figure 8: Input device footprints. The displacement car-

ried out by the device is split in two states: navigation

(N {P

,T h}) and selection (S {P

,T h}).The boxes

show the interquartile range with the median as the horizon-

tal bar. The whiskers extend to the minimum and maximum

of the data.

We also measured the movement carried out by

the user. In all cases, our system requires a lower

amount of movement by the user. The analysis shows

that the movement done in DAPPMed is signiﬁcantly

less than CP for P

(p = 0.009, F = 8.72) and for

Thickness (p = 0.03, F = 5.62).

Thus, we can conclude that our system behaves

better than CP.

Concerning the accuracy, the mean values show

better performance for our technique. However, we

did not ﬁnd signiﬁcant statistical differences. The

reason could be that with CP technique you can get

enough precision if you know exactly the point you

have to select. In order to do a deeper analysis, we

perform another test which is closer to a real medical

scenario since we use points with anatomical signiﬁ-

cance.

We have also tracked another set of variables, such

as the number of points selected before validating

them, or the number of exploration versus selection

phases, but we could not extract any behavior or pat-

tern from those.

4.3 Test 2: Anatomic-based Points

We have carried out a second experiment where the

workﬂow is closer to a real medical environment.

The selection points are points with speciﬁc anatomic

meaning commonly used to place anchor points or

measuring distances. The objectives of this test were

twofold: a) Testing if our VR application was as

accurate than the desktop application and, b) ﬁnd-

ing whether DAAPMed technique was more accurate

than the CP technique.

To carry out this analysis we use a speciﬁc

desktop-based application for the morpho-analysis of

the abdominal air (Accarino et al., 2009). In this ap-

plication, users had to mark a set of points on the

skeleton of a model in order to infer some measures.

Using this application, we marked the set of points

shown in Figure 9. The exact location of these points

were saved to a ﬁle in order to evaluate the accuracy

reached in a VR setup. Users, with the helping of the

image shown in Figure 9, have to locate these points

as accurately as possible.

Data Preparation. The model used was taken from a

medical dataset. It has a resolution of 512×512×369

and a voxel dimension of 2.042

× 3.56mm. The

desktop-application uses a mesh representation of the

data extracted from the volume dataset using the

Marching Cubes algorithm. Our application uses the

same volume model visualized with a DVR algo-

rithm.

Subjects and Procedure. 13 subjects participated in

this test: 10 male and 3 female, ranging between 23

Figure 9: Model used in the second test. This test conﬁrms

that DAAPMed is as accurate as a clinical environment in

a desktop-based computer and more accurate than the CP

technique.

DAAPMed:AData-awareAnchorPointSelectionToolforMedicalModelsinVREnvironments

315

and 40.

In this experiment we proceed in the same way as

the ﬁrst test. Each participant performs the test once.

Before the experiment, users were provided with a

short (1-3 min.) training session. The test was di-

vided into two blocks, one for each technique: CP

and DAAPMed. The order of the blocks was chosen

randomly in order not to introduce some learning ef-

fect.

4.3.1 Results

The results are summarized in Table 2. The ﬁrst and

second rows show the mean and the standard devia-

tion for both techniques. The third and fourth rows

show the statistical signiﬁcance information (p anf

F). For all the points introduced (except P

and P

the DAAPMed technique shows signiﬁcant statistical

difference with respect CP. We do not have a clear

idea on the lack of signiﬁcance of points P

and P

but it might be that the speciﬁcation of their corre-

sponding positions was not as clear as with the others.

Figure 10 shows a boxplot for all the tasks performed.

We also compared the accuracy with the desktop

approach and obtained errors that did not differ sig-

niﬁcantly, which demonstrates that we may achieve

comparable results in a VR environment. In both

cases, the error performed was below the voxel size.

Figure 10: Accuracy by technique. The boxes show the

interquartile range with the median as the horizontal bar.

The whiskers extend to the minimum and maximum of the

data.

4.4 Post-questionnaire Results

To complete the information, we also asked the sub-

jects to ﬁll some questionnaires, to know the prefer-

ences of the users between the two techniques. All

responses in the post-questionnaire were measured on

a Likert scale of 1-5, where 1 meant the worst value

and 5 was the best value. The results are shown in Fig-

ure 11. The answers seem to indicate that DAAPMed

metaphor is more suitable than the CP technique.

The users noted two major problems with respect

Figure 11: Results obtained from a personal preference

evaluation questionnaire. These results show that the users’

perceptions are quite positive with our tool.

to our technique. The ﬁrst one is the inherent jitter-

ing of the tracker, that made selection affect user per-

formance. Only two users agreed in that it seems to

produce a more relevant effect to the ray-based selec-

tion than to the plane-based. Furthermore, in all the

experiments, the ray-based approach showed a better

behavior than the clipping-planes system. The sec-

ond issue was the lack of ray reﬁnement: most users

suggested that a ﬁne tuning of the ray, after its initial

positioning would be welcome. We let this work for

future improvements.

5 CONCLUSIONS AND FUTURE

WORK

We have presented a new interaction technique for

selecting points in a volume dataset. This selection

technique follows the ray casting paradigm, enhanced

with an automatic calculation of the set of suitable

points of interest by an on-the-ﬂy determination of the

isosurfaces along the ray path. The feedback with the

interaction is enhanced with a meaningful visualiza-

tion called Helper Views that provides context for the

ray selection and shows occluded detected candidate

points that would be otherwise invisible to the user

without posterior and ad-hoc volume manipulation.

The user study showed that our technique is easy

to learn and to use. Despite the limited precision of

the 3D input devices, our technique achieves a precise

3D interaction thanks to the automatic anchor point

calculation provided by the system. Users felt more

comfortable and achieved better results with our sys-

tem than with the clipping plane technique.

In the future we would like to continue working

on the improvement of the accuracy with the current,

imprecise devices, and we also want to carry out a

study of the overall system with medical experts.

GRAPP2013-InternationalConferenceonComputerGraphicsTheoryandApplications

316

Table 2: The overall statistical results of the evaluation shown as means and standard deviations of the tolerance error. We

can clearly see how the DAPPMed metaphor provides better results for all the tasks than the CP method.

CP 2.944 ± 1.305 3.018 ± 1.495 3.171 ± 1.729 2.336 ±0.879 2.066 ± 1.095 2.070 ± 1.169

DAAPMed 1.293 ± 0.671 1.701 ± 0.703 1.605 ± 0.501 1.766 ± 0.570 1.789 ± 0.424 0.280 ± 0.078

p,F 0.002 - 16.55 0.011 - 9.01 0.005 - 11.58 0.187 - 1.96 0.385 - 0.81 0.001 - 17.42

ACKNOWLEDGEMENTS

The authors want to thank all the participants involved

in the user study. This work has been supported by the

project TIN2010-20590-C01-01 of the Spanish Gov-

ernment.

REFERENCES

Accarino, A., Perez, F., Azpiroz, F., Quiroga, S., and

Malagelada, J. (2009). Abdominal distention results

from caudo-ventral redistribution of contents. Gas-

troenterology, 136:1544–1551.

Bowman, D., Kruijff, E., LaViola, J., and Poupyrev, I.

(2004). 3D User Interfaces: Theory and Practice.

Addison-Wesley,Pearson Education.

Bowman, D. A., Johnson, D. B., and Hodges, L. F. (1999).

Testbed evaluation of virtual environment interaction

techniques. In Proceedings of the ACM symposium on

Virtual reality software and technology, pages 26–33.

Bowman, D. A., Wingrave, C. A., Campbell, J. M., Ly,

V. Q., and Rhoton, C. J. (2002). Novel uses of pinch

gloves for virtual environment interaction techniques.

Virtual Reality, 6:122–129.

Gallo, L., De Pietro, G., and Marra, I. (2008). 3d interac-

tion with volumetric medical data: experiencing the

wiimote. In Proceedings of the 1st international con-

ference on Ambient media and systems, pages 14:1–

14:6.

Grossman, T. and Balakrishnan, R. (2006). The design and

evaluation of selection techniques for 3d volumetric

displays. In Proceedings of the Symposium on User

interface software and technology, pages 3–12. ACM.

Hagedorn, J., Joy, P., Dunkers, S., and Peskin, A. (2007).

Measurement tools for the immersive visualization en-

vironment: Steps toward the virtual laboratory. Jour-

nal of Research of the National Institute of Standards

and Technology, 112(5):257–270.

Hanson, A. J. (1992). The rolling ball. In Graphics Gems

III, pages 51–60. Academic Press.

Hastreiter, P., Tom, B., Eberhardt, K. E. W., and Ertl,

T. (1998). Fast analysis of intracranial aneurysms

based on interactive direct volume rendering and ct

angiography. In CTA, Medical Image Computing

and Computer-Assisted Intervention (MICCAI), pages

660–669. Springer.

Hinckley, K., Pausch, R., and Goble, J. (1994a). A three-

dimensional user interface for neurosurgical visualiza-

tion. In In the SPIE Conf. on Medical Imaging, pages

126–136. SPIE.

Hinckley, K., Pausch, R., Goble, J., and Kassell, N. (1994b).

A survey of design issues in spatial input. In Proceed-

ings of the Symposium on User interface software and

technology, pages 213–222. ACM.

onig, A. H., Doleisch, H., Gr

oller, E., and Brain, T. H.

(1999). Multiple views and magic mirrors - fmri visu-

alization of the human brain.

Mine, M. (1995). Virtual environment interaction tech-

niques. Technical report, UNC Chapel Hill CS Dept.

Monga, O., Deriche, R., Malandain, G., and Cocquerez,

J. P. (1990). Recursive ﬁltering and edge closing: two

primary tools for 3-d edge detection. In Proceedings

of the ﬁrst european conference on Computer vision,

pages 56–65. Springer-Verlag.

Olwal, A. and Feiner, S. (2003). The ﬂexible pointer: An in-

teraction technique for augmented and virtual reality.

In Proceedings of ACM Symposium on User Interface

Software and Technology, pages 81–82.

Preim, B., Tietjen, C., Spindler, W., and Peitgen, H. O.

(2002). Integration of measurement tools in medical

3d visualizations. In Visualization ’02, pages 21–28.

IEEE Computer Society.

Qi, W. and Martens, J.-B. (2005). Tangible user interfaces

for 3d clipping plane interaction with volumetric data:

a case study. In Proceedings of the 7th international

conference on Multimodal interfaces, pages 252–258.

Reitinger, B., Schmalstieg, D., Bornik, A., and Beichel, R.

(2006). Spatial analysis tools for virtual reality-based

surgical planning. 3D User Interfaces, 0:37–44.

ossling, I., Cyrus, C., Dornheim, L., Boehm, A., and

Preim, B. (2010). Fast and ﬂexible distance measures

for treatment planning. International Journal of Com-

puter Assisted Radiology and Surgery, 5:633–646.

Song, P., Goh, W. B., Fu, C.-W., Meng, Q., and Heng, P.-A.

(2011). Wysiwyf: exploring and annotating volume

data with a tangible handheld device. In Proceedings

of the 2011 annual conference on Human factors in

computing systems, pages 1333–1342. ACM.

Ware, C. and Slipp, L. (1991). Using velocity control to

navigate 3d graphical environments: A comparison of

three interfaces. In Human Factors and Ergonomic

Studies (HFES) Meeting, pages 25–32.

DAAPMed:AData-awareAnchorPointSelectionToolforMedicalModelsinVREnvironments

317