Assessment of 3D Scanners for Modeling Pectus Carinatum

Corrective Bar

António H. J. Moreira

1,2

, João Gomes Fonseca

, Pedro L. Rodrigues

1,2

, Jaime C. Fonseca

1,5

A. C. M. Pinho

, Jorge Correia-Pinto

, Nuno F. Rodrigues

3,4

and João L. Vilaça

1,2,3

Centro ALGORITMI, School of Engineering, University of Minho, Guimarães, Portugal

ICVS/3B’s - PT Government Associate Laboratory, Braga/Guimarães, Portugal

DIGARC, Polytechnic Institute of Cávado and Ave, Barcelos, Portugal

HASLab / INESC TEC –University of Minho, Braga, Portugal

DEI, Industrial Electronics Department, University of Minho, Guimarães, Portugal

DEM, Mechanical Department, University of Minho, Guimarães, Portugal

Keywords: 3D Scanner, Kinect, FastSCAN, Pectus Carinatum, Surface Mesh, 3D Reconstruction.

Abstract: Pectus Carinatum (PC) is a chest deformity consisting on the anterior protrusion of the sternum and

adjacent costal cartilages. Non-operative corrections, such as the orthotic compression brace, require

previous information of the patient chest surface, to improve the overall brace fit. This paper focuses on the

validation of the Kinect scanner for the modelling of an orthotic compression brace for the correction of

Pectus Carinatum. To this extent, a phantom chest wall surface was acquired using two scanner systems –

Kinect and Polhemus FastSCAN – and compared through CT. The results show a RMS error of 3.25mm

between the CT data and the surface mesh from the Kinect sensor and 1.5mm from the FastSCAN sensor.

1 INTRODUCTION

Pectus Carinatum (PC) is a chest deformity

characterised by the anterior projection of the

sternum and adjacent costal cartilages. The

deformity is caused by a disproportionate growth of

the costal cartilages, compared to the thoracic

skeleton, resulting in a protrusion with symmetric or

asymmetric overgrowth (Golladay ES., 2003).

Nowadays, several authors propose non-

operative corrections, such as the body cast or the

orthotic compression brace (OCB) (Frey et al.,

2006).

Due to the symmetric or asymmetric nature of

the deformity, some studies report better outcomes

with a custom-fitted OCB (Egan et al., 2000).

Currently, this procedure is performed using

rough measurements, which may lead to future

adjustments in OCB design; or CT-scan with

radiation dosage; or, in few cases, precision 3D

scanners which are costly (Philippe et al., 2007).

In this paper our primary focus is the validation

of the Kinect sensor as a handheld scanner for

modelling Pectus Carinatum OCB. To this extent, it

is compared and discussed the differences between

the 3D scans and errors that affect the OCB

modelling from a software using Kinect

(ReconstructMe), the mesh from an available

handheld scanner (Polhemus FastSCAN) and the

mesh from the CT-Scan as reference.

The paper proceeds with the description of the

acquisition procedure and reliability assessment for

a phantom model, and also the measurements for the

OCB modelling. In Section 3, the errors obtained

from different scans and OCB measurements are

presented. The paper concludes with some

observations and future improvements in Section 4.

2 METHODS

2.1 3D Scanning

2.1.1 FastSCAN

The Polhemus FastSCAN™ Cobra is a handheld

scanner which uses a camera and a laser to

triangulate a 3D stripe. An electromagnetic tracking

system is used to track the scanner wand location in

the 3D space.

Resolution along the laser line depends on wand-

object range, typically 0.5mm at 200mm range and it

122

H. J. Moreira A., Gomes Fonseca J., L. Rodrigues P., C. Fonseca J., C. M. Pinho A., Correia-Pinto J., F. Rodrigues N. and L. Vilaça J..

Assessment of 3D Scanners for Modeling Pectus Carinatum Corrective Bar.

DOI: 10.5220/0004300901220125

In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2013), pages 122-125

ISBN: 978-989-8565-48-8

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

can reach resolutions as low as 0.1mm. The distance

between transmitter (small ranger) and wand is

limited to a radius of 310mm, so a good accuracy is

achieved (0.75mm) within a 600mm sphere centered

on the reference source (Polhemus, 2012).

2.1.2 3D Kinect (Reconstructme Software)

Kinect is a device composed by one Infra-Red (IR)

projector, one IR camera and one RGB camera. The

IR projector and IR camera are used to triangulate

the points in space, and to estimate the depth by

measuring the disparities captured by the IR camera

(Smisek et al., 2011); (Khoshelham and Elberink,

2012).

The operating range of the sensor is between 0.4

meters to 5 meters. At the range of 2 meters, one

level of disparity corresponds to 1 cm. Thus, to

increase the depth resolution for acquisitions with

Kinect, the acquisition range is limited to 0.4 meters

up to 1.2 meters. According to Khoshelham et. al.,

the standard deviations of depth resolution at 1.5

meters can be as high as 0.5 cm.

The software ReconstructMe (Non-commercial

version 405), developed by PROFACTOR GmbH,

was used to build the surface meshes.

(ReconstructMe, 2012). Essentially, ReconstructMe

uses depth acquisitions to represent 3D points,

which characterize a 3D scene.

2.2 Reliability Assessment

In order to access and validate the differences

between scanners capability to scan the human chest

wall, a phantom (Training Model “ABDFAN” -

Kyoto Kagaku Co., Ltd) was used in this analysis.

The usage of Kinect is then evaluated for OCB

measures by assessing its similarity with FastSCAN

and CT-Scan results.

The surface mesh reconstructed from the CT-

data is used as the ground-truth in this study. The

volume resolution is 512×512×241 with voxel

dimensions of 0.684×0.684×1mm, the 241 axial

slices were acquired with the HiSpeed CT/e™ (GE

Medical Systems).

The surface contours from the segmented slices

are used to reconstruct the final mesh, see Figure 1.

Figure 1: Surface mesh from CT-scan.

Two different setups were performed:

- Movement - the scanner moves around a static

object;

- Static - the scanner is fixed and the object moves

in front of it.

To improve the static mode, the object is fixed in

a support which allows 360 degrees rotation. For

each mode and scanner, 10 meshes were acquired.

2.2.1 Repeatability

The FastSCAN scanner is operator dependent, since

the mesh precision depends of the distance between

the wand and the reference. Occlusion is another

problem which brings the necessity of extra sweeps.

Therefore, to overcome these limitations, some post-

processing was applied to the meshes. First, the

sweeps were slightly registered to decrease the

distances between them. Then, smooth and decimate

operators were applied to the merged meshes

The repeatability was also studied in Kinect

based on ReconstructMe software with default

settings.

To analyse the repeatability, the CloudCompare

software running the ICP algorithm was used to

align the meshes. Two hundred thousand sampling

points were used. Then, the registration was

validated by Root-Mean-Square (RMS) error

between meshes. The distance between meshes was

computed assigning each point of the compared

mesh to the nearest-neighbour point in the reference

mesh.

Four different setups were defined and, for each,

10 meshes were acquired. To compute the

repeatability, the described process was applied to

all meshes. To reduce the influence of the

registration in error measurements, due to different

number of vertexes per mesh, the comparisons were

made through the combination of all meshes, using

all of them as reference. In each setup 90

comparisons were computed.

2.2.2 Accuracy

This subsection describes how the meshes accuracy

was accessed. Accuracy represents the distance

between the corresponding points of the surface

mesh acquired from the scanner and the ground-truth

surface mesh.

Here, they were applied the same steps of the

repeatability, however, for this case, the acquired

meshes were compared with ground-truth mesh built

from the CT-scan. The accuracy is measured and

compared in the four setups, resulting in a total of 40

comparisons.

Assessmentof3DScannersforModelingPectusCarinatumCorrectiveBar

123

2.3 OCB Modelling

Usually, the OCB is modelled by taking

measurements from CT-Scan or, if there is no

available patient CT-Scan, measured manually in the

patient. In the CT-data, one slice is chosen at the

point of greatest protrusion. The measurements for

modelling the OCB are the transverse diameter of

the thorax (Figure 2 – A), the right and left

hemithorax distance (Figure 2 – B and C) and the

thorax perimeter. The curvature of the anterior and

posterior elements of the OCB (Figure 2 – D

and

) are modelled following the lateral tangential

curvature of the chest.

Figure 2: Measurements for modelling the OCB: A –

transverse diameter of the thorax; B and C –

anteroposterior distance of the right and left hemithorax.

and D

– anterior and posterior elements of the OCB, E

– Contact pillow/support.

To realise if the OCB modelling can be achieved

using 3D scanners, the defined measurements were

computed in both scan meshes and compare to the

CT-Scan mesh in the greatest protrusion point.

3 RESULTS AND DISCUSSION

For each scan preformed with the Kinect, the

phantom was turned around 360º, at least 3 times, to

minimize random noise.

When the mean and standard deviation were

computed, the outliers were eliminated using

99.73% (3σ

) of the total data for each mesh

comparison.

3.1 Repeatability

The results reveal that the Kinect acquisition based

on ReconstructMe has higher errors for both modes.

The RMS errors, obtained from registration, are

reported in Figure 3.

As the differences are not substantial, it is possible

to mention that FastSCAN performs repeatable

Figure 3: Repeatability - Boxplot results of the RMS error

for the four setups. Mean value is represented by the

yellow dot.

acquisitions in both modes. Using Kinect, the results

reveal that repeatability is higher when the sensor

stays static and the object moves. When the Kinect

is used as a handheld scanner (movement mode), the

influence of the operator is verified - on average

higher than 0.5mm.

3.2 Accuracy

The accuracy results are presented in Figure 4,

revealing that FastSCAN is more accurate than

Kinect.

Figure 4: Accuracy - Boxplot results of the RMS error for

the four setups for. Mean value is represented by the

yellow dot.

On average, the accuracy differences between

scanners are higher than 1.8mm in RMS error.

Although, when comparing acquisition modes

(movement and static), these do not greatly

influence the resulting meshes in either scanners.

3.3 Corrective Bar Similarity

Thorax perimeter, transverse diameter, left and right

hemithorax distances are reported in table 1. This

VISAPP2013-InternationalConferenceonComputerVisionTheoryandApplications

124

table describes the mean values of the 10 meshes of

each setup.

Table 1: OCB thorax mean distances, in millimetres.

Setup TP TTD

Left

HTD

Right

HTD

CT (Ref) 858,02 338,38 218,06 216,69

Kinect

(MM)

842,30

±6,28

332,29

±1,72

211,16

±1,49

211,03

±3,87

97,65

±0,93

Kinect

(SM)

855,60

±7,07

∆

328,38

±0,96

206,00

±1,74

216,10

±0,85

∆

97,74

±0,57

FastSCAN

(MM)

865,85

±2,78

339,77

±0,49

∆

217,93

±1,02

∆

218,88

±0,54

□

100,57

±0,30

FastSCAN

(SM)

853,35

±3,54

□

334,45

±0,71

□

215,63

±0,95

□

213,21

±0,61

98,89

±0,33

TP – thorax perimeter; TTD - thorax transverse distance; HTD –

Hemithorax distance; % - percentage of similarity with CT-data;

SM – static mode; MM – movement mode. ∆ - best result; □ –

second best result.

The best performance was achieved using the

FastSCAN in MM setup (see Table 1 - %). Also, the

overall best results are achieved with FastSCAN

scanner in both modes, MM and SM.

Observing the difference of similarity in Kinect

(~2.5%), this can result in mean error between 5mm

and 16mm, affecting the OCB modelling. In the

FastSCAN case, with difference of similarity lower

than 1.2%, the worst mean error is 7.8mm.

4 CONCLUSIONS

FastSCAN has revealed to be the most accurate and

precise scanner. Kinect, with ReconstructMe

software, has proved to be a well capable system for

the acquisition of 3D objects, demonstrating a RMS

accuracy error up to 3mm, higher than FastSCAN

(~1.5mm), when compared to ground-truth. Also, it

shows less level of detail than FastSCAN.

Since Kinect is a static acquisition system, it

shows more variability when used as a handheld.

Unlike it, FastSCAN remains stable in both motion

setups, SM and MM.

One major drawback of FastSCAN system is its

cost when compared to Kinect.

Future improvements in Kinect registration and

depth field sensor can expand the usage of this

scanner as a low-cost handheld device allowing for

fast and precise remote scans for custom-fitted OCB

modelling.

ACKNOWLEDGEMENTS

The authors acknowledge to Foundation for Science

and Technology (FCT) - Portugal for the fellowships

with the references: UMINHO/BI/95/2012;

SFRH/BD/68270/2010; SFRH/BD/74276/2010 and

SFRH/BPD/46851/2008. This work was also

supported by FCT R&D project PTDC/SAU-

BEB/103368/2008.

REFERENCES

Golladay, E. S., 2003. Pectus carinatum and other

deformities of the chestwall. In Ziegler, M. M.,

Azizkhan, R. G., Weber, T. R. Operative pediatric

surgery. New York (NY) McGraw-Hill, p. 269-77.

Frey, A. S., Garcia, V. F., Brown, R,L,, Inge, T. H.,

Ryckman, F.C., Cohen, A. P., Durrett, G., Azizkhan,

R. G., 2006. Nonoperative management of pectus

carinatum. In J Pediatr Surg, 41, p. 40-5.

Egan, J. C., DuBois, J. J., Morphy, M., et al., 2000.

Compressive orthotics in the treatment of asymmetric

pectus carinatum: a preliminary report with an

objective radiographic marker. In J Pediatr Surg, 35,

p. 1183-6.

Philippe Poncet, Dragan Kravarusic, Tessa Richart,

Rhiannon Evison, Janet L. Ronsky, Ali Alassiri, David

Sigalet, Clinical impact of optical imaging with 3-D

reconstruction of torso topography in common anterior

chest wall anomalies, Journal of Pediatric Surgery,

Volume 42, Issue 5, May 2007, 898-903.

Smisek, J., Jancosek, M., Pajdla T., 2011. 3D with Kinect.

In IEEE International Conference on Computer Vision

Workshops, p.1154-60.

Khoshelham, K., Elberink, S. O., 2012. Accuracy and

Resolution of Kinect Depth Data for Indoor Mapping

Applications. In Sensors, 12, p. 1437-54.

Polhemus, 2012. http://www.polhemus.com/ (accessed

August 2012).

ReconstructMe, 2012. http://reconstructme.net/ (accessed

August 2012).

Assessmentof3DScannersforModelingPectusCarinatumCorrectiveBar

125