Basic Science
Fully Automatic Three-Dimensional Quantitative
Analysis of Intracoronary Optical
Coherence Tomography: Method and Validation
Kenji Sihan,1,2 BSc, Charl Botha,2 PhD, Frits Post,2 PhD, Sebastiaan de Winter,2 BSc,
Nieves Gonzalo,1 MD, Evelyn Regar,1 MD, PhD, Patrick J.W.C. Serruys,1 MD, PhD,
Ronald Hamers,1 PhD, and Nico Bruining,1* PhD
Objectives and background: Quantitative analysis of intracoronary optical coherence
tomography (OCT) image data (QOCT) is currently performed by a time-consuming
manual contour tracing process in individual OCT images acquired during a pullback
procedure (frame-based method). To get an efficient quantitative analysis process, we
developed a fully automatic three-dimensional (3D) lumen contour detection method
and evaluated the results against those derived by expert human observers. Methods:
The method was developed using Matlab (The Mathworks, Natick, MA). It incorporates
a graphical user interface for contour display and, in the selected cases where this
might be necessary, editing. OCT image data of 20 randomly selected patients,
acquired with a commercially available system (Lightlab imaging, Westford, MA), were
pulled from our OCT database for validation. Results: A total of 4,137 OCT images
were analyzed. There was no statistically significant difference in mean lumen areas
between the two methods (5.03 6 2.16 vs. 5.02 6 2.21 mm2; P 5 0.6, human vs. auto-
mated). Regression analysis showed a good correlation with an r value of 0.99. The
method requires an average 2–5 sec calculation time per OCT image. In 3% of the
detected contours an observer correction was necessary. Conclusion: Fully automatic
lumen contour detection in OCT images is feasible with only a select few contours
showing an artifact (3%) that can be easily corrected. This QOCT method may be a valua-
ble tool for future coronary imaging studies incorporating OCT. VC 2009 Wiley-Liss, Inc.
Key words: angiography; coronary; diagnostic cardiac catheterization; quantitative
vascular angiography
INTRODUCTION
Optical coherence tomography (OCT) has been rap-
idly accepted as an additional invasive coronary imaging
tool [1]. It allows highly detailed imaging of the coro-
nary lumen and vessel wall morphology at resolutions of
10 times higher than what current intracoronary ultra-
sound (ICUS) can offer. The details shown within OCT
images are close to histopathology, allowing accurate
evaluation of by example apposition of stent struts
and––in longitudinal studies––tissue-coverage of drug-
eluting stents (DES) [2]. This advantage has resulted in
an increasing use of OCT in studies evaluating new ther-
apeutic treatments, in addition to ICUS, the current de
facto reference method for intravascular imaging [3,4].
To apply an imaging method in studies, accurate
quantitative analysis tools are required, as has been
proven by quantitative angiography (QCA) [5] and
quantitative ICUS (QCU) [6,7]. Recently, results of
1Thoraxcenter, Delft, The Netherlands
2Department of Cardiology, Erasmus MC, Rotterdam, The
Netherlands and TU-Delft, Delft, The Netherlands
Conflict of interest: Dr. Hamers is an employee of CURAD BV,
Wijk Bij Duurstede, The Netherlands.
*Correspondence to: Nico Bruining, PhD, Erasmus MC, P.O. Box
1738, 3000 DR Rotterdam, The Netherlands. E-mail: n.Bruining@
erasmusmc.nl
Received 25 March 2009; Revision accepted 23 April 2009
DOI 10.1002/ccd.22125
Published online 11 June 2009 in Wiley InterScience (www.
interscience.wiley.com)
V
C 2009 Wiley-Liss, Inc.
Catheterization and Cardiovascular Interventions 74:1058–1065 (2009)

computer-assisted quantitative OCT (QOCT) have been
reported showing good results for coronary lumen
measurements [8]. However, because an OCT examina-
tion contains hundreds of individual images, such anal-
ysis can, despite the use of computer-assisted tools, be
a tedious and time-consuming process [8]. However,
because the lumen-intima interface is so clearly visual-
ized in OCT, far better than in ICUS, a fully automatic
lumen contour detection approach could be feasible.
This article reports on a novel in-house developed fully
automatic three-dimensional (3D) OCT lumen contour
detection approach and its validation using expert human
observer computer-assisted analysis as a reference.
MATERIALS AND METHODS
OCT Imaging System
OCT imaging was performed with a commercially
available system (Lightlab imaging, Westford, MA). This
system uses a 1,310-nm broadband light-source generated
by a super luminescent diode with an output power in the
range of 8.0 mW. The average tissue penetration depth is
approximately 1.5 mm with an axial and lateral resolution
of 15 and 25 lm, respectively. The imaging probe has the
size of a guide-wire with a maximum outer diameter of
0.019 inch (ImagewireTM, LightLab Imaging). The wire
contains a single-mode fiber optic core within a translu-
cent sheath. The imaging-wire is connected to an imaging
console, similar as ICUS, which performs the real-time
image data processing, visualization, and image storage.
Systematic imaging of a coronary segment is also ana-
logue to ICUS by an automatic continuous speed pullback
(between 1 and 3 mm/sec) of the imaging wire [6]. OCT
images are generated at a rate of 10–20/sec (cf. ICUS 30
frames/sec). The accuracy of OCT for dimensional meas-
urements––determined using a phantom––has been
recently reported to be excellent (mean difference in meas-
ured length of
0.03 mm with 0.02 mm precision) [8].
Patients and OCT Image Acquisition
For validation of the automated method, we made a
random selection of 20 OCT cases from our database
of patients participating in different studies. In all cases
a standard femoral approach with 7F guiding catheters
was used. Before imaging, all patients received weight-
adjusted heparin intravenously to maintain an activated
clotting time of >300 sec as well as intravenous anal-
gesics. To be able to see the coronary vessel wall, the
coronary artery must be cleared of blood by replacing
it with lactated Ringer’s solution. This procedure is
performed by occlusion of the artery with a dedicated
occlusion catheter (Helios, Goodman, Japan) including
a short balloon (6.0 mm length) that can be inflated by
a low pressure (0.3 atm). The Ringer’s is infused
distally from the balloon at a rate of 0.5 ml/sec at a
temperature of 37C). Sufficient occlusion of the coro-
nary is checked by contrast injection via the guiding
catheter and the balloon pressure is increased (at 0.5
atm increments) when necessary. The images were dig-
itally stored in the AVI file format on DVD’s and were
translated later into the DICOM medical imaging
standard by in-house developed software.
Automated Quantitative OCT Method
The automated OCT lumen contour detection method
was developed in the Matlab environment (The Math-
works, Natick, MA). The method has five stages:
Preprocessing. Each individual OCT frame is prepro-
cessed to remove speckle noise and gaps and to adapt
the contrast for proper image normalization. Prepocessing
consists of the application of a Gaussian filter and using
different relative thresholds (Fig. 1B). The relative
thresholds are applied on the pixel values to remove
extreme values and to improve image normalization.
Edge detection. A Canny [9] filter is applied to
detect edges in the OCT image. The final lumen con-
tour is the result of appropriately selected edges, which
are positioned on the lumen-intima border only.
Straightforward application of the standard Canny filter
to the OCT images leads to many false and/or missed
contours. Therefore, we iteratively apply the Canny fil-
ter to match the constraints necessary for OCT images
(percentage of image pixels classified as true edges).
This percentage is based on a test-set where it is set in
such a way that the lumen contour is detected while
detecting as little noise as possible. Within this proce-
dure, the threshold of the Canny filter is optimized
using a binary search algorithm.
Lumen edge selection. The result of the Canny
edge detection stage includes some edges that do not
belong to the lumen-intima interface, for example radi-
ally behind bright areas (Fig. 1C). These false edges
are mostly due to noise caused by the catheter and by
speckle noise which was not be removed by the pre-
processing step without significantly impairing image
details. The majority of these false edges are identified
by two constraints:
1. The angle between the gradient orthogonal to the
line segment and the line connecting it to the cathe-
ter center should be smaller than a certain threshold.
2. The length of the edge should be longer than a cer-
tain threshold (Fig. 1).
Lumen edge linking. The final lumen contour is the
optimal combination of the resulting true edges. This is
Quantitative Analysis of Intracoronary OCT 1059
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).

performed using a quality score determined by the re-
sultant lumen area, length (relative to the area), and
gaps in the contour (relative to the length). For all pos-
sible combinations of edge selections, the quality score
is calculated and the combination with the highest
score results in the final lumen contour.
Postprocessing. Postprocessing is applied to over-
come possible errors introduced by the pre-processing
and those inherent to the nature of OCT imaging itself
(such as large side-branches, gaps caused by guide-
wire artifacts, etc.). The postprocessing is divided into
several subprocesses:
Contour correction and smoothing. The iterative
application of the Canny filter does not select the edges
with local maximum gradient. In a postprocessing step,
the maximum gradient search is automatically per-
formed within a 5 pixel radius from the initially found
contour. Subsequently, the contour is smoothed, weigh-
ing neighbor coordinates with the gradient magnitude
in a normal distribution.
Side-branch gaps and out-of-range borders. The
resulting luminal contours may still have gaps, which
are mostly caused by side-branches and/or guide-wire
artifacts. Furthermore, often in OCT images the lumen
border is out of range, in large vessels, or is not pro-
nounced enough to produce an edge (Fig. 2). To close
these gaps and omissions, a mathematical circular cor-
rection model is applied (Fig. 2).
Replacement of falsely detected contours (3D analy-
sis). In the case of large side-branches, heavy noise or
large parts of missing visible lumen data within the
OCT images (Fig. 3), it is still possible that the auto-
mated contour detection does not result in the desired
contour. For each consecutive image, the enclosed area
of the lumen contour is calculated. Frames showing a
relatively large deviation in areas compared to their
neighbors are labeled as incorrect. A search and substi-
tute algorithm replaces these contours by the closest
available correct contour in the longitudinal direction.
Final approval and correction. It is very difficult, or
even impossible, to develop a 100% accurate fully auto-
matic detection method in medical image processing. The
large differences found in coronary morphology will
always result in unexpected images that could not have
been foreseen during development. Therefore, the results
must be validated by an expert. To facilitate this, a user
interface was developed, similar to that used for QCU [10].
Validation
For validation of the automated method, the quanti-
tative results were compared against those derived by
application of a computer-assisted lumen detection
Fig. 1. This figure shows the results of the different process-
ing steps. (A) An original OCT image is presented. (B) The
image after the application of a Gaussian filter. (C) The
detected edges as a result of the iteratively applied Canny fil-
ter. (D) The remaining edges after application of the angle con-
straint. Short edges are removed after application of a length
threshold (E). The postprocessed and smoothed final contour
is presented in (F). [Color figure can be viewed in the online
issue, which is available at www.interscience.wiley.com.]
1060 Sihan et al.
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).

Fig. 2. To close the remaining gaps, the straight line at the right-hand side (A), a circular arc
interpolation is performed automatically using the center of gravity of the contour as the cen-
ter of the circle (B). For the radius, linear interpolation is used from r1 to r2. The repaired
contour is presented in (C).
Fig. 3. This figure presents a few different lumen morpholo-
gies which are difficult to detect correctly fully-automated. (A)
A large side-branch can be appreciated (indicated by SB). (A’)
The detected automated contour, which has been corrected by
the human observer (A00). (B) In addition to a side-branch also a
case were part of the lumen is out of range for the OCT cathe-
ter. The automated contour detection applied the automated
circular correction (B’) and the human observer corrected for
the large side-branch artifact (B00). Finally, in (C) a guide-wire
artifact is presented (C, GA). This relatively small gap is auto-
matically repaired by the correction algorithms (C’). Again, also
in this example the observer corrected for the large side-
branch (C00). [Color figure can be viewed in the online issue,
which is available at www.interscience.wiley.com.]
Quantitative Analysis of Intracoronary OCT 1061
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).

method (CURAD vessel analysis, CURAD BV, Wijk
bij Duurstede, The Netherlands) [8]. The expert (N.G.)
was blinded for the automated results.
Statistical Analysis
Quantitative data are presented as mean  standard
deviation (SD). Comparison between the methods was
performed by the two-tailed paired Student’s t-test. A
P value < 0.05 was considered statistically significant.
In addition, regression analysis and the method as pro-
posed by Bland and Altman [11] was performed.
RESULTS
In the 20 OCT cases a total of 4,167 frames were
analyzed (208  92 frames on average per patient). In
each case, the OCT images were analyzed consecu-
tively, without intervals (Fig. 4). Although the human
analysis time was not measured, it is well known that
this is usually a lengthy process. The automated
method required on average approximately 2–5 sec of
calculation time per frame. In 125 OCT frames (3%),
the automatic results had to be corrected.
The computer-assisted manual analysis showed a
mean lumen area of 5.0  2.2 mm2 versus 5.1 
Fig. 4. A typical analysis example is presented. In (A) an indi-
vidual cross-sectional image is shown and in (A’) the longitudi-
nal reconstruction of the pullback examination. In (B and B’),
the same images are presented with the automated contour
detected result superimposed. It can be appreciated that the
cross-sectional OCT images present a lot of details of the
lumen-intima morphology. (C) The regression analyses of the
area measurements of all frames with the manual results on
the x-axis and the automated results on the y-axis. Finally in
(D), all the area measurements are presented of both methods.
[Color figure can be viewed in the online issue, which is avail-
able at www.interscience.wiley.com.]
1062 Sihan et al.
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).

2.2 mm2 for the fully automatic method, with no statisti-
cal significant difference (P ¼ 0.26). The relative differ-
ence was 0.4%  1.8%. Regression analysis yielded r ¼
0.999 (Fig. 5).
Bland and Altman analysis revealed a single outlier
(Fig. 6). All other cases were within an interval of
2%. Inspection of the outlier showed a deviated con-
tour detection by the expert observer, which resulted in
a relative difference of 6%. If this outlier is disre-
garded, the results are 5.1  2.2 mm2 (expert) versus
5.1  2.2 mm2 (automated), P ¼ 0.52; relative differ-
ence 0.02%  1.1%.
DISCUSSION
This study shows that fully automatic 3D lumen
contour detection for quantitative OCT analysis is
Fig. 5. The linear regression analysis for the 20 OCT cases analysed computer-assisted
(e.g., manual) versus fully automatic. [Color figure can be viewed in the online issue, which is
available at www.interscience.wiley.com.]
Fig. 6. The relative differences according to the method as proposed by Bland and Altman
[11]. [Color figure can be viewed in the online issue, which is available at www.interscience.
wiley.com.]
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).
Quantitative Analysis of Intracoronary OCT 1063

feasible. The quantitative results are similar to those
derived manually by an expert.
To evaluate new interventional therapies, quantita-
tive imaging tools are mandatory, such as QCA [5]
and QCU [6,10]. Although a computer-assisted method
for QOCT has been presented showing excellent results
[8], the advantages of fully automatic contour detection
are obvious. It does not consume the valuable time of
an expert neither does it suffer from possible interob-
server- and intraobserver-related deviations. Further-
more, as the outlier case in this report shows, contours
which are not precisely positioned on the lumen border
can result in a relatively large deviation (in this case
6%). The human expert will most likely be more moti-
vated to analyze all images in a pullback (and not a
limited subset of images selected at by example 1 mm
intervals, e.g. every 15th frame) if the majority of the
contours are already correctly detected and the human
effort required is purely for inspection and a limited
number of corrections (Fig. 3).
OCT is a relatively new coronary imaging technique,
but it has gained considerable enthusiasm in a very
short period. It can reveal much more information of
the region around the lumen border than is achievable
with the current available ICUS technology. However,
on the down-side, the penetration depth into the coro-
nary vessel wall and present plaques is still limited (1–
2 mm). Therefore, at this moment an automated outer
vessel contour detector cannot be developed. Further-
more, because OCT cannot be used to perform coro-
nary plaque measurements, the advantages and disad-
vantages of OCT in several different clinical scenarios
have been described recently [8].
At present, several different commercial OCT systems
are available. For this study the system of Lightlab was
used [12,13]. This system has an integrated quantitative
analysis tool that is limited to single frames. Further-
more, to our knowledge the method has not been pub-
lished. This single frame approach requires a time con-
suming and operator dependent manual frame selection
at 1 mm intervals. It has been reported that depending on
the length of the analyzed region it can take up 2–4 hr to
complete an analysis [8].
Independent third party quantitative software tools
are, to our knowledge, not yet reported, except for one
study presented by Tanimoto et al. [8]. In this article, a
computer-assisted dedicated OCT analysis tool was
reported, showing good inter and intraobserver quantita-
tive results (1.57%  0.05% for lumen areas). However,
only well-visualized OCT images were included. Images
suffering from motion-induced artifacts, dissections, and
side-branches––hampering analysis––were excluded
(9%). In this study, all available imaging data (real world
data) was analyzed and no exclusions were made. To our
knowledge, to date, no other reports have been published
concerning fully automatic QOCT methods.
Limitations
Unfortunately, because of the nature of OCT, the
penetration depth is currently to low to be able to visu-
alize the coronary vessel wall in diseased segments and
therefore only the coronary lumen could be quantified
in this study. Developments of newer OCT methods,
such as OFDI, and application of other light sources,
could possibly enhance the penetration depth making it
hopefully possible to visualize advanced coronary
plaques in the near future.
The number of cases included in this study is lim-
ited, a larger number of cases must reveal if the excel-
lent score of fully automatic detection in 97% of the
images can be maintained for larger populations. How-
ever, we evaluated almost 4,200 individual OCT
images of very different lumen morphology, because
of the high spatial resolution.
Future Developments
This full-automated approach to quantify the coronary
lumen by OCT is the first step towards further develop-
ments of highly anticipated additional quantification
tools. On the requirements list are currently: detection of
stent struts, protrusion of plaque contents through stent
struts, in-stent thrombi, fibrous caps (detection and thick-
ness measurements), and plaque composition. If these
requirements could be detected automatically remains
topic for further research and developments. However,
measurements of in-stent intima hyperplasia, lumen-ec-
centricity and -remodeling (if base-line and follow-up
OCT measurements are available) are already possible
using the automated approach (for lumen) in combination
by computer-assisted tools (for stent contouring) [8].
The described method has been developed in a
generic mathematical research environment on a nor-
mal desktop personal computer running Microsoft Win-
dows. The processing time of 2–5 sec could most
likely be reduced in the near future if the software
were to be ported to a dedicated QOCT environment.
Many of the OCT images suffer from motion artifacts,
which are caused by the low temporal resolution of cur-
rent OCT systems as compared to the relatively rapid
motion of the heart. The heart motion also causes the
saw-tooth shaped appearance of the coronary vessel wall
in the longitudinal reconstructions (Fig. 4). These motion-
related artifacts could probably be overcome by the appli-
cation of optical frequency domain imaging (OFDI) to
coronary vessel imaging. However, these systems are still
in the research phase and not commercially available yet.
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).
1064 Sihan et al.

CONCLUSION
This study shows that fully automatic lumen contour
detection in OCT images is feasible with only a few
contours showing an artifact (3%) that can be easily
corrected. This QOCT method may be a valuable tool
for future coronary imaging studies incorporating OCT.
REFERENCES
1. Regar E, Schaar JA, Mont E, Virmani R, Serruys PW. Optical
coherence tomography. Cardiovasc Radiat Med 2003;4:198–
204.
2. Prati F, Zimarino M, Stabile E, Pizzicannella G, Fouad T,
Rabozzi R, Filippini A, Pizzicannella J, Cera M, De Caterina R.
Does optical coherence tomography identify arterial healing
after stenting? An in vivo comparison with histology, in a rabbit
carotid model. Heart 2008;94:217–221.
3. Guagliumi G, Sirbu V. Optical coherence tomography: High
resolution intravascular imaging to evaluate vascular healing after
coronary stenting. Catheter Cardiovasc Interv 2008;72:237–247.
4. Ormiston JA, Serruys PW, Regar E, Dudek D, Thuesen L,
Webster MW, Onuma Y, Garcia-Garcia HM, McGreevy R,
Veldhof S. A bioabsorbable everolimus-eluting coronary stent
system for patients with single de-novo coronary artery lesions
(ABSORB): A prospective open-label trial. Lancet 2008;371:
899–907.
5. Haase J, Escaned J, van Swijndregt EM, Ozaki Y, Gronenschild
E, Slager CJ, Serruys PW. Experimental validation of geometric
and densitometric coronary measurements on the new genera-
tion Cardiovascular Angiography Analysis System (CAAS II).
Cathet Cardiovasc Diagn 1993;30:104–114.
6. Mintz GS, Nissen SE, Anderson WD, Bailey SR, Erbel R, et al.
American College of Cardiology Clinical Expert Consensus
Document on Standards for Acquisition, Measurement and
Reporting of Intravascular Ultrasound Studies (IVUS). A report
of the American College of Cardiology Task Force on Clinical
Expert Consensus Documents. J Am Coll Cardiol 2001;37:
1478–1492.
7. von Birgelen C, Mintz GS, Nicosia A, Foley DP, van der Gies-
sen WJ, Bruining N, Airiian SG, Roelandt JRTC, de Feyter PJ,
Serruys PW. Electrocardiogram-gated intravascular ultrasound
image acquisition after coronary stent deployment facilitates on-
line three-dimensional reconstruction and automated lumen
quantification. J Am Coll Cardiol 1997;30:436–443.
8. Tanimoto S, Rodriguez-Granillo G, Barlis P, de Winter S,
Bruining N, Hamers R, Knappen M, Verheye S, Serruys PW,
Regar E. A novel approach for quantitative analysis of intra-
coronary optical coherence tomography: High inter-observer
agreement with computer-assisted contour detection. Catheter
Cardiovasc Interv 2008;72:228–235.
9. Canny J. A computational approach to edge detection. IEEE
Trans Pattern Anal Mach Intel 1986;8:679–714.
10. Hamers R, Bruining N, Knook M, Sabate M, Roelandt JRTC. A
novel approach to quantitative analysis of intravascular ultra-
sound images. Comput Cardiol 2001:589–592.
11. Bland JM, Altman DG. Statistical methods for assessing agree-
ment between two methods of clinical measurement. Lancet
1986;1:307–310.
12. Matsumoto D, Shite J, Shinke T, Otake H, Tanino Y,
Ogasawara D, Sawada T, Paredes OL, Hirata K, Yokoyama M.
Neointimal coverage of sirolimus-eluting stents at 6-month fol-
low-up: Evaluated by optical coherence tomography. Eur Heart
J 2007;28:961–967.
13. Takano M, Inami S, Jang IK, Yamamoto M, Murakami D,
Seimiya K, Ohba T, Mizuno K. Evaluation by optical coherence
tomography of neointimal coverage of sirolimus-eluting stent
three months after implantation. Am J Cardiol 2007;99:1033–
1038.
Catheterization and Cardiovascular Interventions DOI 10.1002/ccd.
Published on behalf of The Society for Cardiovascular Angiography and Interventions (SCAI).
Quantitative Analysis of Intracoronary OCT 1065