1.1 Coronary arteries disease . . . . . . . . . . . . . . . . . . . . . . . . . . . [608735]

Contents
1 Introduction 2
1.1 Coronary arteries disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Intravascular imaging modalities . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Motivation 4
3 Methods 5
3.1 OCT processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.1 Diameter sequence computation . . . . . . . . . . . . . . . . . . . . . 6
3.1.2 Unreliable slices detection . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.3 Rejected contours reconstruction . . . . . . . . . . . . . . . . . . . . 9
3.2 Angiogram processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Centerline points distribution . . . . . . . . . . . . . . . . . . . . . . 9
3.2.2 Bifurcation removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Angio-OCT diameter co-registration . . . . . . . . . . . . . . . . . . . . . . 12
3.4 3D reconstruction of OCT . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4.1 Pointset translation to tridimensional space . . . . . . . . . . . . . . 15
3.4.2 Meshing the vessel wall . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.5 Hemodynamic analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4 Co-registration app 17
5 Conclusions 17
1

1 Introduction
1.1 Coronary arteries disease
Coronary artery disease(CAD) is a condition which manifests through a reduction of the
blood
ow to cardiac muscles due to a narrowing of the blood vessel. This narrowing, also
called a stenosis, can be generated by deposits of tissues such as fatty, calcied, brous,
etc, on the vessel wall. Such deposits are called atherosclerotic plaques which may grow in
dimension keeping the heart mucles from recieving the right quantity of blood, thus harming
the heart pumping eciency. Further, the entire circulatory system may malfunction leading
to a wide range of physiological abnormalities, and, as the blood
ow decrease, a heart attack
is very likely to occur.
1.2 Intravascular imaging modalities
Cardiac catheterization is a family of invasive procedures widely used in coronary artery
disease(CAD) diagnostication and treatment, on which a thin
exible tube (called catheter)
is guided through the blood vessels to the heart. Based on the type of the catheter being
used, the clinician will be able to retrieve useful information about the circulatory system
and cardiac cycle based on which he can assess the functionality of the heart.
Access to the blood vessels is gained most commonly through the femoral artery, to look at
the left side of the heart and at the arterial system; or the jugular or femoral vein, to look
at the right side of the heart and at the venous system[1].
Angiography is a X-Ray imaging technique that uses a radio-opaque dye to see inside an
organ, a heart chamber or the lumen of a blood vessel.
Figure 1: Coronary angiography of the left coronary artery (LCA).
In a coronary angiography, during cardiac cathterization, a dye is injected into the ves-
2

sel through the catheter, mixing up with blood. This special kind of dye is also known as
contrast medium or contrast agent due to the property of having a high attenuation power,
thus becoming visible on X-Ray images using techniques such as
uoroscopy. Hence, in
X-Ray images, the lumen of the artery will be denoted by dark pixels compared to the much
brighter background, thus providing a good contrast (Figure 1).
In order to enhance the information provided by angiograms, two images of the same
vessel will be aquired in the same time, but from dierent angles. This is required since
the lumen does not have a cylindric shape, especially due to the presence of atherosclerotic
plaques.
Intravascular Optical coherence tomography(IVOCT) is a light based imaging modality
that generates high-resolution cross-sectional images of tissue microstructure [2]. From the
catheter tip, a near infrared light is emmited to the vessel wall, and then, structural infor-
mation as a function of depth can be retrieved by measuring the delay time of the light that
is re
ected from tissue. Following this procedure, high resolution cross-sectional images are
obtained (Figure 2), providing a very accurate view of the lumen boundaries and internal
structures of the vessel wall.
Figure 2: Cross sectional image aquired with OCT.
Percutaneous coronary intervention(PCI) is a non surgical procedure widely used to treat
coronary artery disease. Guided by an angiography, the clinician reach the obstructed area
with a balloon catheter which is going to be in
ated in order to restore the normal blood

ow. If needed, devices called stents are placed in the stenosed area to keep the lumen
expanded after the balloon's catheter withdraw.
3

2 Motivation
Calcied atherosclerotic plaques undergoing PCI are associated with higher procedural com-
plcations compared to non-calcied plaques. Due to the rigidity of calcium, the stent can not
be optimally expanded which leads to a series of complications such as unachievable luminal
gain, longer procedures, the need of multiple stents, etc. Even post-PCI, complications as
restenosis and stent thrombosis are very likely to occur due to the stent poor expansion.
Despite the fact that coronary angiography is to goal standard of imaging guiding PCI,
only approximately 50% of all calcied plaques can be detected by this modality (moderate
and severe). Moreover, image quality is crucial for calcied plaques detection on coronary
angiographies (Figure 3a), extending the patient x-Ray exposure time along with the ad-
ministration of a higher dose of contrast medium.
(a) Calcied plaque detected on coronary angiog-
raphy.
(b) Calcied plaque detected on OCT.
Figure 3: Calcied plaque visualization.
At the opposite pole, optical coherence tomography provides a way better visualization of
calcied plaques, being the gold standard -at the moment of writing this thesis- for thickness
and angle assessment. In OCT cross-sectional images, calcied plaques are characterized
by sharp boundaries deliniating a darker homogeneuous region 3b. However, the major
drawback of this modality is that it cannot be used interventionally. More, an OCT catheter
is expensive compared to an angiography catheter.
Therefore, each of the upmentioned modalities has its own advantages which makes itself
widely adopted to address a certain task, but, the clinician often needs cumulated informa-
tion in order to provide an accurate diagnosis or treatment. Hence, the current work propose
a framework for co-registration between OCT and angiography. Based on the information
extracted from both images, we register the two modalities, thus providing enhanced infor-
mation to support the clinician. Following this approach, micrometric resolution imaging
provided by OCT may be used interventionally in combination with coronary angiographies
to reduce the risks associated with unidentied calcied atherosclerotic plaques during a
PCI.
4

3 Methods
The underlying concept of the proposed framework is 1D sequence registration. By means
of image processing, we extract a mean diameter from each cross sectional slice in OCT,
obtaining an uniformly distributed 1D sequence, which can be registered with the same
kind of information extracted from angiographies. In IVOCT, cross-sectional 2D images are
aquired while the cathehter is pulled back from the target artery at a constant speed such
that the sampling distance equals 0 :2mm.
3.1 OCT processing
Each IVOCT slice is composed by multiple regions (see 2). The upper side of the image
holds plain text information inputted by clinician before starting the procedure (e.g. artery
name, aquision date and time, location, etc.). The lower side of the image contains imagistic
data showing a transversal section of the vessel. Even the IVOCT provides a sequence of
2D cross-sectional slices, this transversal section holds 3D information such as length of the
vessel, position of the current frame, evolution of the lumen diameters as a function of depth,
etc. To that extent, this rgion may be seen as an imaginary z-axis.
The center of the image is the region of interest for the current work, since all the
information regarding cross-sectional diameters can be found within it. It can be directly
extracted from the entire slice by using information available in the DICOM le.
Digital Imaging and Communications in Medicine (DICOM) is the standard for com-
munication and management of medical imaging information and related data. DICOM is
most commonly used for storing and transmitting medical images enabling the integration of
medical imaging devices such as scanners, servers, workstations, printers, network hardware,
and picture archiving and communication systems (PACS) from multiple manufacturers. It
has been widely adopted by hospitals.[3]
Therefore, each OCT slice as presented to the clinician is made up by three major
regions, each of them having its own characteristics. It is very important to carefully identify
them while processing the OCT to provide an accurate result since each of them has a
dierent spatial resolution, so any spatial knowledge extracted from image solely rely on the
correctudiness of choosing the scaling factor.
Hence, we extract the region of interest (see 4) on which we are going to perform seg-
mentation in order to extract the lumen contour. A contour can be seen as a curve joining
all the continuous points along the boundary, having the same color and intensity.
OCT slices are RGB high resolution images, on which the lumen is already deliniated
by a continuous enclosed green curve. Since the lumen boundary is the only thing we need
to compute diameters, all the processing is done in the green channel. First, a threshold is
applied on the image, resulting a binary image on which the lumen segmentation is already
available. The threshold value has been empirically set to 90 % from the maximum intensity
pixel in the green channel previously extracted. In the second step, all the contours visible
5

Figure 4: Lumen segmentation in OCT frames.
in the region of interest (ROI) are being extracted, and, intuitively, the contour having the
maximum area is the one deliniating the lumen of the coronary artery (see 4).
Further, for each independent contour we compute the centroid using the spatial image
moments provided by OpenCV library1.
Cx=M10=M 00;Cy=M01=M 00 (1)
3.1.1 Diameter sequence computation
In order to obtain the lumen diameter corresponding to each cross-sectional slice, two dif-
ferent approaches have been deployed.
Mean distance between diametrically opposed sampled contour points
As depicted in the next sections, a 3D reconstruction based on OCT contours is implemented
by means of triangulation. This feature impose keeping a list of ordered contour points
sampled from the original lumen contour. Hence, the application holds a list of ordered points
sampled every 5 degrees, so, each cross-sectional contour will be represented by 72 samples.
This sampling procedure is based on modied Breshenham drawing lines algorithm[4]. As
input data, the algorithm has a starting point, which is the centroid previously computed
(equation 1), and the angle (2[0,360] with step 5) at which to draw the line. The
procedure goes on until the line intersect a non-zero pixel, returning its location on the
pixmap.
Therefore, a distance between each diametrically oposed pair of points can be computed
using the second order Euclidean norm. In order to get the spatial distance and not the
distance in the image space, coordinates have been multiplied by the appropiate scale factors
provided in DICOM metadata. Following this procedure, for each cross-sectional lumen
contour we will end up having 36 diameters sampled every 5 degrees, from which we can
compute the mean diameter used for co-registation (Figure5).
1OpenCV (Open Source Computer Vision) is a cross platform library mainly used in developement of
real time computer vision applications.
6

d(A;B) =p
(A1B1)2(A2B2)2 (2)
whereA1-xcoordinate of point sampled angle multiplied by the xscale factor, A2-y
coordinate of point sampled angle multiplied by the yscale factor, B1-xcoordinate of
point sampled angle + 180 multiplied by the xscale factor, B2-ycoordinate of point
sampled angle + 180 multiplied by the yscale factor.
Diameter computation from contour area.
Contour area can be computed using the function contourArea provided by OpenCV, or can
be given by the rst moment of contour segmentation image M[0m000].
Therefore, for each cross-sectional slice we compute the area, based on which an estimation
of mean diameter can be done. Assuming that the lumen has the shape of a circle, we can
estimate a mean diameter using the following equation:
d=p
4A= (3)
wheredis an estimated mean diameter and Ais the area of the lumen in the current slice.
Figure 5: Diameter sequence extraction pipeline.
Both approaches have been implemented to estimate the vessel cross-section diameter,
but, during experiments on multiple data-pairs, the rst one has shown more robustness in
co-registration.
3.1.2 Unreliable slices detection
The software controlling the OCT equipment has the functionality of deliniating the lumen
boundary by a green continuous curve (6a). Anyway, when the aquisition is poor or when the
aquisition is done nearby a bifurcation, the boundary information provided by the equipment
is not reliable (6b). Moreover, it can be even missing from some slices (6c), in which case
the proposed segmentation framework will fail.
To overcome this issue, the user has the posibility of rejecting slices from OCT manually
by pressing a button, or semi-automatically. In the latest, a similarity measure between
7

(a) Good segmentation on
OCT.
(b) Wrong segmentation due
to a bifurcation.
(c) Missing annotation.
Figure 6: Lumen segmentation fail cases.
each two consecutive slices is computed. When this similarity measure is below a certain
threshold, the user is being prompted regarding the correctudiness of the segmentation. This
is possible because, during the pullback, a slice is aquired every 0 :2mm, so, intuitively, the
lumen area cannot vary that much in such a short distance. When the user rejects on slice,
he will be prompted about each of the following ones, until he use the accept button. At this
point, the algorithm retakes the control, further checking the similarity between consecutive
slices.
Similarity is computed between two binary images iandj, where each backgound pixel
has a null value while pixels whitin the lumen will be equal to 1. Based on this information,
a variance based distance measure[5] can be computed using equation 4:
Dvari= (b+c)=4(a+b+c+d) (4)
wherea;b;c anddare computed accordingly to table 1.
Based on this quantity, we dene a similarity measure as follows:
Svari= 1Dvari (5)
whereSvarishows how similar the images iandjare. IfSvariwill have a value lower than
0.75, the contour will be marked as possible incorrect and the user will be employed in taking
the nal decision.
Table 1: Similarity indicators computation.
i,j 1(Presence) 0(Absence)
1(Presence) a=i_j b =ij
0(Absence) c=ij d =ij
In the end, all contours within rejected slices will be reconstructed by means of interpo-
lation.
8

3.1.3 Rejected contours reconstruction
After the user is done with the selection procedure, all the contours will be reconstructed by
interpolating the sampled contour points of the previous and the following accepted contour
(equation 6). Thus, incorrect segmentations will be replaced by linearly evolving ones.
mCnt [i] = (1)pCnt [i] +nCnt;i = 1;N (6)
Where,
Nis the number of consecutive rejected slices,
pCnt is the previous accepted contour,
nCnt is the following accepted contour,
is the interpolation step, being equal to
= 1=(N+ 1) (7)
At each iteration, the last accepted contour – pCnt – is set to the previous interpolated
one,N=N1 andwill be recomputed. After interpolation is done, all the reconstructed
slices are being processed again in order to update the all the stored data such as diameters,
centroids, etc.
3.2 Angiogram processing
A coronary angiography consists in two syncronized images (g 7), simultaneously aquired
at dierent angles. By detecting a set of corresponding points, manually or automatically,
a 3D reconstruction can be computed2.
To the extent of the current co-registration framework, having a 3D centerline extracted
from angiograms along with corresponding diameters is sucient, so this information will
be loaded as input data from a xml le.
3.2.1 Centerline points distribution
Anyway, angio centerline is not uniformly distributed in terms of distance between each
two consecutive points. On the other hand, as OCT slices have been aquired once every
0:2mm, their corresponding centroids are uniformly distributed, so an inconsistency occur
between the two modalities, which leads to an oset appeareance. To overcome this issue,
the centerline extracted from angriogram may be seen as a spline interpolation function3.
2The 3D angiography reconstruction implementation is not a part of this work, so all the information
regarding angiography represents input data for the proposed framework.
3In this special kind of interpolation, the interpolator is a special kind of piecewise polinomial called
splines, thus enabling the estimation of complex curves through relatively low level polinomials.
9

Figure 7: 3D reconstruction from angio views.
Having a number of N points describing the centerline [( xi;yi) ,i= 0;1;:::N ], also called
control points, we interpolate between each two consecutive points ( xi;yi) and (xi+1;yi+1).
One of the advantages of spline interpolation is that the derivatives are taken into an account
when computing the interpolators, resulting a smooth curve (gure 8)). As the centerline
points in agiography are three-dimensional, we proceed at applying a 1D spline interpolation
three times, one for each dimension of the points.
Figure 8: Interpolation: spline vs. linear.
Therefore, by using spline interpolation we create a large number of points within the
same vessel segment. Further, new uniformly distributed centerline points can be choosen.
Listing 1: Angio centerline sampling algorithm.
10

t o t a l D i s t a n c e = 0
f o r index = 1 to ( n points1) do
distance = compute distance ( point [ index ] , point [ index +1])
t o t a l D i s t a n c e += distance
i f t o t a l D i s t a n c e 0.2>0 then
hold point [ index +1]
t o t a l D i s t a n c e = 0
continue
end i f
end f o r
Following this algorithm, we get a good precision in sampling equidistantly centroids by
increasing the number of generated points, which may lead to poor eciency in terms of
computation time. To that extent, we only generate a large enough number of points to
provide a distance between consecutive points larger than 0 :2mm, on which we apply the
following strategy.
Letdtdenote the desired distance (0 :2mm) between each two consecutive points. We
compute the actual distance using the equation 2, denoted by d.
Assuming that the starting point has the coordinates ( xi;yi) and the point following it
is denoted by ( xi+1;yi+1), we can compute the coordinates of the point ( xt;yt), situated at
exactlydtmmfrom the starting point using the following equations:
xt= (1t)xi+txi+1;yt= (1t)yi+tyi+1 (8)
Where,
t=dt=d (9)
Following this approach, each two consecutive points of the centerline will be exactly
0:2mm far away one from each other, removing the oset between the two sequences, and
thus facilitating the co-registration.
3.2.2 Bifurcation removal
Since the coronary arteries detected in agiograms may have bifurcations (gure 9 – left), it
is important to eliminate them before trying to co-register the information extracted from
the two modalities. This is so important because the OCT aquisition always corresponds to
the only one path. Hence, in order to eliminate bifurcation we use the following procedure:
First, we extract the information available in angiographies. This information is held
as a coronarian tree where each node corresponds to a reconstructed point. Each node
has information regarding it's parrent, lumen radius, position, etc.
We order the nodes such that each child node is correctly associated with its parrent
node. A parrent node may have multiple child nodes.
11

A bifurcation occurs when a parrent node has multiple childs. Based on this, we nd
all the bifurcations.
For each bifurcation, we only keep the longest one.
At the end of this steps, all the bifurcations will be removed making the centerline look
as the one depicted in gure 9 (right).
Figure 9: Angio centerline's bifurcations removal.
3.3 Angio-OCT diameter co-registration
In order to register the information extracted from each modality, a series of similarity
measures have been implemented and assessed. The underlying concept of this measures is
a distance computation between the two sequences.
The co-registration is done by determining the highest similarity between a xed sequence
and a moving one, using one of the follwing measures:
Jaccard
d(u;v) =CTF+CFT
cTT+CFT+CTF(10)
Compare histograms
d(u;v) =tobewritten (11)
Dice
d(u;v) =CTF+CFT
2cTT+CFT+CTF(12)
12

Figure 10: Co-registration pipeline.
Euclidean
d(u;v) =jjuvjj2 (13)
Cosine
d(u;v) = 1uv
jjujj2jjvjj2(14)
Correlation
d(u;v) = 1(uu)(vv)
jjuujj2jjvvjj2(15)
Canberra
d(u;v) =X
ijuivij
juij+jvij(16)
Chebyshev
d(u;v) = max
ijuivij (17)
Bray-Curtis
d(u;v) =P
ijuivijP
ijui+vij(18)
Minkowski distance
d(u;v) =jjuvjjp= (X
ijuivijp)1
p (19)
Cityblock (Manhattan) distance.
d(u;v) =X
ijuivij (20)
Squared-chord
d(u;v) =X
i(pupv)(pvpu) (21)
Pearson
d(u;v) =X
i((uv)(uv))i
vi(22)
13

Neyman
d(u;v) =X
i((uv)(uv))i
ui(23)
ChiSquared
d(u;v) =X
i((uv)(uv))i
ui+vi(24)
Topsoe
d(u;v) =X
iuilog2ui
ui+vi+vlog2vi
ui+vi(25)
Wave Hedges
d(u;v) =X
i1minui;vi
maxui;vi(26)
Czekanowski
d(u;v) =P
ijuivijP
iui+vi(27)
Intersection
d(u;v) = 0:5X
ijuivij (28)
Soergel
d(u;v) =P
ijuivijP
imaxui;vi(29)
Clark
d(u;v) =r
juvj
u+vjuvj
u+v(30)
To compare the similairy for each diameter, a zero padding having the same length as
OCT diameter sequence is added before angio diameters. Further, OCT diameter sequence
play the role of a moving window overlaying on angio sequence in an interative fashion.
At each iteration, we get the intersection of diemeters (added padding is not taking into
a consideration) on which a similarity measure is computed. For example, in the rst
iteration, we compute the distance between the rst diameter in angio sequence and the last
diameter in OCT sequence, in the second iteration we compute the distance between rst
two diameters in angio sequence and last two in OCT, and the procedure continues until
we compute the distance between the last diameter in angio and the rst diameter in OCT.
At each step we compute one of the above similarity measures on the intersection diameters
from each sequence. Further, the algorithm rejects all the similarity scores below a certain
threshold, and then, it recomputes the scores in such a way that the longer sequences will
have a higher weight (eq 31).
s=scrtd
dmax(31)
14

Figure 11: Maximum similarity result.
3.4 3D reconstruction of OCT
As the information regarding the luminal structures extracted from OCT is more accurate
than the one extracted from Angio, while the Angio is better than OCT in terms of artery
layout assessment, a combination between the two may signicantly enhance the imaging,
thus supporting doctors in clinical work
ows. Hence, the current work propose a 3D recon-
strction of luminal contours extracted from OCT based on the 3D spatial information from
angiograms.
3.4.1 Pointset translation to tridimensional space
As the OCT data resulted after processing is described as a 2D pointset representing the
lumen contours, in the proposed framework it is translated to 3D space by taking advantage
of the 3D centroids available in angiography. This can be done followind the procedure
below:
First of all, an extra-dimension is added to each point, playing the role of the z-axis.
In other words, each point ( x;y) in OCT will become ( x;y;i ), whereiis the index of the
cross-sectional slice from which ( x;y) has been extracted.
Secondly, we compute the relative coordinates of each contour point with respect to the
corresponding centroid.
relCnt [i] =octcentroid [i]octcnt[i];i=1;N (32)
whereNis the number of OCT slices.
Further, each OCT contour is centered on the corresponding angio centroid using the
following equation
nCnt [i] =angio centroid [i]relCnt [i];i=1;N (33)
At this point, all the OCT contours are centered on 3D angio centerline, but their
orientation is wrong, generating a sequence of parallel planes. In order to x the orientation
15

Figure 12: OCT contour points translated in angio space.
issue, for each contour, we consider the normal vector as being the vector between the current
angio centroid and the one following it.
n=ci+1ci (34)
wherenis the desired normal vector and ciis theithangio centroid.
Further, a transformation matrix needs to be computed in order to re-orient OCT contour
points with respect to the angio centerline. Since all the points centered on angio centerline
using equation 33 only form a bunch of parralel planes, the normal vector of each of those
contours is denoted as i= (0;0;1). Thus, we can compute the following vectors:
u=ni (35)
whereuwill hold the x-axis rotation parameters.
v=un (36)
wherevwill hold the y-axis rotation parameters.
In order to compute the transformation matrix, vectors u;vandnneed to be normalized
using the following formula:
16

x=xp
x2
1+x2
2+x2
3(37)
Therefore, we dene the transformation matrix as:
M=2
64u1v1n1c1
u2v2n2c2
u3v3n3c33
75 (38)
where (c1;c2;c3)Tare the coordinates of the angio centroid on which we translate the
OCT contour.
Using the matrix 38, we can transform a point x= (x1;x2;x3) tox0= (x0
1;x0
2;x0
3) such
that its orientation will be xed with respect to the angio centerline (see gure 12).
2
6664×0
1
x0
2
x0
3
13
7775=2
64u1v1n1c1
u2v2n2c2
u3v3n3c33
752
6664×1
x2
x3
13
7775(39)
3.4.2 Meshing the vessel wall
3.5 Hemodynamic analysis.
What solvers do we use, what info they carry, etc …
4 Co-registration app
5 Conclusions
17

List of Figures
1 Coronary angiography of the left coronary artery (LCA). . . . . . . . . . . . 2
2 Cross sectional image aquired with OCT. . . . . . . . . . . . . . . . . . . . . 3
3 Calcied plaque visualization. . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 Lumen segmentation in OCT frames. . . . . . . . . . . . . . . . . . . . . . . 6
5 Diameter sequence extraction pipeline. . . . . . . . . . . . . . . . . . . . . . 7
6 Lumen segmentation fail cases. . . . . . . . . . . . . . . . . . . . . . . . . . 8
7 3D reconstruction from angio views. . . . . . . . . . . . . . . . . . . . . . . . 10
8 Interpolation: spline vs. linear. . . . . . . . . . . . . . . . . . . . . . . . . . 10
9 Angio centerline's bifurcations removal. . . . . . . . . . . . . . . . . . . . . . 12
10 Co-registration pipeline. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
11 Maximum similarity result. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
12 OCT contour points translated in angio space. . . . . . . . . . . . . . . . . . 16
18

List of Tables
1 Similarity indicators computation. . . . . . . . . . . . . . . . . . . . . . . . . 8
19

References
[1] Angiography, Coronary Angiography or arteriography | Wikipedia, The Free Encyclo-
pedia , 2004. [Online; accessed 21-March-2019].
[2] G. J. Tearney and et al., Consensus Standards for Acquisition, Measurement, and Re-
porting of Intravascular Optical Coherence Tomography Studies . Elsevier Inc., 2012.
[3] DICOM, \Digital imaging and communications in medicine | Wikipedia, the free en-
cyclopedia," 2010. [Online; accessed 08-May-2019].
[4] K. I. Joy, \On-line computer graphics notes – breshenham's algorithm," University of
California, Davis , 1999.
[5] S.-S. Choi, S.-H. Cha, and C. Tappert, \A survey of binary similarity and distance
measures," Department of Computer Science, Pace University, New York, US , 2010.
20