X Ray Analysis Of Carotid Plaque

Published Date: 02 Nov 2017

Group 13

X-ray: Analysis of carotid plaque

Jussi Nupponen

Kaisa Liimatainen

185482

April 15, 2013

Introduction

Atherosclerosis is a disease where accumulation of fatty materials thickens vessel walls and

causes luminal stenosis. The stenosis in carotid arteries can cause ischemic stroke. Plaques

can be divided into three groups according to Houns

eld unit (HU) values: lipid,

brous

and calci

ed plaques.

1

Each plaque type has di#erent risk of ischemic events.

2

Computed

tomography angiography (CTA) can accurately re

the extent of carotid stenosis.

3

This essay focuses on X-ray computed tomography instead of traditional X-ray imaging. Multidetector computed tomography (MDCT) is an advanced method for acquiring

high quality CTA images. In all the articles concerning CT this essay refers to MDCT

machines were used. Amount of detector rows in modern CT machines is usually 16 or

64 depending on the manufacturer: GE Healthcare provides a machine with 16 detector

rows whereas Philips machine has 64 detector rows.

4

This essay is based on several scienti

c articles and an interview of Florentino Santos,

TUT, who is currently developing a method for automatic analysis of carotid plaque from

CTA images.

Implementation scale

CTA is a routine clinical method for detecting carotid plaque. Analysis is currently done

by physicians but automatic methods for detecting carotid plaque are being developed.

Thus CTA images of carotis are also needed for scienti

c research.

4

Imaging protocol

Imaging protocol of CTA varies depending on medical center and equipment. The values

of imaging parameters in this chapter are collected from di#erent sources, not just a

single protocol. Thus they also give an answer to the question of requirements of imaging

conditions.

CTA scan is generally obtained from aortic arch to the circle of Willis.

5

Tube voltage

for CTA is usually 120 kVp but tube current varies a lot. The amount of current should be

determined by the patient's body size. Bigger current delivers bigger amount of radiation

to the patient.

6

Usually the tube current is between 180 and 240 mAs.

1;3;5;7

.

Collimation presents the thickness of a slice along the longitudinal axis of the patient.

Thus it also determines the minimum slice thickness in reconstructed image. Usual collimation for 16-slice scanner is 16 x 0.75 mm and for 64-slice scanner 64 x 0.6 mm.

1;6;7;8

Pitch is generally determined by tabel travel per rotation divided by collimation.

6

The

value of pitch is usually one or below if overlapping is wanted.

1;5

The rotation time for 16-slice scanner is 0.5 or 0.6 seconds and for 64-slice scanner

0.4 seconds

3;5

. The recommended settings are smallest possible slice thickness and fastest

possible gantry rotation, although fastest rotation may create some artifacts.

5

A iodinated contrast agent like Iopamidol (370 mgI/ml) or Iodixanol (320 mgI/ml) is

used in CTA for optimizing vascular visualization. The injection rate is usually 4 ml/s,

amount injected is 100 ml and injection site is right arm.

5;7

The protocol for image reconstruction de

nes image matrix size as 512 x 512 and slice

thickness 0.5 - 1.0 mm. For slice thickness of 1.0 mm the slice increment of 0.5 - 0.6 mm

is used, for smaller thicknesses the increment is usually less. Field of view varies from 100

to 190 mm.

1;7;8;9

1

3D reconstruction and image enhancement

CTA images are reconstructed according to reconstruction protocol presented in previous chapter and with a certain reconstruction method. Popular methods are maximal

intensity projection (MIP), multiplanar reconstruction (MPR) and perspective volume

rendering (VR).

5;10

In MIP only the pixels with maximum intensity are shown which

enables identifying bright objects embedded inside another object

11

. For creating MPR

images the mean density of all pixels is selected. VR displays all pixels in a 3D display

thus using all the data. MIP and MPR methods are better for detecting calci

es plaques

than VR.

5

Physicians often use smoothing for CTA images which can remove small indications

of carotid plaque. This is a problem since even small lipidic plaque can cause an attack if

it releases into blood stream. In development of automatic methods raw CTA images are

used to get optimal results.

4

If motion artifacts appear in data set and thus images are noisy, images can be

smoothed with 50% overlap. This problem does not usually appear with fast scanners

if minimum slice thickness is used.

5

Image segmentation

The lumen segmentation is the

rst step in assessing stenosis and segmentation of vascular structures is very valuable for diagnosis assistance and treatment and surgery planning.

7;12

Most angiographic clinical routines are still operator dependent which causes

inter-operator variation and requires excessive amount of the operators time.

4;12

Vascular

segmentation is challenging because blood vessels soft tissue that form complex networks

and are often tightly connected to surrounding organs or bones. The bifurbication make

the segmentation of the vessels hard because it involves setting limits to external and

internal carotid distance. Segmenting the bifurbication involves decision making in the

matter of when are the two carotid arteries to be considered two separete vessels.

4;7

According to Kirbas and Quek vessel segmentation algorithms and techniques can

be divided into six main categories which are pattern recognition techniques, modelbased approaches, tracking-based approaches, articial intelligence-based approaches, neural network-based approaches, and miscellaneous tube-like object detection approaches.

13

Here the

rst two approaches are discussed in more detail and the others only in general

level.

The pattern recognition techniques consist of many subcategories of which one popular

category is region growing algorithms

4;13

. A region growing algorithm takes one seed pixel

and selects incrementally adjacent pixels according to some prede

ned criteria which is

usually based on similarity or proximity of the pixels. A region growing algorithm makes

assumptions that if pixels are close enough to each other and have similar values then they

probably belong to a same object. Region growing algorithms usually require operator

dependent seed point as a seed pixel.

4;13

One model based approach is an active contour technique. The active contour technique

nds object contours by using set of forces that could be external forces which are

derived from the image and internal forces that are based on models that set rules to

contour geometry and its regularity

12

. An active contour technique that is optimized for

vessels is called snakes. Internal forces set smoothing constraints to the image and external

forces pull the snake closer to lines and edges. Other model based approach is parametric

models that use prede

ned parameters that set rules for the objects of interest. Some

2

parametric models use overlapping ellipsoids or circular objects as a model of a vessel

but these models are too simple to correctly recognize pathological irregular shapes and

vessel bifurbications. Optimized solution can be achieved by e.g. Markov Random Field

algorithm.

13

Tracking based approaches detect vessel centerlines or boundaries by investigating

pixels in an orthogonal direction compared to tracking direction. Arti

cial Intelligencebased approaches di#erent knowledge sources for image segmentation, e.g. knowledge

about acquisition device and a model of the vessels in question. Neural network approach

uses pattern recognition with a learning dataset that is used to train the classi

er to

recognize wanted objects. Miscellaneous tube-like object detection approaches consists of

techniques that are suitable for vascular segmentation but are mainly developed for other

tube-like objects.

13

Image analysis

The images produced by the CTA can be analysed with methods that include maximal

intensity projection (MIP), multiplanar reconstruction (MPR) and volume rendering.

5

Image analysis consists of segmenting vessels,

nding bifurbication, classifying plaque

hardness. Plaque can be classi

ed to three groups that are calci

cation,

brous tissue and

lipid core. HU values for di#erent types of plaque are presented in the Parameters section.

Calci

ed areas are detected from lumen by setting cut-o# value to 200 HU. Lipid core is

detected if area of interest presents only necrotic tissue, hemorrhage or lipid, and presence

of calci

ed or

brous plaque is considered false positive for lipid core.

1

Automated methods for

nding the vessels of interest use e.g. thresholding and seeding

to de

ne the extent of the vessels and combining this with curved MPR normal vessel size,

maximun stenosis and stenosis percent can be acquired.

5

Stenosis can also be found using a

region growing algorithm and selecting the point of lumen that has the most of narrowing.

4

Quantitative parameters

An important quantitative parameter obtained from CTA is the degree of stenosis. According to North American Symptomatic Carotid Endarterectomy Trial (NASCET) criterion

it is calculated with pattern

degreeofstenosis(%) =

D d

D

# 100

where D is the lumen diameter of distal internal carotid artery (ICA) and d is the

minimal residual lumen.

14

Houns

eld unit attenuation is used to identify the type of the plaque. Houns

eld values

(HV) for each plaque tissue vary slightly according to source. The values are around -20 -60 HU for lipidic tissue, 60 - 120 HU for

brous tissue and >120 HU for calci

ed tissue.

1;2;9

The size of di#erent plaque components can also be determined.

1

Message of the parameters, normal values?

The severity of stenosis is determined by NASCET as follows: normal (0%), mild stenosis

(1% - 29%), low moderate stenosis (30% - 49%), high moderate stenosis (50% - 69%), high

stenosis (70% - 99%) and occluded (100%). The degree of stenosis considered with other

risk factors e.g. contralateral carotid occlusion, a history of diabetes and high diastolic

3

blood pressure helps physicians determine the risk of stroke or death so the most bene

cial

treatment can be decided.

15

Requirements of imaging conditions

The answer to this question is presented in the Imaging protocol section.

Duration and cost of the process

The MDCT scan of carotid artery takes less than half a minute plus little wait time

before imaging can start due to spreading of the contrast agent. Lumen segmentation

takes approximately 1-15 minutes depending on the algorithm used.

7

Compared to MRA

that requires approximately 15 minutes of time to produce image of similar area, MDCT

is signi

cantly faster. Compared to traditional angiogram CTA method is more cost

e#ective

16

and compared to MRA it is approximately 10 times cheaper

4

.

Interview of a professional

As a part of this essay we interviewed Florentino Santos who works as a researcher for

Tampere University of Technology. He has special knowledge about analysis of carotid

plaque with CTA and he is also involved in developing better models of carotid plaques

with the help of micro-CT scans of surgically removed carotid plaques.

Mr. Florentino gave us presentation of his operator independent automated software

that analyses carotid artery. His software utilized region growing algorithm for vascular

segmentation and was developed with Matlab. He also helped us understand the basics

of CTA imaging and answered to the questions that we had prepared for him. We have

used the answers that he provided as one the references while writing this essay.

Diagnostic criteria for patient selection

Transient ischemic attack (TIA) is often the

rst sign of weakened blood

Symptoms of TIA are e.g. numbness in arms, sluggish speak and uncontrolled blinking.

The attack is de

ned as transient if the symptoms disappear. TIAs tell that a screening

for carotid plaques is needed.

4

The

rst screening method for evaluating carotid disease is carotid duplex ultrasound.

If the results are inconclusive or indicate an intermediate or severe stenosis CTA or magnetic resonance angiography (MRA) may be used to further study the carotis. Since CTA

provides excellent spatial resolution and high anatomic detail of both carotid artery and

the surrounding tissue it is preferred by surgeons.

5

Other available methods

Other available methods for detection of carotid plaque are ultrasound, MRA and conventional angiography.

As previously mentioned, ultrasound is the

rst screening method in analysis of carotid

plaque. It is cheapest of the methods mentioned above but might not give as good results

especially in characterization of the plaque. Calci

ed plaque produces an acoustic shadowing in ultrasound imaging which makes the assessment of stenosis degree di#cult.

2

Other

4

drawbacks of carotid ultrasound are its dependency of operator, variance of

ndings and

di#culty to reproduce the

ndings.

10

MRA is as good a method as CTA for analyzing carotid plaques but signi

cantly more

expensive. MR images can have better tissue contrast resolution than CTA images but

until the usage and maintaining of MR machines becomes cheaper CTA is the preferred

method.

4

Conventional angiography uses X-rays for imaging but uses a di#erent method for

injecting the contrast agent. The contrast agent is injected to the artery under study by a

catheter, a small tube-like device

11

. This makes the method invasive and increases risk of

complications

5

. In CTA the agent is injected to a vein in arm with a needle which is a safe

and noninvasive technique. Thus conventional angiography is reserved for cases where the

combination of ultrasound and CTA or MRA give inconclusive results.

5

Possible use of analysis in other imaging modalities

MRA produces three-dimensional images consisting of slices similar to those acquired

with CTA. Thus it is possible to CTA image processing techniques also for the analysis of

MRA images. The implementation for MRA image analysis can be done by adaptation

and teaching of the software.

4