Добавил:
Sekretar
kiopkiopkiop18@yandex.ru
t.me/Prokururor I Вовсе не секретарь, но почту проверяю
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5829_Библиотеки_им_академика_М_И_Перельмана.pdf
196 F. Ciompi et al.
Tabl e 2 Performance as
MEAN(STD)ofIvtDand
pSFFS methods when
classifying in-vitro data. The
overall accuracy (A), the
sensitivity (S), specificity (K)
and precision (P)are
computed when
discriminating fibrotic (fib),
lipidic (lip)andcalcified (cal)
plaque
100
98
96
94
performance %
92
IvtD pSFFS
A 89.93 (0.29) 91.59 (0.47)
S
fib
S
lip
S
cal
K
fib
K
lip
K
cal
P
fib
P
lip
P
cal
90.46 (0.60) 92.58 (0.95)
80.49 (0.80) 85.21 (0.59)
92.20 (0.46) 92.29 (0.75)
92.16 (0.41) 93.85 (0.34)
95.50 (0.17) 96.47 (0.37)
96.15 (0.48) 96.21 (0.38)
93.34 (0.33) 94.72 (0.30)
69.48 (0.86) 74.78 (2.09)
92.48 (0.86) 93.23 (0.62)
J
Accuracy
Sfibrotic
Slipidic
Scalcified
90
88
1 2 3 4 5 6 7 8 9 10 11 12
Iteration [k]
Fig. 11 Example of thepSFFS algorithm performance up to 12th iteration. This case is specific for
a generic pth necro case. J is the value achieved at each iteration during the data selection process,
while the overall accuracy and sensitivity (S) for each tissue type are computed by characterizing
p
data at each iteration
the X
val
4.5 Test In Vitro
Tabl e 2 shows the classification parameters for the pSFFS methods when characterizing in vitro data, compared with the initial plaque characterization performance
using exclusively in vitro data (IvtD) as training set. In pSFFS, the classifier has been
trained by using the enhanced data set obtained by applying the data fusion method.
Note that the algorithm provides a mean improvement in performance parameters
Coronary Atherosclerotic Plaque Characterization By Intravascular Ultrasound 197
Fig. 12 Plaque characterization on IVUS data. First column: original IVUS images; second
column: histologically validated ground-truth plaque segmentation; third column: classification
result in the regions defined by histology. Yellow, green and blue (corresponding to light, medium
and dark grey) colors indicate lipidic, fibrotic and calcified tissue, respectively
of 1.9% with respect to the IvtD case. Figure 11 shows an example of the changes of
performance parameters during thepSFFS algorithm for a generic patient. Figure12
shows some examples of plaque characterization in IVUS images where tissues are
labeled using the enhanced classifier.
4.6 Test In Vivo
The results on Table 2 prove that the enhanced classifier outperforms a classifier
trained and validated only on in vitro cases. Note that the last approach has been
198 F. Ciompi et al.
Tabl e 3 Performance as
MEAN (STD) of Ivt vs
pSFFS methods in classifying
in-vivo data. The overall
accuracy (A),thesensitivity
(S), specificity (K) and
precision (P) are computed
when discriminating fibrotic
(fib), lipidic (lip) and calcified
(cal) plaque
A 74.05 (1.34) 87.09 (0.33)
S
fib
S
lip
S
cal
K
fib
K
lip
K
cal
P
fib
P
lip
P
cal
IvtD pSFFS
88.49 (3.36) 90.66 (0.43)
27.87 (3.50) 58.46 (1.61)
99.62 (0.11) 99.37 (0.22)
67.54 (1.53) 85.86 (0.55)
97.53 (1.49) 98.56 (0.09)
96.67 (0.26) 96.32 (0.24)
55.90 (1.27) 73.85 (0.72)
83.93 (7.95) 92.59 (0.33)
94.74 (0.38) 95.82 (0.26)
followed by almost all the plaque characterization methods presented in the last ten
years, on the hypothesis that the differences between in vitro and in vivo data are
negligible.
In the data fusion approach, the training data set is designed by fusing in vivo
and in vitro IVUS data. It is then reasonable to assume that the enhanced classifier
could exhibit better performance in discriminating in vivo data with respect to a
classifier trained only with in vitro data. Note that this assumption does not imply
any hypothesis on the similarity between in vivo and in vitro data; it is in fact
naturally deduced since the enhanced data set includes examples of in vivo data,
thus providing to the classifier a knowledge about the spatial distribution of in vivo
points. In order to prove the last hypothesis, the classification results of both the
IvtD and the pSFFS classifiers when discriminating plaques in in vivo cases are
compared: Table 3 shows the obtained classification performance in both cases.
Given the impossibility of obtaining the absolute knowledge on in vivo tissue
composition by histological analysis, it is clear that the obtained results cannot
be considered as a reliable, though qualitative results. Notwithstanding that, the
increment of about 13% in overall accuracy when the pSFFS classifier is used
indicates the benefits achieved in using an enhanced data set instead of a pure in
vitro training set.
References
1. Shah PK (2003) Mechanism of plaque vulnerability and rupture. JACC 41(1):15–22
2. Burke AP, Farb A, Malcom GT, Liang Y, Smialek J, Virmani R (1997) Coronary risk factors
and plaque morphology in men with coronary disease who died suddenly. New Engl J Med
336:1276–1281
3. Fuster V, Moreno PR, Fayad ZA, Corti R, Badimon JJ (2005) Atherothrombosis and high-risk
plaque. JACC 46(6):937–54
4. Ehara S, Kobayashi Y, Yoshiyama M, Shimada K, Shimada Y, Fukuda D, Nakamura Y,
Yamashita H, Yamagishi H, Takeuchi K, Naruko T, Haze K, Becker AE, Yoshikawa J, Ueda
M (2004) Spotty calcification typifies the culprit plaque in patients with acute myocardial
infarction: An intravascular ultrasound study. Circulation 110:3424–3429
Coronary Atherosclerotic Plaque Characterization By Intravascular Ultrasound 199
5. Willerson JT, Wellens HJJ, Cohn JN, Holmes DR (2007) Atherosclerotic vulnerable plaques:
Pathophysiology, detection, and treatment. Cardiovasc Med 10:621–639
6. Davies MJ, Thomas AC (1985) Plaque fissuring – the cause of acute myocardial infarction,
sudden ischaemic death, and crescendo angina. Br Heart J 53(4):363–73
7. Schaberle W (2004) Ultrasonography in vascular diagnosis. Springer, Berlin
8. Gorge G, Ge J, Baumgart D, von Birgelen C, Erbel R (1998) In vivo tomographic assessment
of the heart and blood vessels with intravascular ultrasound. Basic Res Cardiol 93(4):219–240
9. Nissen SE, Yock P (2001) Intravascular ultrasound: Novel pathophysiological insights and
current clinical applications. Circulation 103:604–616
10. Ellis SG, Holmes DR Jr (eds) (2006) Strategic approaches in coronary intervention, 3rd edn.
Lippincott Williams & Wilkins, PA
11. Siegel RJ (ed) (1998) Intravascular ultrasound imaging in coronary artery disease Informa
Healthcare; 1st edition (January 15, 1998)
12. Zhang X, McKay CR, Sonka M (1998) Tissue characterization in intravascular ultrasound
images. TMI 17(6):889–899
13. Caballero KL, Barajas J, Pujol O, Salvatella N, Radeva P (2006) In-vivo ivus tissue classification: A comparison between rf signal analysis and reconstructed image. Progr Pattern Recognit
Image Anal Appl 4225/2006:137–146
14. Caballero KL, Barajas J, Pujol O, Rodriguez O, Radeva P (2007) Using reconstructed ivus
images for coronary plaque classification. EMBS, pp. 2167–2170
15. Moore M et al (1998) Characterisation of coronary atherosclerotic morphology by spectral
analysis of radiofrequency signal: In vitro intravascular ultrasound study with histological and
radiological validation. Heart 79(5):459–467
16. Nair A, Kuban BD, Tuzcu EM, Schoenhagen P, Nissen SE, Vince DG (2002) Coronary
plaque classification with intravascular ultrasound radiofrequency data analysis. Circulation
106:2200–2206
17. Nair A, Kuban BD, Obuchowski N, Vince GD (2001) Assessing spectral algorithms to predict
atherosclerotic plaque composition with normalized and raw intravascular ultrasound data.
UMB 27(10):1319–1331
18. Katouzian A, Sathyanarayana S, Baseri B, Konofagou EE, Carlier SG (2008) Challenges
in atherosclerotic plaque characterization with intravascular ultrasound (ivus): From data
collection to classification. TITB 12(3):315–327
19. Bedekar D (2003) Atherosclerotic plaque characterization by acoustic impedance analysis of
intravascular ultrasound data. IEEE Ultrason Symp, pp. 1524–1527
20. Sathyaranayana S, Carlier S, Wenguang L, Thomas L (2009) Characterization of atherosclerotic plaque by spectral similarity of radiofrequency intravascular ultrasound signals. EuroIntervention 5:133–139
21. Kawasaki M, Takatsu H, Noda T, Ito Y, Kunishima A, Arai M, Nishigaki K, Takemura G,
Morita N, Minatoguchi S, Fujiwara H (2001) Noninvasive quantitative tissue characterization
and two-dimensional color-coded map of human atherosclerotic lesions using ultrasound
integrated backscatter. JACC 38(2):486–492
22. Kawasaki M, Takatsu H, Noda T, Sano K, Ito Y, Hayakawa K, Tsuchiya K, Arai M,
Nishigaki K, Takemura G, Minatoguchi S, Fujiwara T, Fujiwara H (2002) Invivo quantitative tissue characterization of human coronary arterial plaques by use of integrated
backscatter intravascular ultrasound and comparison with angioscopic findings. Circulation
105:2487–2492
23. Kawasaki M, Sano K, Okubo M, Yokoyama H, Ito Y, Murata I, Tsuchiya K, Minatoguchi
S, Zhou X, Fujita H, Fujiwara H (2005) Volumetric quantitative analysis of tissue characteristics of coronary plaques after statin therapy using three-dimensional integrated backscatter
intravascular ultrasounds. JACC 45(12):1946–1953
24. Murashige A, Hiro T, Fujii T, Imoto K, Murata T, Fukumoto Y, Matsuzaki M (2005)
Detection of lipid-laden atherosclerotic plaque by wavelet analysis of radiofrequency intravascular ultrasound signals: In vitro validation and preliminary in vivo application. JACC
45(12):1954–1960
200 F. Ciompi et al.
25. Gonzalez RC, Woods RE (2001) Digital image processing, 2nd edn. Prentice Hall, NJ
26. Mehta SK, McCrary JR, Frutkin AD, Dolla WJS, Marso SP (2007) Intravascular ultrasound
radiofrequency analysis of coronary atherosclerosis: An emerging technology fot the assessment of vulnerable plaque. Eur Heart J 28:1283–1288
27. Tobocman W, Santosh K, Carter JR, Haacke EM (1994) Tissue characterization of arteries with
4MHZ ultrasound. Ultrasonics 33:331–339
28. Milunski MR, Mohr GA, Perez JE, Vered Z, Wear KA, Gessler CJ, Sobel BE, Miller
JG, Wickline SA (1989) Ultrasonic tissue characterization with integrated backscatter.
acute myocardial ischemia, reperfusion, and stunned myocardium in patients. Circulation
80:491–503
29. Ciompi F, Pujol O, Gatta C, Rodriguez-Leor O, Mauri-Ferre J, Radeva P (2010) Fusing in-vitro
and in-vivo intravascular ultrasound data for plaque characterization. IJCI 26:763–779
30. Carlier SG, Tanaka K, Katouzian A (2007) Atherosclerotic plaque characterization from radio
frequency ultrasound signal processing. US Cardiovascular Disease, pp 54–56
31. Katouzian A, Baseri B, Konofagou EE, Laine AF (2008) Automatic detection of blood versus
non-blood regions on intravascular ultrasound (ivus) images using wavelet packet signatures.
Proc SPIE 6920, pp 1–8
32. Addison PS (2002) The illustrated wavelet transform handbook. Publishing, Institute of
Physics
33. Escalera S, Pujol O, Mauri J, Radeva P (2008) Ivus tissue characterization with sub-class errorcorrecting output codes. J Signal Process Syst 55:35–47
34. Ciompi F (2008) Ecoc-based plaque classification using in-vivo and ex-vivo intravascular
ultrasound data. Master thesis
35. Dietterich TG, Bakiri G (1995) Solving multiclass learning problems via error-correcting
output codes. JAIR 2:263–286
36. Rifkin R, Klautau A (2004) In defense of one-vs-all classification. JMLR 5:101–141
37. Pudil P, Ferri FJ, Novovicova J, Kittler J (1994) Floating search methods for feature selection
with nonmonotonic criterion functions. ICPR 2:279–283
38. Allwein EL, Schapire RE, Singer Y (2000) Reducing multiclassto binary: A unifying approach
for margin classifiers. Proceedings of the Seventeenth International Conference on Machine
Learning, pp 9–16
39. Schapire R (2002) The boosting approach to machine learning: an overview. MSRI Workshop
on Nonlinear Estimation and Classification, pp 1–23
Francesco Ciompi F. Ciompi received the MS degree
in Electronic Engineering from the Universit`adiPisain
2006 and the MS in Computer Vision and Artificial Intelligence from the Universitat Aut´onoma de Barcelona
in 2008. He is currently a PhD student at the dept. of
Applied Mathematics and Analysis at the Universitat
de Barcelona. His research interests include the use of
machine learning techniques for the classification and
segmentation of Intravascular Ultrasound data.
Coronary Atherosclerotic Plaque Characterization By Intravascular Ultrasound 201
Oriol Pujol Oriol Pujol is associate professor at dept.
of Applied Mathematics and Analysis of University of
Barcelona since 2008, and senior researcher at Computer
Vision Center. He has been since 2004 an active member
in the organization of several activities related to image
analysis, computer vision, machine learning and artificial
intelligence. In applied research he is working in medical
image analysis, object recognition, wearable computing
and augmented reality.
Josepa Mauri Ferr´e Josepa Mauri received the title of
MD in 1982 at Universitat Aut`onoma de Barcelona. In
1992 she received the Laurea summa Cum Laude in
Medicine at the Universitat de Barcelona. Since 2000,
she is the Director of Cardiac Catherization Laboratory
in the Hospital Universitari “Germans Trias I Pujol de
Badalona”. From 2002 to 2005 she was the President
of the Diagnostic Intracoronary Technics/IVUS Working
Group of the Spanish Society of Cardiology. From 2006
to 2009 she was the President of the Spanish Working
Group in Cardiac Interventions of the Spanish Society of Cardiology. Her clinical
research areas have been in coronary angioplasty and Dilated Cardiomyopathy,
dilated Cardiomiopathy, coronary angioplasty, stents, endothelial disfunction and
IVUS studies. She is currently involved in international educational projects in
interventional cardiology.
Petia Radeva Petia Radeva (PhD 1998, Universitat
Aut`onoma de Barcelona, Spain) is a senior researcher
at UB. She has more than 150 publications in interna-
tional journals and proceedings. Her present research
interests are development of learning-based approaches
(in particular statistical methods) for computer vision
and medical imaging. She has led one EU project and
several Spanish projects. She has 12 patents in the field
of medical imaging. Currently, she is heading projects in
the field of cardiac imaging and wireless endoscopy in collaboration with Spanish
hospitals and international medical imaging companies.
Three-Dimensional Ultrasound Plaque
Characterization
Jos´e Seabra, Jasjit S. Suri, and Jo˜ao Miguel Sanches
Abstract The chapter proposes a framework for extending the analysis of the
atherosclerotic disease to a three-dimensional perspective. Different data acquisition
systems, either based on a robotic arm setup or free-hand are proposed, in order
to collect image sequences that completely describe the plaque anatomy. A 3D
reconstruction method is proposed, comprising a Rayleigh based de-speckling
approach and interpolation. As a consequence, 3D maps accounting for plaque
echogenicity and texture, according to appropriate local Rayleigh estimators are
obtained. Furthermore, the applicationof a segmentation approach which makes use
of the Graph-cuts method, provides an efficient way to segment and locally identify
unstable regions throughout the plaque. This information, complemented with a
more accurate inspection of plaque morphology, may have an important clinical
impact in disease diagnosis.
1 Introduction
Atherosclerosis is a disease which generally affects large and medium-sized arteries and its most important feature is plaque formation due to progressive
sub-endothelial accumulation of lipid, protein, and cholesterol esters in the blood
vessel wall.
Several studies recognize that the degree of stenosis, obstruction to the blood
flow, is an important physiological landmark of stroke but that other parame-
J. Seabra () • J.M. Sanches
Institute for Systems and Robotics, Department of Bioengineering from the Instituto Superior
T´ecnico/Technical University of Lisbon, Portugal
e-mail: mail2jseabra@gmail.com; jmrs@ist.utl.pt
J.S. Suri
Biomedical Technologies, Inc., Denver, CO, USA
Idaho State University (Affiliated), Pocatello, ID, USA
e-mail: jsuri@comcast.net
J.M. Sanches et al. (eds.), Ultrasound Imaging: Advances and Applications,
DOI 10.1007/978-1-4614-1180-2
9, © Springer Science+Business Media, LLC 2012
203
204 J. Seabra et al.
ters, such as plaque echo-morphology and texture should also be considered for
designing a plaque risk profile. Additionally, it has been observed that vulnerable
plaques are usually associated with fibrous cap thinning and infiltration of inflammatory cells consequently leading to rupture. Other studies reported a positive
correlation betweenthe presence offatty contents and hemorrhage with neurological
symptoms, thus suggesting that inflammatory activity potentially determines plaque
instability. In Pedro et al. [1, 2], the location and extension of these regions are
identified as sensitive and relevant markers of stroke risk.
Numerous research groups conducted studies aiming at characterizing and
identifying the main features of the symptomatic lesion [1,3–8]. Among these, the
gray-scale median and P
can be used to characterize the plaque from an echogenic
40
viewpoint. However, the interpretation of these parameters values will fail to reveal
possible unstable foci within the plaque, specially when plaques are heterogeneous
or present significant hypoechogenic areas.
The risk assessment of plaque rupture through conventional 2D techniques is
limited to a subjective selection of a representative image of plaque structure and
it is hardly reproducible. An accurate diagnostic procedure based on 3D is known
to be valuable but has not yet been adopted in clinical practice, mainly because
such technology is not usually available in most medical facilities. Recently, less
operator-dependent methods based on 3D US have been proposed from better
assessment of plaque vulnerability [9, 10]. These studies aim at quantifying the
plaque volume, degree of stenosis [11], and the extension of surface ulceration [12].
The focus of this chapter is to assess the atherosclerotic disease on a 3D
perspective providing better visualization of the lesion and characterization of
potential risk. The carotid disease study in 3D is rooted on the reconstruction of 3D
maps starting with 2D information extracted from noisy BUS images. To achieve
the proposed objectives, different 3D image acquisition methods are explored
and powerful methods for identifying vulnerable foci within the plaque volume
are used.
2 Development of DAQ Systems for 3D Ultrasound
Medical US has benefited from major advances in technology and is considered an
indispensable imaging modality due to its flexibility and non-invasive character.
Currently, there are accurate methods to assess the disease severity based on
CT [13]orMRI[14]. However their application is expensive, time consuming,
and requires equipment which is not yet available and accessible in most clinical
facilities. On the other hand, 2D ultrasound is widely available and provides real
time data acquisition and visualization, so it has been so far the preferred technique
in the diagnosis and monitoring of the disease.
Although all anatomy is 3D in form, the vast majority of US imaging is 2D.
Most of the times, this technique provides sufficient information for diagnosis but
there are clearly identifiable limitations, such as, non-ability to perform quantitative
Three-Dimensional Ultrasound Plaque Characterization 205
volume measurements or to obtain optimal 2D scan views of the anatomical
ROI (Region of Interest). Consequently, 3D US is a logical solution to allow better,
more completeand objective diagnostic results. In this imagingmodality, the 2D US
images are combined by a computer to form an objective 3D image of the anatomy
and pathology. This data can be manipulated and measured in 3D both in real time
or later off-line. Moreover, unlike CT and MR imaging, in which 2D images are
usually acquired at a slow rate as a stack of parallel slices, in a fixed orientation, US
provides images at a high rate (15–60 s
−1
) and in arbitrary orientations.
Most 3D US systems make use of a conventional transducer to obtain a
sequence of images by sweeping the probe along the anatomical ROI, and differ
only in acquisition and position sensing [9]. In this way, images can be acquired
mechanically, free-handed with or without an optical or electromagnetic spatial
locator and using 2D arrays. Some of these systems were validated in various
clinical applications, such as obstetrics, cardiology, and vascular imaging in order
to increase the diagnosis confidence [15].
2.1 Robotic Arm Prototype
Depending on the organ or tissue to be scanned, it is necessary to apply different
scanning strategies or protocols which comprise linear, rotatory, and free-hand
scanning, just to name the most common ones. Most imaging systems are not
optimally adapted for such a wide range of applications. Hence, a 3D US prototype
robotic system which can control, standardize, and accurately perform the acquisition process is presented. This system may assist the operator in defining suitable
scanning paths for each patient according to the ROI to be scanned. Different
acquisition properties can be assigned in each examination, such as the duration
and rate of image acquisition. Moreover, each image is assigned with its spatial
information allowing to further perform follow-up studies or to reconstruct and
segment the tissues or organs scanned with higher degree of confidence.
Robotic systems can be regarded as an important diagnosis tool because they
can simultaneously control and standardize the image acquisition process. Thus,
they can be very suitable for quantifying and accurately monitor the development
of cardiovascular diseases, namely, the progression of atheromatous plaques by
scanning the carotid or coronary arteries [16]. In addition, the ability to remotely
position the US probe with the robotic arm could also be used in telemedicine [17].
Given this, a prototype medical robot is described which can easily be integrated
with common ultrasound scanning equipment and provides clinicians with their
regular scanning operations. The prototype robotic arm is schematically shown
in Fig. 1a and comprises four main components. The first unit is the robotic arm
(Scorbot-ER VII, Intelitek, USA) with six degrees of freedom which is operated
from the robot controller. The second element is an US portable scanner (Echo
Blaster 128, Telemed, LT), equipped with alinear array probe. This probe isattached
to the tip of the robotic arm, together with an electromagnetic position sensing
206 J. Seabra et al.
Fig. 1 Robotic Arm Prototype. (a) Block diagram of the system components. (b) Experimental
setup. The robotic arm carries the probe and the position sensing receptor from the US scanner and
the spatial locator, respectively. The workstation personal computer controls the movements of the
robotic arm with the aid of a joystick. Images are tagged with their spatial location and showed on
the screen. (c) Acquisition modes: linear (a), fan-like (b), and rotatory (c) scans
device (Fastrak, Polhemus, VT). The last component of the system is the computer
workstation, which holds a joystick and the interface to control the medical robot
and the US scanner.
A user-friendly graphical interface provides access and setting of the robotic arm
controls and movements as well as visualization and tuning of acquisition results.
The robotic arm is manipulated using a joystick which allows to move the US probe
towardthe ROI to be scanned. Moreover,the system features a learningmode which
enables to store in memory a suitable scan path, and a replay mode to reproduce the
manually taught path. This attribute of the robotic system is suitable to guarantee
reproducible and personalized results since a scan path can be assigned for each
patient with controlled speed and accurate position information provided by the
spatial locator.