rebuilding ivus images from raw data of the rf signal: a...

6
Rebuilding IVUS Images From Raw Data Of The RF Signal Exported by IVUS Equipment Marco Aurélio Granero¹ , ³, Marco Antônio Gutierrez², Eduardo Tavares Costa¹ ¹ Department of Biomedical EngineeringDEB/FEEC/UNICAMP, Campinas, Brazil ² Division of Informatics/Heart Institute HCFMUSP, São Paulo, Brazil ³ Federal Institute of Education, Science and Technology S. Paulo IFSP, São Paulo, Bra Abstract - The study of composition and classification of atherosclerotic plaque has been a very active research field, both in cardiology and image processing. Intravascular ultrasound (IVUS) is an effective tool, which can insights about the cross- section of blood vessels, with sufficient accuracy to allow an accurate assessment of CT slices. This enables information about blood vessel structures to be determined. During an IVUS medical examination, physicians subjectively adjust a set of parameters to improve the visualization of a Region Of Interest (ROI) and produce corresponding images in Digital Imaging and Communications (DICOM) format, for later analysis and study. DICOM is appropriate for storage, transportation and access, but limits subsequent changes to image parameters, such as contrast or brightness. This makes comparison across patient populations difficult and restricts image processing operations. This paper details an alternative to using DICOM, which is to rebuild IVUS images from raw radiofrequency signal (RF) data. The main advantage of this process is the independence of the acquisition parameters adjusted during the exam. This advantage makes possible the comparison between exams and can be used to monitor the evolution of cardiovascular disease. Beyond this, once the reconstructed images and the RF signal are stored, operations relating to texture and spectral analysis can be carried out and automatic classifiers employed. From a clinical point of view these reconstructed images share the same characteristics as DICOM images with an advantage that the former have a higher contrast than the latter, allowing deeper regions to be seen. Keywords: RF signal, IVUS image, rebuilding process, ultrasound image, RF raw data. 1. Introduction Among the different modalities of medical images, ultrasound is arguably the most difficult in which to perform segmentation. This is evident from a study of the first papers on segmentation, in which it was only possible to apply a threshold to the image in order to separate the background from foreground due to the poor quality of the acquired data (Noble, 2010). At the same time, subsequent technological development has greatly increased the quality of ultrasound images, especially in terms of signal to noise ratio (SNR) and contrast to noise ratio (CNR), resulting in improvements to image quality. Several studies have been highlighted that aim to develop algorithms for the design of edges on objects contained in ultrasound images (Noble, 2010). Ultrasonic Tissue Characterization (UTC) has become a well-established research field since its first publication (Mountford and Wells, 1972). Thijssen (2003) defines UTC as the assessment by ultrasound of quantitative information about the characteristics of biological tissue, and their pathology. This quantitative information is extracted from echographic data from RF data. UTC applications abound in the literature and include classification of breast tissue (Tsui et al. 2008 and Molthen et. al. 1998), liver (Molthen et al., 1998), heart (Clifford et al., 1998 and Nillesen et. al., 2008), eyes (Lizzi et. al., 1983), skin (Raju et. al., 2003), kidney (Engelhorn et. al., 2012) and prostate (Moradi, 2008). Szabo (2004) defines two general goals for ultrasonic tissue characterization which can be applied to the above areas (Szabo, 2004): i. Reveal the properties of tissues by analyzing the RF signal backscattered by ultrasound transducer and ii. Use information about the properties of the tissue to distinguish between the state of tissue (healthy or diseased), or to detect changes in these properties when subjected to stimuli or long periods of time in response to natural processes or medication. Reaching these goals can be challenging since the interaction between biological tissue and sound waves is extremely difficult to model and the process evolved in image segmentation is strongly influenced by the quality of data and by the different parameters used during the acquisition process of an image. Parameters like contrast, brightness and gain are adjusted by physicians to improve the visualization of regions during the examination. These changes determine the DICOM images that are recorded and the result cannot be changed after the image has been acquired. This greatly complicates the comparison between patients and the use of images in studies of groups of patients.

Upload: doantuyen

Post on 07-Mar-2018

217 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

Rebuilding IVUS Images From Raw Data Of The RF

Signal Exported by IVUS Equipment

Marco Aurélio Granero¹,³, Marco Antônio Gutierrez², Eduardo Tavares Costa¹

¹ Department of Biomedical Engineering– DEB/FEEC/UNICAMP, Campinas, Brazil

² Division of Informatics/Heart Institute – HCFMUSP, São Paulo, Brazil

³ Federal Institute of Education, Science and Technology S. Paulo – IFSP, São Paulo, Bra

Abstract - The study of composition and classification

of atherosclerotic plaque has been a very active

research field, both in cardiology and image

processing. Intravascular ultrasound (IVUS) is an

effective tool, which can insights about the cross-

section of blood vessels, with sufficient accuracy to

allow an accurate assessment of CT slices. This

enables information about blood vessel structures to be

determined. During an IVUS medical examination,

physicians subjectively adjust a set of parameters to

improve the visualization of a Region Of Interest (ROI)

and produce corresponding images in Digital Imaging

and Communications (DICOM) format, for later

analysis and study. DICOM is appropriate for

storage, transportation and access, but limits

subsequent changes to image parameters, such as

contrast or brightness. This makes comparison across

patient populations difficult and restricts image

processing operations. This paper details an

alternative to using DICOM, which is to rebuild IVUS

images from raw radiofrequency signal (RF) data. The

main advantage of this process is the independence of

the acquisition parameters adjusted during the exam.

This advantage makes possible the comparison

between exams and can be used to monitor the

evolution of cardiovascular disease. Beyond this, once

the reconstructed images and the RF signal are stored,

operations relating to texture and spectral analysis can

be carried out and automatic classifiers employed.

From a clinical point of view these reconstructed

images share the same characteristics as DICOM

images with an advantage that the former have a

higher contrast than the latter, allowing deeper

regions to be seen.

Keywords: RF signal, IVUS image, rebuilding

process, ultrasound image, RF raw data.

1. Introduction

Among the different modalities of medical images,

ultrasound is arguably the most difficult in which to

perform segmentation. This is evident from a study of

the first papers on segmentation, in which it was only

possible to apply a threshold to the image in order to

separate the background from foreground due to the

poor quality of the acquired data (Noble, 2010).

At the same time, subsequent technological

development has greatly increased the quality of

ultrasound images, especially in terms of signal to

noise ratio (SNR) and contrast to noise ratio (CNR),

resulting in improvements to image quality. Several

studies have been highlighted that aim to develop

algorithms for the design of edges on objects contained

in ultrasound images (Noble, 2010).

Ultrasonic Tissue Characterization (UTC) has

become a well-established research field since its first

publication (Mountford and Wells, 1972). Thijssen

(2003) defines UTC as the assessment by ultrasound of

quantitative information about the characteristics of

biological tissue, and their pathology. This quantitative

information is extracted from echographic data from

RF data.

UTC applications abound in the literature and

include classification of breast tissue (Tsui et al. 2008

and Molthen et. al. 1998), liver (Molthen et al., 1998),

heart (Clifford et al., 1998 and Nillesen et. al., 2008),

eyes (Lizzi et. al., 1983), skin (Raju et. al., 2003),

kidney (Engelhorn et. al., 2012) and prostate (Moradi,

2008).

Szabo (2004) defines two general goals for

ultrasonic tissue characterization which can be applied

to the above areas (Szabo, 2004):

i. Reveal the properties of tissues by analyzing

the RF signal backscattered by ultrasound

transducer and

ii. Use information about the properties of the

tissue to distinguish between the state of

tissue (healthy or diseased), or to detect

changes in these properties when subjected to

stimuli or long periods of time in response to

natural processes or medication.

Reaching these goals can be challenging since the

interaction between biological tissue and sound waves

is extremely difficult to model and the process evolved

in image segmentation is strongly influenced by the

quality of data and by the different parameters used

during the acquisition process of an image.

Parameters like contrast, brightness and gain are

adjusted by physicians to improve the visualization of

regions during the examination. These changes

determine the DICOM images that are recorded and

the result cannot be changed after the image has been

acquired. This greatly complicates the comparison

between patients and the use of images in studies of

groups of patients.

Page 2: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

Thus, to avoid these complications and make

image reconstructed independent of the parameters set

by the physician a reconstruction method from IVUS

images is proposed. This method is based on the RF

signal stored by the equipment during medical imaging

examinations of intravascular ultrasound.

The process of rebuilding starts with applying a

band-pass filter to the RF signal to eliminate signals

that do not come from the transducer. In the next step,

a time gain compensation (TGC) function is applied to

compensate for attenuation loss. After this, the

envelope of the signal is computed and the result is

log-compressed and normalized in a grayscale image.

After the process of rebuilding, the grayscale

image, in polar coordinators, is submitted to a Digital

Development Process (DDP) responsible for enhancing

the contrast and edge emphasis. So, the image is

interpolated to cartesian coordinators. The cartesian

image is further processed with an intensity

transformation function to improve the contrast of the

final cartesian grayscale IVUS image.

The above processes are described in more detail

in section 2 and the results obtained are shown in

section 3. In section 4 a comparison is made between

reconstructed images and DICOM images from an

examination. Finally, section 5 shows conclusions and

possibilities for future work.

2. Method for IVUS image

reconstruction

An IVUS examination is carried out by inserting a

catheter into coronary arteries via femoral or brachial

vessels. At the tip of this catheter there is an ultrasound

emitter and a piezoelectric transducer that collect the

echoes reflected by internal structures of the vessel as

RF signal.

A schematic representation of the execution of an

IVUS examination is shown in Figure 1(a), where the

IVUS equipment collects data from patient and stores

it in the workstation. Figure 1(b) shows an IVUS

rotational catheter.

During an IVUS exam, the acquired images are

stored in DICOM format and exported to the databank

of the clinical centre to be used for clinical diagnosis.

In addition to the images in DICOM format, the

equipment allows the RF signal to be recorded, which

are used in the manufacture of images in a proprietary

format.

The proposal of this paper is to process the RF

signal data according to the steps shown in Figure 2.

These steps are detailed below.

Figure 1: (a) IVUS in-vivo analysis typical scenario, (b) Rotational IVUS catheter. Extracted from Ciompi, (2008).

Figure 2: Block diagram of reconstruction

process.

2.1 – RF dataset

The data was taken from examinations in the

Department of Hemodynamics in the Heart Institute of

the Medical School of the University of São Paulo

(Heart Institute – HCFMUSP), Brazil, using iLab

IVUS (Boston Scientific, Fremont, USA), equipped

with a 40 MHz catheter Atlantis SR 40 Pro and

anonymised to avoid the identification of the patient

and used only for research purposes.

The RF File Reader (designed by Boston

Equipment) is an xml file that contains information

about the examination. This file allows us to identify

the number of rows, columns and frames from each

exam. Beyond this, the reader contains the distance

from each pixel in the image, in millimetres.

Once image attributes have been found using the

RF File Reader, it is possible to extract the data. These

data were placed in a tri-dimensional matrix. The rows

of this matrix represent the lines in A-mode, each line

with radial information about the vessel, the columns

represent the distance to the tip of catheter and the

slices, third dimension, represent each time frames of

the exam. The study of IVUS used in this work results

RF raw data

Bandpass filter

TGC

Log-compression

DDP

Polar Image

Envelope

Page 3: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

in a 3D matrix, where the dimensions represents the

size of each image and the third dimension being the

number of frames.

After this, each frame was submitted to the

reconstruction shown in Figure 2.

2.2 - Bandpass filter

A Butterworth bandpass filter was applied to

dataset in order to eliminate frequencies that do not

come from the transducer. The manufacture of

transducer describes the central frequency emitted by

transducer at the tip of catheter as 40 MHz and

frequency sample rate as 200 MHz.

Each line in A-mode was filtered by a Butterworth

finite impulse response filter (FIR filter).

The frequency range was adjusted between 20 and

60 MHz as can be viewed in Figure 3(a).

Figure 3: (a) Frequency response of the

Butterworth FIR filter and

(b) Profile of TGC function.

2.3 - TGC (Time Gain Compensation)

The ultrasound beam is attenuated as it penetrates

the tissue. To compensate for this loss in signal

intensity TGC is applied to each line in A-mode, which

is defined as

rerT 1)( (1)

where is the coefficient of attenuation and is the

radial distance from tip of catheter.

The range of the radial distance was extracted

from the RF File Reader of exam ranging until 4.48cm.

In Ciompi (2008), RF signal of in-vivo and ex-

vivo was used to develop a multiclass classifier to the

problem of characterization of the atherosclerotic

plaque. They define a value for the coefficient of

attenuation as , which was

adopted in this work.

The profile of TGC function is shown in

Figure 3(b).

2.4 - Signal envelope

To show the changes stemming from the texture

and not from the wave profile, the envelope of the

signal is obtained simply applying the Hilbert

transform to each line in A-mode from the RF signal.

Figure 4: RF envelope is shown as a black line over the

wave profile of RF signal gray line (Data).

2.5 - Log-compression

The next stage in the procedure described in

Figure 2, is to normalize the RF signal providing

values between 0 and 1 in order to permit work with a

homogeneous range for all IVUS images. After this,

the RF signal undergoes a transformation whose

purpose is to map a narrow range of grayscale values

in an input image to a wide range of output levels

(Gonzalez and Woods, 2004). This transformation is

defined as

nor

t Iet

I 11log1

log (2)

where norI represents the RF signal normalized and t

is empirically obtained to improve the log-

compression.

2.6 - Digital Development Process

In order to emphasize the edges borders and

improve contrast gain, a Digital Development Process

(DDP) (Gonzalez and Woods, 2004) was applied to the

RF signal.

(a)

(b)

Page 4: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

Each pixel value of an image was modified by

equation (3) to produce an image with better Contrast

Noise Ratio (CRN).

baX

XkY

ji

ji

ji

}{ (3)

where }{ jiX is obtained by applying a Gaussian low-

pass filter to the original image and the parameters k ,

a , and b were empirically determined to improve the

CRN.

After this, the image was converted and

interpolated to cartesian form, resulting in an image

with 512x512 pixels and 256 gray levels.

Finally, an intensity transformation was applied to

image in order to expand the saturation of the gray

level dynamic band and the image was Gaussian

filtered.

3. Experimental Results

The results of the rebuilding process of the IVUS

images are shown in Figure 5, which the mayor

structures visible in an IVUS examination are

identified.

Figure 5(a) shows the segmentation of lumen and

the media-adventitia borders. 5(b) the stent and an

artifact generated by the wire guide. 5(c) shows a

region with calcification and the acoustic shadow

behind it, with an arrow pointing to an artifact

generated by the wire guide.

Figures 5(e) and (h) show a bifurcation region,

with calcification. A stent is visible in 5(d) and (i), and

it is possible to identify the malposition of the stent in

5(i).

Figure 5(f) shows the shadow of the pericardium

and 5(g) the acoustic shadow of a big calcification and

the lumen and media-adventitia borders.

Figure 5: Images from the reconstruction process.

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Page 5: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

Figure 6 show both the rebuild image and the DICOM

images.

Figure 6: (a), (b), (c) and (d) DICOM images.

(e), (f), (g) and (h) Rebuilt images.

As can be seen, the rebuilt images show the same

structures as DICOM images, and in all cases the

contrast of the rebuild images are better than DICOM

images. What is perhaps most noticeable is the

difference in visibility in the outer region of the lumen.

The reconstructed image shows fine detail where the

DICOM shows only a black region.

4. Discussion, Conclusion and Future

Work

IVUS is an examination that can provide a good

quality image of the cross-section of blood vessel

allowing the assessment of inner structures.

In an IVUS medical examination, sets of hundreds

or even thousands of images are acquired and used as

the basis for a medical diagnosis.

These images are subject to a variability of

interpretation inter and intra operator because a set of

parameter are adjusted to improve the visualization of

a ROI. Once the images are acquired these parameters

cannot be changed, restricting the comparison between

different examinations or patients.

To avoid this limitation, this article describes a

methodology for reconstructing IVUS images from RF

raw data, which are independent of the parameters

adjusted by the physician during the exam and which

can be processed to improve the CNR of the image.

The RF signal is processed according to the

theoretic model proposed in section 2 and illustrated in

Figure 2. The parameters used in the model were

adjusted to maximize CNR enabling identification of

the main structures of the vessel.

The results of the proposed model were presented

in Figures 5 and 6 and compared with the DICOM

images generated by the equipment. The proposed

model produces images with superior CNR which can

be used for clinical purposes.

In the figures it is possible to see the main

structures of the vessel and this result can be used to

perform segmentation to help the physician in

diagnosis process. Beyond this, it is possible to identify

bifurcations and calcifications regions to be submitted

a percutaneous coronary intervention - PCI.

Considering the data used in this work, the

propose method was proved to be robust with regard to

fidelity in the reconstruction of structures in

comparison with DICOM image and, in all cases the

CNR in reconstructed images was greater than DICOM

images, figure 6.

5. References

[1] Noble, J. A., "Ultrasound image segmentation and

tissue characterization", Proceedings of the Institution

of Mechanical Engineers, Part H: Journal of

Engineering in Medicine vol. 224: 307, 2010. DOI:

10.1243/09544119JEIM604

[2] Mountford, R. A. and Wells, P. N. T. (1972).

“Ultrasonic liver scanning: The A-scan in the normal

and cirrhosis”. Phys. in Med. & Biol. 17, 261–269.

[3] Thijssen, J. M., "Ultrasonic speckle formation,

analysis and processing applied to tissue

characterization", Pattern Recognition Letters, vol. 24,

659-675, 2003

(a) (e)

(b) (f)

(c) (g)

(d) (h)

Page 6: Rebuilding IVUS Images From Raw Data of the RF Signal: A ...worldcomp-proceedings.com/proc/p2014/IPC3515.pdf · Rebuilding IVUS Images From Raw Data Of The RF ... ² Division of Informatics/Heart

[4] Tsui, P.-H., Yeh, C.-K., Chang, C.-C., and Liao,

Y.-Y. "Classification of breast masses by ultrasonic

Nakagami imaging: a feasibility study", Phys. Med.

Biol., 53, 6027-6044, 2008.

[5] Molthen, R. C., Shankar, P.M., Reid, J.M.,

Forsberg, F.,Halpern, E. J., Piccoli, C. W., and

Goldberg, B. B. "Comparisons of the Rayleigh and K

distribution models using in vivo breast and liver

tissue", Ultrasound in Medicine and Biology, 24, 93–

100, 1998.

[6] Clifford, L., Fitzgerald, P., and James, D. "Non-

Rayleigh first-order statistics of ultrasonic backscatter

from normal myocardium", Ultrasound in Medicine

and Biology, 19, 487-495, 1993.

[7] Nillesen, M. M., Lopata, R. G. P., Gerrits, I. H.,

Kapusta, L., Thijssen, J. M., and de Korte, C. L.,

"Modeling envelope statistics of blood and

myocardium for segmentation of echocardiographic

images". Ultrasound in Medicine and Biology, 34(4),

674–680, 2008.

[8] Lizzi, F. L., Greenbaum, M., Feleppa, E. J.,

Elbaum, M., and Coleman, D. J., "Theoretical

framework for spectrum analysis in ultrasonic tissue

characterization", Journal of Acoustic Society

America, 73, 1366–1373, 1983.

[9] Raju, B. I., Swindells, K. J., Gonzalez, S., and

Srinivasan, M. A., "Quantitative ultrasonic methods

for characterization of skin lesions in vivo", Ultrasound

in Medicine and Biology, 29(6), 825–838, 2003.

[10] Engelhorn, A. L. D. V., Engelhorn, C. A., Salles-

Cunha, S. X., Ehlert, R., Akiyoshi, F. K. e Assad, K.

W., "Ultrasound tissue characterization of the normal

kidney", Ultrasound Quarterly, vol. 28, nº 4, December

2012.

[11] Moradi, M,. “A New paradigm for Ultrasound-

Based Tissue Typing in Prostate Cancer”. Tese de

doutorado. School of Computing, Queen's University.

2008.

[12] Szabo, T. L., Diagnostic Ultrasound Imaging

Inside Out, Hartford, Connecticut, Elsevier, 2004.

[13] Ciompi, F. “Ecoc-based plaque classification

using in-vivo and ex-vivo intravascular ultrasound

data”. Master thesis. Computer Vision Center.

Universitat Autonoma de Barcelona. 2008.

[14] Gonzalez, R. C., Woods, R. E. e Eddins, S. L.

Digital Image Processing Using Matlab. Prentice Hall.

2004.

ACKNOWLEDGEMENTS

This work is supported by the Brazilian National

Institute of Science and Technology in Medicine

Assisted by Scientific Computing (INCT - MAAC)

and National Council for Scientific and Technological

Development (CNPq).