Author + information
- Received June 13, 2004
- Revision received October 19, 2004
- Accepted October 25, 2004
- Published online June 21, 2005.
- Akihiro Murashige, MD,
- Takafumi Hiro, MD, PhD⁎ (, )
- Takashi Fujii, MD, PhD,
- Koji Imoto, MD,
- Takashige Murata, MD,
- Yusaku Fukumoto, MD and
- Masunori Matsuzaki, MD, PhD
- ↵⁎Reprint requests and correspondence:
Dr. Takafumi Hiro, Division of Cardiovascular Medicine, The Department of Medical Bioregulation, Yamaguchi University Graduate School of Medicine, 1-1-1 Minami Kogushi, Ube, Yamaguchi 755-8505, Japan
Objectives This study examined the feasibility of using a wavelet analysis of radiofrequency (RF) intravascular ultrasound (IVUS) signals in detecting lipid-laden plaque.
Background Wavelet analysis is a new mathematical model for assessing local changes in a geometrical profile of time-series signals.
Methods Radiofrequency IVUS signals of 85 arbitrarily selected vectors were acquired from 27 formalin-fixed noncalcified atherosclerotic plaques from human necropsy with a digitizer at 500 MHz with 8-bit resolution by use of a 40-MHz IVUS catheter. Wavelet analysis of these RF signals was performed using a Daubechies-2 wavelet to obtain a color-coded map of the correlation coefficient with the wavelet reconstructed over the x-y plane of the wavelet scale and the distance from the IVUS catheter. The plaque segment was then examined histologically after being stained with Masson’s trichrome stain. This technique also was applied in vivo in 29 human coronary plaque segments. These segments were excised subsequently by directional coronary atherectomy and processed for histologic analysis.
Results In the in vitro study, histologic examination revealed lipid-laden segments in 29 vectors. When performing a wavelet analysis with the Daubechies-2 wavelet, the color-coded mapping revealed a different pattern in lipid-laden plaques compared with other types of plaque. Using this wavelet analysis, lipid-laden plaque could be detected with a sensitivity of 83% (24 of 29) and a specificity of 82% (46 of 56). In the in vivo study, fatty plaque could be detected with a sensitivity of 81% (13 of 16) and a specificity of 85% (11 of 13) with this method.
Conclusions Wavelet analysis of RF IVUS signals enabled in vitro as well as in vivo detection of lipid-laden plaque. This method may be useful in assessing plaque vulnerability in patients with coronary artery disease.
Because lipid-rich plaques with thin fibrous caps have been shown to be vulnerable to rupture as a major cause of acute coronary events (1,2), several attempts have been made to develop an imaging modality to identify such plaques before they rupture. Intravascular ultrasound (IVUS) imaging provides a detailed arterial cross section with accurate morphometric representation of atherosclerotic plaque dimensions in vitro and in vivo (3–15). However, subsequent studies have demonstrated significant limitations in tissue characterization by IVUS intensity patterns alone, especially in discriminating fibrous and fatty tissues (16–19). To overcome these limitations, many authors (19–24) have proposed methods of quantitative tissue characterization to discriminate fibrous and fatty plaque. However, none of these methods has been sufficiently well recognized as of yet for the appropriate equipment to be installed in commercially available IVUS machines.
Wavelet analysis is a new mathematical model for assessing local changes in the geometrical profile of time-series signals (25). Wavelet analysis is one of the time-frequency domain analyses of signals. This method discriminates a local unique wave pattern within a complex signal. The purpose of this study was to investigate the feasibility of using wavelet analysis of radiofrequency (RF) IVUS signals to detect lipid-laden plaque. The reliability of this method was first examined with in vitro atherosclerotic plaque segments from human necropsy. The parameters evaluated in this in vitro model were applied to an in vivo clinical setting and tested against the histology of the coronary segments excised with directional coronary atherectomy. The histology of the excised tissue was compared with the results of the wavelet analysis of RF IVUS signals.
In vitro IVUS study
Twenty-seven formalin-fixed noncalcified atherosclerotic plaques that were obtained from human femoral and coronary arteries excised from 10 patients at necropsy were imaged using a 40-MHz Atlantis Plus IVUS catheter (CVIS/Boston Scientific, Sunnyvale, California) in saline at room temperature. Eight of these patients died of heart failure with ischemic cardiomyopathy or old myocardial infarction, and two died of noncardiac events. The imaged arteries had plaques with a thickness >0.5 mm. The lumen area of the examine vessel was 10.48 ± 5.78 mm2(range, 1.57 to 25.2 mm2).
Calcified plaques were not studied in the present study because calcified tissue is identified readily by visual inspection with high sensitivity and specificity (13). The current concern for tissue characterization of plaque is how to discriminate between fibrous and fatty tissue. An acoustic reference point was determined by suturing a surgical needle into the wall of the artery perpendicular to the long axis. This technique ensured that the same cross section was imaged for all studies and that the ultrasound images corresponded exactly to the cross section chosen for histologic analysis.
The entire length of the artery was imaged initially by visual inspection using conventional IVUS video monitoring to find an optimal portion of atherosclerotic plaque that provided no significant change in tissue composition or structure within at least a 0.5-mm length of the artery. Care was taken to position the catheter centrally and coaxially. The ultrasound images were recorded on super VHS tape.
We sampled in vitro cross-sectional images of 27 noncalcified plaques in 21 atherosclerotic formalin-fixed artery specimens (coronary: n = 9; femoral: n = 12) with a commercially available IVUS machine (ClearView Ultra System, CVIS/Boston Scientific) and a 40-MHz IVUS catheter. The RF IVUS signals of 256 radial vectors, which completely surrounded (360°) the catheter center with an equal angle span (1.4°), were obtained from these plaques using an analog-to-digital converter installed inside the IVUS machine (Fig. 1).The analog-to-digital board was specially designed and installed by the IVUS manufacturer. Each cross section comprised these 256 RF IVUS signals, which were sampled in real time at 500 MHz in 8-bit resolution with a digitizer and then stored on hard disk for further analysis. Each cross-sectional IVUS image also was recorded on videotape. On a video screen, a radial line from the catheter center was superimposed on a conventional cross-sectional IVUS image to enable recognition of the location of each vector. The vectors analyzed were first obtained from the thickest portion of the plaque imaged. We next selected another two or more vectors at least 15 degrees away from the vector first selected. In other words, there were at least 10 vectors in between these vectors. Only the plaque portions, the thickness of which was more than 0.5 mm, were selected. The RF signals were excluded when the signals were from the regions with significant nonuniform rotational distortion, calcification, or drop-out in the conventional IVUS image. A total of 85 vectors were analyzed from all plaques imaged.
We analyzed the IVUS RF signals offline by wavelet analysis (25) using MATLAB data processing software (The MathWorks, Natick, Massachusetts). Wavelet analysis is a signal-processing tool that enables the detection of a special geometric pattern within a localized area of a signal. A wavelet is a short segmental waveform of limited duration that has an average value of zero. Wavelet patterns that meet various mathematical criteria have been proposed for comparison, such as Daubechies, Meye, and Mexican hat. Wavelet analysis involves the breaking up of a signal into shifted and scaled versions of the original (or mother) wavelet. The continuous wavelet transform is defined as the sum over time of the signal multiplied by scaled, shifted versions of the wavelet function:
This results in many wavelet coefficients, C, which are a function of scale and position. Multiplying each coefficient by the appropriately scaled and shifted wavelet yields the constituent wavelets of the original signal. Wavelet analysis then produces a time-scale view of a signal. “Scaling a wavelet” means stretching (or compressing) it. The greater the scale factor, the more the wavelet is stretched. This scale is related to the frequency of the signal. “Shifting a wavelet” simply means delaying (or hastening) its onset.
To obtain a wavelet analysis, the following steps are performed (Fig. 2)
1. Take a wavelet and compare it to a section at the start of the original signal.
2. Calculate C, the coefficient between the section and the wavelet, which represents how closely correlated the wavelet is with this section of the signal. The higher Cis, the greater the similarity. The results will depend on the shape of the wavelet selected.
3. Shift the wavelet to the right and repeat steps 1 and 2 until the whole signal is covered.
4. Scale (stretch) the wavelet and repeat steps 1 through 3.
5. Repeat steps 1 through 4 for all scales.
This process produces wavelet coefficients (C ) that are a function of scale and position. The commercially available program for wavelet analysis used in this study automatically selected the minimal scale of the wavelet to correspond to the minimum sampling interval.
After taking these steps, the coefficients are produced at different scales by different sections of the signal. The coefficients constitute a regression of the original signal performed on the wavelets. The results can be represented graphically, in which the x-axis represents position along the signal (time), the y-axis represents scale, and the color at each x-y point represents the magnitude of the wavelet coefficient C. In this map, correlation coefficients are shown using a blue-pink scale, in which pink represents higher values of the coefficient and blue represents lower values.
Preliminary in vivo application
The same technique was applied in vivo to 29 coronary plaque segments from 13 patients (65 ± 6 years; range, 54 to 74 years) with coronary artery disease (7 patients with stable angina, 6 with acute coronary syndrome). The RF IVUS signals were obtained from the thickest part of the plaque imaged. These segments were excised by directional coronary atherectomy (FLEXI-CUT/Guidant, Indianapolis, Indiana) and processed for histologic analysis. Plaque segments were excluded, which were insufficiently debulked by the atherectomy, leaving a residual plaque area of more than one-third of the original plaque area. In the histologic preparation, the specimens were stained with hematoxylin-eosin stain and Azan stain. This study was approved by the institutional review committee, and patients gave informed consent.
In the in vitro study, after the arteries were imaged by IVUS, the needle for acoustic reference was removed, and the needle site marked with India ink. The specimens were processed for histology and stained with Masson’s trichrome stain. The IVUS and the histologic examinations were performed by different observers. A plaque was defined as lipid-laden by visual inspection, when a lipid-core was >50% of the total plaque area. A lipid core was defined as a contiguous area of lipid-containing foam cells, extracellular lipids, cholesterol crystals, a lipid pool, or necrotizing material. A plaque was defined as fibrous when it had no distinct lipid core but had a fibroacellular matrix with dense collagen bands. The thickness of the lipid core had to be >0.3 mm and >50% of the total plaque area to be included in this study.
In the in vivo study, the directional coronary atherectomy specimens were stained with hematoxylin-eosin and Azan stains. This study only included typical fatty-dominant or fibrous-dominant plaques. The fatty-dominant plaques contained a lipid core >80% of total plaque area. The fibrous-dominant plaques contained a fibrous area >80% of total plaque area.
Values were expressed as means ± standard deviation. Receiver operating curve analysis was performed to discriminate the optimal criteria in the interpretation of the results of this wavelet analysis.
In vitro study
The mean thickness of plaque examined in this study was 1.42 ± 0.47 mm. Histologic examination revealed that 29 of 85 vectors of RF signals analyzed were from a lipid-laden plaque. Representative examples of wavelet analysis of RF IVUS signals from a lipid-laden plaque and from a fibrous plaque are shown in Figure 3.Wavelet analysis of the RF signals with a Daubechies-2 wavelet function provided an apparently different pattern in the color-coded mapping between scale 20 and scale 30. In this time-scale graphic representation of wavelet analysis of RF IVUS signals from a plaque with a lipid core, a different pattern of pink mapping was observed that was not observed from a fibrous plaque without a lipid core. A lipid-laden zone frequently was present, when the wavelet coefficient (C ) was more than a certain value compared with a wavelet whose scale is between 20 and 30. The ROC analysis revealed that the optimal value of this wavelet coefficient was 0.6 to discriminate a lipid-laden plaque (Fig. 4).Using this criteria, the lipid-laden plaque was detected in this in vitro setting with a sensitivity of 83% (24 of 29) and a specificity of 82% (46 of 56) (Table 1).Many other wavelet approaches (approximately 50 types) were analyzed, and none provided the sensitivity and specificity of the Daubechies-2 method.
In vivo study
Histologic examination from the directional coronary atherectomies revealed that 16 of 29 coronary segments were fat-dominant (lipid-laden). No apparent fatty area was observed histologically in the remaining 13 segments. In the lipid-laden plaques, the wavelet analysis with the Daubechies-2 wavelet function revealed a similar pattern as the in vitro results (Fig. 5).Using the same criteria of the wavelet analysis as in the in vitro study, fatty plaque could be detected from the clinical material with a sensitivity of 81% (13 of 16) and a specificity of 85% (11 of 13).
The present study is the first report of in vitro as well as in vivo tissue characterization of atherosclerotic plaque using a wavelet analysis of RF IVUS signals. The major finding of this study is that this wavelet method is accurate in detecting lipid-laden atherosclerotic plaque. This method may be useful in assessing plaque vulnerability in patients with coronary artery disease.
Advantages of wavelet analysis
The theoretical basis of wavelet analysis was first developed by Grossmann and Morlet in 1983 (25). Wavelet analysis is a time-frequency domain analysis of signals. The most well known of these is Fourier analysis, which breaks down a signal into constituent sinusoids of different frequencies. The Fourier transform was modified into a transform to analyze only a small section of the signal at a time by looking at “windows” of the signal. This short-time Fourier transform provides some information about when and at what frequencies a signal event occurs. The major drawback of this method is that once a particular size for the time window is chosen, that window is the same for all frequencies. If the window size is changed to a shorter one to increase time (space) resolution, the frequency resolution is compromised. Wavelet analysis was proposed in an attempt to overcome the problems in resolution.
Wavelet analysis represents a windowing technique with variable-sized regions. Wavelet analysis allows the use of long-time intervals when more precise low-frequency information is needed and shorter regions when high-frequency information is needed. One major advantage of wavelets is their ability to analyze a localized area of a larger signal. In this study, the Daubechies-2 wavelet proved best for detecting a lipid-laden plaque. An empirical selection of wavelet has to be made when applying wavelet analysis in a novel field of data. If a new wavelet family is developed, the sensitivity and specificity for detection of fatty tissue may be improved.
Wavelet scales 20 and 30 correspond to wavelengths of 32 and 47 μm, respectively. A scale of <20 is less than conventional IVUS resolution (26) or the ultrasound pulse wavelength. The results from wavelet analysis with a wavelet scale <20 would measure artificial noise only. A higher value of wavelet correlation coefficient represents an acoustic signal derived from a more complicated structure. Compared with a fibrous area, a fatty area usually is composed of various kinds of tissue, such as lipid-laden foam cells, cholesterol crystals, extracellular lipids, necrotizing material, and fibers, which may be intermingled in a way that could produce complex acoustic impedance mismatches inside the plaque (17). Therefore, a lipid-laden area provides a higher value of wavelet correlation coefficient with a shorter scale of wavelet.
Comparison with other methods of tissue characterization
It was originally expected that tissue components within plaque could be identified from the video-intensity pattern of IVUS images (4,5,7,12–15). Subsequent studies, however, demonstrated significant limitations of tissue characterization by IVUS intensity patterns alone, especially in discriminating fibrous and fatty tissues or in assessing plaque vulnerability (16–19). To overcome the limitations, some authors (20–23) have proposed several methods of quantitative tissue characterization to discriminate fibrous and fatty plaque, including RF signal analysis, such as integrated backscatter analysis, attenuation slope mapping (19,24), and spectral analysis (27). Recently, IVUS elastography was proposed as a novel modality of tissue characterization with IVUS (28). Our laboratory previously reported that color mapping of the angle-dependent echo-intensity was useful for detecting fibrous caps within plaques (29). However, this method has difficulties in detecting other type of tissues. Because none of these previously reported techniques has become available commercially, no study has yet compared their clinical feasibility using the same subjects.
For the in vitro study, the arteries were imaged after they were fixed in formalin at room temperature. It is unknown whether formalin fixation or change in temperature will alter the results of this analysis. Another limitation was the use of nonpressure-distended arteries. When removed from physiologic pressure, atherosclerotic arteries contract. This contraction could significantly alter the architecture, which might affect the wave pattern of the RF IVUS signal. However, the in vivo application of the wavelet analysis also offered similar sensitivity and specificity for identifying a lipid-laden plaque as in the in vitro study. Therefore, these effects appear to be negligible in this study.
This wavelet analysis was performed for one single vector. The single vector analysis is subject to mismatch because of rotation of the images. To minimize any mismatch, we superimposed a radial line from the catheter center onto a conventional cross-sectional IVUS video image to enable the recognition of the location of each vector. In the in vitro study, all the plaques analyzed had a thickness >0.5 mm, and any lipid core had a thickness >0.3 mm. Therefore, we do not know whether it is possible to analyze thinner plaques or to identify very thin lipid cores with this method. Furthermore, the presence of blood and phasic pressure within the lumen as well as any noncoaxial alignment of the catheter may impair appropriate analysis in vivo with this method.
This study was performed on the off-line basis, taking an hour or so to obtain each color map. Therefore, a further development is necessary to be able to provide an on-line plaque evaluation during the study so that immediate feedback is given to the operator.
The present study demonstrates the feasibility of in vitro as well as in vivo tissue characterization by wavelet analysis of RF IVUS signals. Using wavelet analysis, lipid-laden plaque could be detected with a sensitivity and specificity of >80%. This method may be useful in assessing plaque vulnerability in patients with coronary artery disease. Currently, there is no reliable, commercially available device that is capable of discriminating fibrous and fatty areas within atherosclerotic plaque. Detection of vulnerable plaque or sequential observations of the stabilizing effect of lipid-lowering therapy on plaque composition with acceptable accuracy in vivo could improve the management of patients with coronary artery disease. Further evaluation of wavelet analysis in comparison with clinical data and inflammatory markers will be necessary to assess its usefulness in clinical practice to predict future cardiac events in patients with coronary artery disease.
This work was partly supported by a grant-in-aid for scientific research of the Ministry of Education, Japan (grant No. 13670715). Health and Labour Sciences research grants: Comprehensive Research on Cardiovascular Diseases from Ministry of Health, Labour, and Welfare of Japan, and Knowledge Cluster Initiative of the Ministry of Education, Japan.
This study was presented in part at the 75th scientific sessions of the American Heart Association, Chicago, Illinois, 2002. Drs. Murashige and Hiro contributed equally to this work.
- Abbreviations and Acronyms
- intravascular ultrasound
- Received June 13, 2004.
- Revision received October 19, 2004.
- Accepted October 25, 2004.
- American College of Cardiology Foundation
- Libby P.
- Yock P.G.,
- Johnson E.L.,
- Linker D.T.
- Gussenhoven E.J.,
- Essed C.E.,
- Lancee C.T.,
- et al.
- Tobis J.M.,
- Mallery J.,
- Mahon D.,
- et al.
- Nissen S.E.,
- Grines C.L.,
- Gurley J.C.,
- et al.
- Potkin B.N.,
- Bartorelli A.L.,
- Gessert J.M.,
- et al.
- Nissen S.E.,
- Gurley J.C.,
- Grines C.L.,
- et al.
- Jeremias A.,
- Kolz M.L.,
- Ikonen T.S.,
- et al.
- Kawasaki M.,
- Takatsu H.,
- Noda T.,
- et al.
- Kawasaki M.,
- Takatsu H.,
- Noda T.,
- et al.
- Nair A.,
- Kuban B.D.,
- Tuzcu E.M.,
- et al.
- de Korte C.L.,
- Sierevogel M.J.,
- Mastik F.,
- et al.
- Hiro T.,
- Fujii T.,
- Yasumoto K.,
- et al.