Copyright
Ultrasound in Med. & Biol., Vol. 23, No. 2, pp. 215-229, 1997 0 1997 World Federation for Ultrasound in Medicine & Biology Printed in the USA. All tights reser& 0301-5629/97 $17.00 + .OO
PI1 SO301-5629( 96) 00201-3
ELSEVIER
@Original Contribution PARAMETRIC IMAGES
(INTEGRATED BACKSCATTER AND ATTENUATION) CONSTRUCTED USING BACKSCATTERED RADIO FREQUENCY SIGNALS (25-56 MHz) FROM HUMAN AORTAE IN VZTRO
BRUNEVAL~ and GENEVII~E BERGER* *Laboratoire d’Imagerie ParamWque URA CNRS 1458, Paris, France; and ‘Laboratoire d’Anatomie Pathologique INSERM U 430. HGpital Broussais, Paris, France
S. LORI BRIDAL, * PAUL FORNI?S,+ PATRICK
(Received 8 May 1996;in jinal form 6 September 1996)
Abstract-Quantitative ultrasonic tissuecharacterization using backscatteredhigh-frequency intravascular ultrasound could provide a basisfor the objective identlilcation of lesionsin vivo. Representationof local measurementsof quantitative ultrasonic parameters in a conventional image format should facilitate their interpretation and thus increasetheir clinical utility. Toward thii goal, the apparent integrated backscatter, the slope of attenuation (25-56 MHz) and the value of the attenuation on the linear fit at 37.5 MHz were measuredusing the backscattered radio frequency signalsfrom in vitro human aortae. Local estimations of these nltrasonic parameters from both normal and atherosclerotic aortic segmentswere displayed in a B-scan format. The morphological features of these parametric imagescorresponded well to features of histological imagesof the sameregions. The attenuation from 25-56 MHz of seven segmentsof the medial layer (both with and without overlying atheroma) were measuredusing the multinarrow-baud backscatter method. The average attenuation in the media at 24°C 2 3°C was 45 rl: 16 dB/cm at 25 MHz and 102 ? 13 dB/cm at 50 MHz. This work represents progress toward the development of quantitative imaging methodsfor intravascular applications. 0 1997World Federation for Ultrasound in Medicine 8~Biology. Key Words: Ultrasound, Ultrasonic tissue characterization, Vascular tissue, Atherosclerosis, Attenuation, Backscatter, Parametric images,Acoustic microscopy. INTRODUCTION
Atherosclerotic disease leads to the majority of ischemit events responsible for myocardial infarction and stroke. Currently, angiography is the prevalent method for the diagnosis of atherosclerosis. This technique provides a view of the lumen from which regions of narrowing may be detected and a classification of the degree of stenosis made. Intravascular ultrasonic images present an annular cross-section of the arterial wall and may be used, subsequent to the detection of a stenosis, to quantify the thickness of the plaque or to monitor cardiovascular intervention. The apparent echogenicity of the image may also be evaluated as a first approach to plaque identification. The potential evolution of an atherosclerotic lesion and its response to interventional therapy may depend not only on its extent, but also on its biomeAddresscorrespondence to: S.Lori Bridal,Ph.D.,Laboratoire d’Imagerie ParamCtrique, URA CNRS 1458, 15 rue de 1’Ecole de MBdecine, 75006Paris,France.
chanical properties, which are linked to its biochemical composition and structure. Lesions with lipidic cores covered by a fibrous cap appear to be susceptible to rupture and thrombosis due to the concentration of stress at the shoulder regions of the caps, while fibrous lesions are considered to be more stable (Davies and Thomas 1985; Lee et al. 1992). Thus, a technique for the quantitative imaging of the arterial wall could provide a means to better predict the potential evolution of atherosclerotic plaque. The ultrasonic parameters of backscatter and attenuation are related to tissue structure and biochemical composition (Lizzi et al. 1988; Nassiri and Hill 1986a; O’Donnell et al. 1981; Olerud et al. 1990) and may, therefore, offer indexes for the identification of lesion types (Nassiri and Hill 1986a; O’Donnell et al. 1981; Wear et al. 1989). Several researchers have shown that the magnitude and frequency dependences of attenuation and backscatter may be used as quantitative indexes of material condition (Bamber 1979; Garra et al. 1994; Insana et al. 1991; Nassiri and Hill
216
Ultrasound in Medicine and Biology
1986b). Beyond this, the size and scattering strength of the effective scatterers in biological tissue may be estimated based on measurements of the backscatter coefficient (Bamber 1979; Garra et al. 1994; Insana et al. 1991; Nassiri and Hill 1986b). Local estimations of ultrasonic material characteristics (attenuation, backscatter coefficient) or parameters describing tissue microstructure (effective scatterer size, net scattering strength) may be displayed in the form of an image (parametric imaging). Recent research has demonstrated the potential diagnostic value of images of several such parameters in tissue-mimicking phantoms (Insana and Hall 1990)) liver (Lizzi et al. 1988 ) and ocular tissues (Silverman et al. 1995). The extension of such techniques for intravascular application could provide a valuable method for atherosclerotic plaque characterization by allowing a quantitative view of lesion structure. Quantitative measurements of acoustic parameters in vascular tissues were first obtained at lower frequencies ( 1- 15 MHz) and have, more recently, been extended to higher frequencies ( 15 -65 MHz) (De Kroon et al. 1991; Lockwood et al. 1991; Wickline et al. 1993, 1994; Wilson et al. 1994). Much of this work has concentrated on the correlation of the apparent backscatter or slope of attenuation with plaque type. Early measurements at lower frequencies provided evidence that apparent backscatter and attenuation could potentially distinguish between arteries burdened with different types of plaques. Measurements at these lower frequencies may be performed on peripheral arteries, such as the carotid, in a totally noninvasive manner. However, the resolution of these lower frequency measurements is limited, and, thus, measurements represent the average properties of the arteries across a relatively large region of the arterial wall. In spite of this limitation, Barzilai et al. ( 1987) and Urbani et al. (1993) were able to differentiate between normal, fibrofatty, fatty and calcific regions of arterial wall using the apparent backscatter, and Picano et al. (1983, 1985) and Greenleaf et al. (1974) observed differences in the attenuation across the walls of arteries having different types of lesions. The use of higher frequency, higher resolution transducers has allowed more local characterization of smaller regions of arteries, such as atherosclerotic plaques. The anisotropy of the directivity and the backscattered power (measured with a 27-MHz broadband transducer) has been reported to be related to plaque type (De Kroon et al. 1991). Using a rabbit model, Wickline et al. (1993, 1994) demonstrated higher backscattered power levels between 30 and 55 MHz from fibrous plaques than from fatty plaques. Although both the apparent backscatter and atten-
Volume 23, Number 2, 1997
uation are, in principle, obtainable from the echographic signal alone, to date, nearly all attenuation measurements reported in arteries have been conducted using either a through transmission or a substitution measurement, which is not feasible in viva and which measures only the average attenuation across the total thickness of the artery wall. In human arteries, Lockwood et al. ( 1991) measured the attenuation between 35 and 60 MHz across the entire artery wall in regions excluding atheroma using a substitution method. They then compensated backscattered radio frequency (rf) signal from regions of interest (ROIs) (adventitia, media and thickened intima) within the artery for this global attenuation. Their results showed that the backscatter coefficient was dependent not only on the layer of the artery wall (adventitia, media and thickened intima), but also on the type of artery (elastic vs. muscular) and the direction of insonification. The first measurements of attenuation in arteries, performed using echographic rf signals only, were made by Wilson et al. (1994) using a medical intravascular system (bandwidth 13-25 MHz) and a multinarrow-band measurement method. Measurements were used to construct parametric images of the attenuation slope with an approximate axial resolution of 900 pm. They observed relative differences in attenuation slope between normal vessel walls and different plaque types. Thus, reports indicate that backscattered power and attenuation provide useful information for plaque discrimination and that, by using higher frequencies ( 15-65 MHz), these parameters may be obtained for local regions of the arterial wall. However, only a few local measurements from different plaque types obtained using echographic signals alone have been,reported (De Kroon et al. 1991; Wickline et al. 1993, 1994; Wilson et al. 1994)) and, therefore, the sensitivity of these measurements to plaque type and their clinical applicability have not been fully established. This article describes a method for the local measurement of quantitative ultrasonic parameters from the backscattered rf signal (attenuation, slope of attenuation and apparent integrated backscatter [ AIBS ] ) . Images of in vitro normal and atherosclerotic human aorta constructed with these parameters are compared to histological images. Measurements of the ultrasonic parameters (AIBS, slope of attenuation and attenuation) from both normal and atherosclerotic aortae are presented in an image format, and several ROIs are selected on these images for which the average value of the imaged parameter is calculated and displayed. The attenuation from regions of the media of both normal and atherosclerotic aortae measured using echographic signals is reported.
Parametric (integrated backscatter and attenuation) images l S. L. BRIDAL et al.
MATERIALS
AND METHODS
Specimens Segments of human thoracic aortae were obtained at autopsy from seven patients (middle-aged men) autopsied <12 h after death. The excised aortae were cut along the longitudinal direction of the artery, rinsed and stored at -80°C prior to ultrasonic measurements. Several researchers have demonstrated that the speed of sound, attenuation and level of backscatter from arteries are unchanged by freezing (Gussenhoven et al. 1989; Lockwood et al. 1991). The sample was brought progressively to room temperature (24°C 2 3°C) and pinned to a rubber backing. Two stainless steel pins were inserted at right angles to each other to form a cross in the plane of the artery surface to be used as an ultrasonic reference point. The specimen was then placed in a 0.9% saline solution with the intimal side of the artery facing the ultrasonic transducer. Acoustic measurements A schematic diagram of the acoustic backscatter microscope system is shown in Fig. 1. This system allows the acquisition of rf data from a three-dimen-
Computer
I
Motion Controller
217
sional volume by displacing the transducer in the x and y planes between scan sites using computer-controlled stepper motors. The transducer used was a Panametrics (Waltham, MA, USA) 75-MHz, 3-mm diameter, 12.7mm focal length PVDF broadband transducer, The wavelengths in the useful bandwidth ranged from 6027 pm. The beam width at the focus was estimated to be 210 pm by measuring the -6 dB full width of the transverse (off-axis) profile of the signal scattered from a point target at the focal distance of the transducer. The -20 dB bandwidth of the reference spectrum from a polished stainless steel reflector at the focal length was 15-70 MHz, with maximum power at 50 MHz. The downshift in frequency between the nominal frequency of the transducer and the peak frequency measured at the focus results from the frequency dependence of the attenuation in the water path between the transducer and the reference reflector. The transducer was focused on the surface of the artery near the reference cross and was displaced until the specular reflections from the needles were found and the center of the cross was located. This point was selected to be (0,O) in the coordinate system of the x and y scan plane. Relative to this reference point, the coordinates at the limits of each scanned region of
400 MHz Digitizing Oscilloscope
Pulser Receiver 1 kHz to 200 MHz
Transducer Specimen Fig. 1. Data acquisition system. The aorta was placed in a 0.9% saline solution with the intimal side facing the transducer. The acoustic backscatter microscopy system consisted of a 75MHz Panametrics 3-mm diameter, 12.7mm focal length PVDF transducer; Microcontrole stepper motors that permit displacement of the transducer in the x and y planes, with l-pm step resolution; a Panametrics 5900 pulserjreceiver with 2OO-MHz pass band; and a LeCroy 9450A 400-MHz digitizing oscilloscope. The data acquisition was controlled by a Dell 486D/66 personal computer.
Ultrasound in Medicine and Biology
Volume 23, Number 2, 1997
DATA
Fig. 2. Histological section stained for hematoxylin-eosinsaffron showing a normal aorta with a thin intima (I), normal elastic media (M) , adventitia (A) and lumen (L) . The black vertical bar is 1 mm long. The histological image shows a slightly smaller length of the arterial segment than is seen in the corresponding ultrasonic images in Fig. 4.
plaque were noted. The transducer was displaced along the axial z direction so that the entire thickness of the plaque and artery wall of interest was within the depth of field of the transducer for which diffraction correction methods have been validated ( 1.5 mm approximately centered at the focal length of the transducer, see Discussion section). Data were acquired across a total x-y scan plane 1 cm x 0.5 mm with 100 pm between rf lines (101 rf lines in the circumferential dimension of the artery, six rf lines in the longitudinal dimension of the artery). Each rf signal was averaged temporally 256 times, digitized (2000 sample points, 400 MHz) and stored for off-line analysis. Histological study The specimen was removed from the saline bath. The limits of the scanned region noted during the acoustic measurement were located by measuring on a grid from the reference point of the crossed needles. India ink was injected at these points to provide a marker visible in histological images. Following fixation in 4% formalin and, if necessary, decalcification, a 3-mm thick slice of the aorta centered upon the inkmarked region of each scan was obtained and embedded in paraffin. Two sections, 5 pm thick, were obtained from each paraffin block and stained with hematoxylin-eosin and orcein (for elastic fibers). Before data analysis, each histological section was viewed with an optical microscope to confirm that the morphological cross-sections between the ink marks corresponded well to the conventional B-scan ultrasonic image obtained from the same specimen. Two histological sections stained for hematoxylin-eosin are shown in Figs. 2 and 3.
ANALYSIS
Dijtkaction correction The high-frequency, tightly focused qualities of the transducer used in this work result in significant diffraction of the beam (Laugier et al. 1985). The method used to compensate for the diffraction was the following. A steel plate was positioned normal to the beam at an axial distance z from the transducer. The echo from the steel plate was gated using a 64-point (120 pm) window, Fourier transformed and squared in magnitude to obtain the reference power spectrum S,(J z), where f is the frequency. This process was repeated with the steel plate at many different distances z from the transducer. After this set of diffraction correction spectra had been obtained, backscattered rf signals were collected from the specimen as described in the Materials and Methods section. Each rf line backscattered from the specimen was gated using a 64-sample (120 pm) sliding Hamming window with 50% overlap. The rf signal in each gated region was Fourier transformed, squared in magnitude and averaged with the power spectra from the corresponding gated regions at the same distance z from the transducer of N neighboring rf lines to provide a spatially averaged apparent scattered power spectrum (S,(f; 2)). The number of rf lines N across which spatial averaging was obtained was 18 for the spectra used to form parametric images. Each average power spectrum (S,(f; z)) was corrected for the system response and diffraction effects using the reference spectrum S,,(f;
Fig. 3. Histological section stained for hematoxylin-eosinsaffron showing an atherosclerotic segment of aorta. Two lipidic cores (LC) surrounded by dense fibrous tissue (F) are visible. The lipidic cores are rich in cholesterol clefts (arrow heads) and are separated from the arterial lumen (L) by a thin fibrous cap (c). Some calcifications are present (arrows). The media (M) is normal. The black vertical bar is 1 mm long. The histological image shows a slightly smaller length of the arterial segment than is seen in the corresponding ultrasonic images in Fig. 5.
Parametric (integrated backscatter and attenuation) images 0 S. L. BRIDAL ef al.
219
z) obtained at the same distance z from the transducer according to:
= 10 ~ogm[Kf >I - &
yielding the corrected spatially averaged power spectra noted as &Af; z)). Apparent integrated backscatter estimation The average power between 25 and 56 MHz of each corrected spatially averaged power spectrum (J&(J; z)) was calculated to measure the AIBS:
*IBSfz)
1 = fmnx - fmin s (&Aft
z))df,
(4)
Assuming that the measurement region is homogeneous, the attenuation may be estimated at each frequency from the slope of a linear regression fit to the log decay of the power at that frequency as a function of the propagation distance in the material (multinarrow-band method). The amplitude attenuation coefficient may be expressed (in dB/unit length) as: a( f ) [ dB/unit
length] = 8.69cx( f ) [ Neper/unit
f,
CfPzr.
length] .
( 5)
(2)
Attenuation slope and attenuation estimation The multinarrow-band method for the estimation of the attenuation was chosen because it allows diffraction correction and provides the measurement of the attenuation at each frequency in the bandwidth as well as the slope of attenuation across the bandwidth. Details of this method have previously been described (Fink and Cardoso 1984), and the method has been validated for the acoustic microscope system (Roberjot et al. 1996). The spatially averaged attenuated power spectrum (S&f; z)) corrected for diffraction and the system response may be expressed as follows:
Estimation of the attenuation coefficient at each frequency in the bandwidth is performed in this manner, and the slope of attenuation p is obtained from the linear fit to the attenuation coefficient as a function of frequency. This fit is of the form a(f) = P(f - fc) + cQ‘~), where fc is the frequency at the center of the useful bandwidth, and, thus, the fit line is not forced to have a y-intercept of zero. In addition to the slope of the linear fit, the value of the attenuation at a selected frequency J on the fit line provides an estimate of the attenuation at that frequency a(J). Calculation of the propagation distance requires knowledge of the velocity in the medium. Velocities measured with IO-MHz transducers across normal human aortae have been reported to range from 1492-1534 m/s at 20°C (Greenleaf et al. 1974). The velocity has also been measured at 50 MHz to be between 1579 and 1628 m/s (Lockwood et al. 199 1 ), but these measurements were obtained at a significantly higher temperature (37°C) than the measurements reported here. A value of 1500 m/s based on the velocity measurements performed at temperatures comparable to the temperature of acoustic measurements presented in this work was used in all calculations.
where B(f) contains the backscatter transfer function of the specimen and a(f) is the amplitude attenuation coefficient in this region in units of Neper/unit length. The distance z is the entire one-way distance (saline and specimen) between the transducer and the region of backscatter, and 2zi is the round-trip propagation distance in the attenuating specimen. The logarithm of eqn (3 ) may be expressed as:
Local estimation of ultrasonic parameters The data acquired from an ultrasonic scan represent the scattering from a volume of the artery (nlongitudinal dimension of the artery by y-circumferential dimension of the artery by z-axial or temporal depth of each r-f line). By analyzing the data from a windowed region of a small number of adjacent rf signals forming a small volume V, a local estimation of an ultrasonic parameter can be obtained. By repeating this procedure on small neighboring volumes within the scanned region, a map of these local parameters can
where the integral is performed numerically using the method of trapezoids. The AIBS has been corrected for the system response and diffraction, but remains apparent in the sense that the backscattered power has not been compensated for the attenuation of the signal across the intervening material between the transducer and the gated region.
220
Ultrasound in Medicine and Biology
be made. Because of the stochastic nature of the echographic signals, a reduction in the size of the local measurement volume V leads to an increase in the variance of the estimated parameter. In tissue such as atherosclerotic arteries, which can be very heterogeneous, the task of optimizing the spatial averaging and the resolution of measurements is difficult. For these measurements, the choice was made to divide the data from each scan into 33 subgroups of 6 X 3 neighboring rf lines (longitudinal 500 pm X circumferential 200 pm). The resolution along the axial direction was limited to 64 sample Hamming gated segments ( 120 pm) with 50% overlap. The Fourier transform of each segment was obtained, and the average power spectrum compensated for the system response and diffraction was calculated for 64 sample segments of each group of 18 rf lines. The AIBS was estimated from each of the resulting spectra. The volume of the artery represented by each of these estimations has circumferential and axial dimensions of 200 pm and 120 ym, respectively, and a dimension of 0.5 mm along the longitudinal dimension of the artery. Because methods for the measurement of the attenuation and the slope of attenuation are based on spectral change with depth, further axial resolution was sacrificed to provide a more stable estimation (Berger et al. 1987). Attenuation slope values and the value of the attenuation (L(A) measured at a selected frequency were estimated based on groups of six consecutive spectra (in the axial direction), thus increasing the axial dimension of the measurement volume to 420 pm. The volumes were displaced along the axis by one spectrum between measurements. This significant axial overlap between the measurement volumes led to artifactual lengthening of features such as interfaces in the resulting images but allowed the visualization of the attenuation slopes resulting from all possible combinations of spectral groups.
Volume 23, Number 2, 1997
The AIBS images were thresholded to have a value of 0 for all AIBS values less than or equal to the approximate maximum level of the noise (-65 dB) and a value of 1 elsewhere. Each thresholded AIBS image was multiplied by the attenuation and attenuation slope images of the same region to mask out the values of the parameters estimated from noisy signals. The attenuation image of (u(h) is displayed from 0 (values 5 0 are black) to 250 dB/cm (values 2 250 are white) _ This range of values was compressed for display using an c~(fi)“~ power law. The slope of attenuation measurements are presented on a linear 256 level grey scale with values zz -2 dB cm-’ MHz-’ mapped to black and values z 8 dB cm-’ MHz-’ mapped to white. The volume represented by a voxel in both types of attenuation images is 500 pm X 200 pm X 420 pm in the longitudinal, circumferential and axial dimensions, respectively, with 85% axial overlap between the measurement volumes. Attenuation slope images
Image representation
The local measured values of the AIBS were displayed in a B-scan format. Each voxel in these images X 200 pm (circumferential) is 500 pm (longitudinal) X 120 pm (axial), with 50% overlap between the volumes of measurement in the axial direction. Values < -65 dB (approximate maximum level of the noise) were mapped to black. Values > -35 dB were mapped to white, with a linear range of 256 grey levels used to represent values between. Examples of AIBS images are shown in Figs. 4b and 5b. Conventional B-scan images based on the envelopes of the rf signals from a central plane of the scanned region are shown in Figs. 4a and 5a for comparison. The images in Figs. 4 and 5 correspond to the aortic segments shown in the histological sections of Figs. 2 and 3, respectively.
Fig. 4. Ultrasonic images of the segment shown in Fig. 2. (a) Conventional B-scan ent integrated backscatter between 25 dB); (c) Slope of attenuation between (in dB cm-’ MHz-‘); and (d) Attenuation dB cm-‘).
of normal media image; (b) Apparand 56 MHz (in 25 and 56 MHz at 37.5 MHz (in
Parametric (integrated backscatter and attenuation) images
Fig. 5. Ultrasonic images of the atherosclerotic artery shown in Fig. 3. (a) Conventional B-scan image; (b) Apparent integrated backscatter between 25 and 56 MHz (in dB ) ; (c ) Slope of attenuation between 25 and 56 MHz (in dB cmMHz-‘); and (d) Attenuation at 37.5 MHz (in dB cm-’ ).
are shown in Figs. 4c and 5c, and attenuation near band center at 37.5 MHz in Figs, 4d and 5d. Regions of interest may be selected on these images for the calculation of an average value of the AIBS, attenuation or slope of attenuation. Such regions and the corresponding measured values are shown in Figs. 4 and 5. Measurements from homogeneous regions of interest Regions of interest in which the attenuation was positive throughout (away from interfaces) were selected from the images of the attenuation at 37.5 MHz. The block of rf lines corresponding to the ROI selected on the attenuation image was extracted from the original data set. Each of these rf lines was gated using a 64-sample sliding Hamming window with 50% overlap. The rf signal in each gated region was Fourier transformed, squared in magnitude and spatially averaged with the power spectra from the gated segments at the same depth of all the other rf lines. The resulting
l S. L. BRIDAL
et ~1.
221
average spectra were compensated for diffraction according to eqn ( 1). Then the slope of attenuation and attenuation for the ROI were calculated based on their decay as described in eqn (4). This approach differed from the measurements from the parametric images in that all the spectra at the same depth but different circumferential and longitudinal positions of the ROI were spatially averaged initially, and the slope of attenuation and attenuation for the ROI were calculated based on the decay of these average spectra with depth across the entire axial extent of the ROI. Thus, this method of analysis provided more initial spatial averaging. Additionally, there is considerable axial overlap between the local volumes used for parametric attenuation and attenuation slope images, resulting in more local values based on spectra at the axial center of the ROI than at the axial limits. Measurements of the attenuation and attenuation slope from the blocks of rf treated as a whole result in a single estimate in which each spectrum is weighted equally. Before accepting a value from a ROI, a method for the identification of regions that are too heterogeneous for reliable parameter estimation was applied. This method (Laugier et al. 1985) is based on the fact that, in a homogeneous medium, the decay of the spectral centroid is linear and, thus, a perturbation in this decay indicates a heterogeneity. The centroid fc( z) of each spectrum (&,,.(f: =;)) was calculated according to:
f,(z) =z,0 where mj( z) is the m’* spectral expressed as:
moment
(6)
Of (s&f;
Z)
>,
(7)
and where fmti and fmnr are the limits of the useful frequency band. If the centroid fell linearly as a function of depth, the region was accepted as homogeneous (Laugier et al. 1985). If there were nonlinear regions near the first or last spectrum, the ROI was reduced in size to eliminate these spectra and the calculation was repeated. If a region in which the centroid fell linearly with depth for at least five spectra ( 192 sample points, 360 ym) could not be found, the ROI was deemed too heterogeneous for meaningful attenuation measurements. In several cases, more than one ROI was se-
222
Ultrasound in Medicine and Biology
lected from the media of a single segment. The average attenuation as a function of frequency for each media segment was found by averaging the attenuation of each ROI weighted by the number of rf lines in the region. RESULTS Images Hematoxylin-eosin stained histological sections (Figs. 2 and 3) can be compared to the corresponding ultrasonic images of a conventional B-scan (a), AIBS (b) , slope of attenuation (c) and attenuation near band center (d) shown in Figs. 4 and 5. The segment of aorta displayed in the first set of images (Figs. 2 and 4) exhibits intimal thickening, but histological examination revealed no evidence of lipid accumulation, which is a defining feature of atherosclerosis (Star-y et al. 1992). The interface between the intima and media is well shown in the attenuation image. Measurement on the images of the AIBS, attenuation and slope of attenuation were made in the marked regions. The AIBS region consists of 96 sample points in the axial direction (two 64-point Hamming windows with 50% overlap) beginning approximately 100 pm beneath the surface of the artery. This depth was chosen to minimize attenuation effects while still excluding the specular echo at the saline-artery interface from the gated region. The AIBS measured in this rectangular region of the intima was found to be -54 2 2 dB, where 2 dB denotes the standard deviation in the region. No region sufficiently thick and homogeneous could be located in the intima for measurements of the attenuation or slope of attenuation using the methods described in this article. The attenuation at 37.5 MHz and the slope of attenuation between 25 and 56 MHz were measured across three rectangular regions of the image within the media. Results are shown in the images in Fig. 4 and in Table 1, where the errors represent the standard error of the average from the N local values (pixels) in the region. The number of pixels N varies, depending upon the size of the ROI selected.
Volume 23, Number 2, 1997
The smallest of these regions contained eight pixels corresponding to a region 480 pm in depth containing 72 rf lines. A second group of images is shown for a case in which the atherosclerotic lesion consists of fibrous tissue on the left of the image and a dome-shaped lipidic core covered by a thin fibrous cap on the right (Figs. 3 and 5 ) . In this case, the average AIBS was measured 100 pm beneath the surface of the plaque in a 96sample point region yielding -43 t 2 dB. The attenuation and slope of attenuation are measured on the image in a region of the fibrous plaque (ROI 4) and a region of media beneath the lipidic core (ROI 5). Results are shown on the images in Fig. 5 and in Table 1. Regions of interest The original blocks of t-f lines corresponding to the ROIs chosen from the attenuation images were also analyzed as a single volume to provide greater initial spatial averaging. The values measured in this manner for the ROIs in Figs. 4 and 5 are summarized in Table 1, where the errors represent the standard error of the average obtained from the N rf lines in the block. The attenuation as a function of frequency for the ROIs delineated in Fig. 4 is shown in Fig. 6a. It can be seen that for each ROI, the centroid exhibits an overall linear decay as a function of depth as shown by the companion curves in Fig. 6b. Thus, these regions were concluded to be sufficiently homogeneous for accurate measurements. Region homogeneity was confirmed by histology. The average attenuation as a function of frequency was measured from seven segments of normal media. Of these media, two were beneath essentially normal, thin intima, and the other five were shadowed by various thicknesses and types of plaques. All attenuation spectra are shown in Fig. 7a, where the results from the two normal arteries are denoted by the filled diamonds and the filled circles. The slope of the linear fit to each attenuation curve shown in Fig. 7a was found, and the seven resulting values were averaged to yield
Table 1. Slope of the attenuation and the attenuation at 37.5 MHz for the regions of interest marked on the images in Figs. 4 and 5, measured on the parametric images and from the original blocks of rf treated as a whole. Region of interest 1 2 3 4 5
Slope of attenuation (dB cm-’ MHz-‘) (parametric image) 2.6 1.9 2.8 2.8 1.9
+- 0.2 2 0.2 t 0.3 t 0.6 t 0.3
Slope of attenuation (dB cm-’ MHz-‘) (block of rt) 2.6 1.9 2.6 3.6 2.2
+ 0.2 5 0.1 +- 0.2 2 0.5 ? 0.5
Attenuation at 37.5 MHz (dB cm-‘) (parametric image) 73 -c 4 52 2 3 76 2 4 120 2 14 70 2 7
Attenuation at 37.5 MHz (dB cm-‘) (block of rf) 19 52 73 103 67
-t- 2 2 1 -c 2 2 7 k 7
Parametric (integrated backscatter and attenuation) images 0 S. L. BRIDAL et al.
Roil
223
ROI 1
. 2.6 f 0.2 dB/cm MHz
36% 0 2
0; 20 25 30 35 40 45 50 55 60 Frequency (MHz) ROI 2
4 6 8 10 Spectrum Number
12
14
ROI 2
150
40
01
’
’
’
’
’
’
’
4+*
J
”
20 25 30 35 40 45 50 55 60 Frequency (MHz)
ROI 3
2
”
”
”
”
”
4 6 8 10 12 Spectrum Number
14
RO13
150 r -_ ,
I
,
,
,
(
,
,
40
, , , , , , , ( , , , , , ,
39 : .+ 38:
44 4.4
37 : 4 01
’
’
’
’
’
’
n
1
20 25 30 35 40 45 50 55 60 Frequency (MHz)
36t~.‘~“‘~*4~“.1 0
2
4
4 6 8 10 Spectrum Number
12 14
(b) (3 Fig. 6. (a). Attenuation as a function of frequency measured from the blocks of rf lines corresponding to the regions of interest shown in Fig. 4. (b) The centroid as a function of depth in each of the regions of interest shown in Fig. 4.
224
Ultrasound in Medicine and Biology
an estimate of the average slope of attenuation from 25-56 MHz in normal media. The resulting average slope of attenuation is 2.3 t 0.4 dB cm-’ MHz-‘. The average of all results with the standard deviations is plotted in Fig. 7b. The attenuation in segments of normal media of both normal and atherosclerotic human aortae was measured to be 45 + 16 dB/cm at 25 MHz and 102 + 13 dB/cm at 50 MHz, where the errors represent the standard deviations of the average values obtained from the seven specimens.
Volume 23, Number 2, 1997
(a) 150
DISCUSSION The goal of this work was to demonstrate the feasibility of a method for the local estimation of ultrasonic parameters and their display in an easily interpreted image format. This study has brought to light several interesting aspects of these parametric images. At marked interfaces between tissue types, the backscattered signal from the tissue in front of the interface is followed by the strong reflected signal at the interface. Although this leads to a&factual negative values of the attenuation, these negative values appear to offer a useful separation between layers of the artery wall. Attenuation images also seem to be particularly useful for the preliminary selection of homogeneous ROIs. Negative values in the slope of the attenuation images also occur at interfaces, but due to the stochastic nature of the echographic signal and the small amount of spatial averaging obtained from any one measurement volume, the variance of measurements was important and negative slopes were also observed within the relatively homogeneous regions of the tissue. Spatial averaging across homogeneous ROIs on these slope of attenuation images appears to overcome this variability, as shown by comparison of the slope of attenuation values obtained from the images and from blocks of t-f treated as a whole shown in Table 1. An important advantage of parametric images over conventional echographic images is that they provide a system-independent, quantitative view of the tissue or material studied. To assure that parameters are truly quantitative, the measurement techniques used for their estimation must be validated. Tests of the methods used in this work have been reported previously (Roberjot et al. 1996). In this previous work, the multinarrow-band method was applied to estimate the attenuation from the signals backscattered from a calibrated phantom. The attenuation measured as a function of frequency across adjoining bands of frequency using three different broadband transducers demonstrated good continuity. The backscattered spectra were compensated using these measured attenuation values to obtain the backscatter coefficient. Using the
20
I
I
25
30
I
I
I
I
35 40 45 50 Frequency (MHz)
I
55
(b) 150 / l
20
I
I
25
30
a
I
I
I
35 40 45 50 Frequency (MHz)
I
55
Fig. 7. (a) Attenuation as a function of frequency from each of the seven segments of media. (b) Average attenuation as a function of frequency from these seven segments. The dashed lines denote the standard deviations of these average values.
manufacturer-provided properties of the phantom, its backscatter coefficient was calculated (Faran 195 1) from Z-60 MHz, and the calculated values were compared to measured values. Very good agreement between both the frequency dependence and the magnitude of the theoretical and experimental backscatter coefficients was observed from 2-60 MHz. This study showed that the methods used to compensate measurements for the system response and diffraction were effective. Because of its particular importance in high-frequency, tightly focused acoustic backscatter microscopy systems, methods for the compensation of beam
Parametric (integrated backscatter and attenuation) images
diffraction deserve closer examination. The diffraction correction method applied in the current work compensated the backscattered signal for diffraction and the system response using spectra obtained from a stainless steel reflector with several transducer-to-reflector separation distances according to eqn ( 1). This method is a simplification of the reference phantom diffraction correction technique described by Fink and Cardoso ( 1984). The method described by Fink and Cardoso ( 1984) has the disadvantage of increasing data acquisition time, but takes into account the role played by isochronous volumes, which is not accounted for using the steel plate diffraction correction method. Multinarrow-band attenuation estimations in a tissue-mimicking phantom (velocity of sound 1540 m/s) obtained using these two different diffraction correction techniques and estimations obtained with no diffraction correction (compensated only for the system response using a reference spectrum from the echo of a stainless steel plate at the focus) are shown in Fig. 8. All measurements were made from the same volume of a phantom placed at different distances from the transducer, such that the axial distance between the transducer and the scattering region ranged from 10.9 - 12 mm (panel a), 12-13.1 mm (panel b) and 12.4-13.5 mm (panel c), at 1.5 mm/+ The nominal focal length of this transducer is 12.7 mm. The attenuation measured using the reference phantom technique (filled circles ) is relatively independent of the field position of the measurement. Between 12 and 13.5 mm (panels b and c), the deviations between the attenuation estimations obtained using the compensation for the diffraction with the steel plate (open circles) and those compensated with a reference phantom (filled circles) were at most 4 dB/cm and were less than the differences observed between measurements obtained without diffraction correction (squares) and using the reference phantom technique (filled circles). The deviations between the steel plate diffraction correction (open circles) and the reference phantom diffraction correction (filled circles) are significantly greater in the region of the field from 10.9-12 mm (panel a). All spectra used to construct parametric images were compensated using steel plate diffraction spectra. The resulting images provide a useful view of the shape and size of the plaques, but ROIs for quantitative measurements should be selected only within an approximately 1.5-mm band toward the axial center of the images, which corresponds to the region of the field from 12- 13.5 mm, where the steel plate method for diffraction correction has been found to agree with that obtained using a reference phantom. To apply the measurement methods tested in tissue-mimicking phantoms in biological tissue, it was necessary to estimate the value of the speed of sound.
0 S. L. BRIDAL
et al.
225
(a)
20
3 50 b requen:; (MHz)
60
lo-,..“‘.,.‘...“.‘.’ 20 30 50 Frequen:; (MHz)
60
lo~~‘..‘~~*“.~*~‘~‘~~ 20 3 50 p:requen$ (MHz)
60
(W
4
(cl
Fig. 8. Attenuation in a phantom measured using spectra: compensated only for the system response using a spectrum obtained from a plane reflector at the focus (squares) ; compensated for the system responseand for the diffraction using diffraction correction spectra obtained from a plane reflector according to eqn ( 1) (open circles); and compensated for the system response and for diffraction using a reference phantom technique (filled circles). The nominal focus of the transducer was 12.7 mm (time of flight of 17 /LSat 1.5 mm/ ps). All measurements were made from the same volume of a phantom placed at different distances from the transducer, such that the axial distance between the transducer and the scattering region ranged from 10.9-12 mm (a), 1213.1 mm (b) and 12.4-13.5 mm (c). at 1.5 mmlps.
226
Ultrasound in Medicine and Biology
To construct parametric images, a speed of sound of 1500 m/s was assumed throughout the arterial wall. Because differences between this assumed value and the actual value would introduce error into the calculation of distances, it is of interest to estimate what order of error may be anticipated. Data (Rooney et al. 1982) reporting a correlation between ultrasonic velocity and the content of calcification, fat and collagen in the artery wall suggest that, excluding calcified regions, the velocity may vary due to differences in the composition of arterial tissue by approximately 10% at most. Another possible source for the variation of the ultrasonic velocity in the specimen is the variation in the ambient temperature (21”C-27°C) between different data runs. Although systematic studies of the velocity of sound in arterial tissue as a function of temperature have not been reported, velocities measured at 37°C (Lockwood et al. 1991) across regions of artery wall excluding significant plaque are approximately 6% higher than those reported at 20°C by other researchers (Greenleaf et al. 1974). The effect of the smaller change in ambient temperature (21”C-27°C) between the measurements of data reported here would be more modest. Thus, a 10% maximum variation in the speed of sound from the assumed value of 1500 m/s is anticipated in the types of plaques (noncalcified) studied in this work. It should also be noted that the diffraction correction method applied in this work assumed that the speeds of sound were the same in the reference medium (in this case, water) and the measured medium (water path followed by arterial specimen). A 10% difference in the velocity in the specimen from the assumed value ( 1500 m/s) could lead to an ambiguity in position of about 100 ,um for a region thought to be 1 mm below the surface of the specimen. This error in the estimation of the position of the backscattered spectra would result in a mismatch between backscattered spectra and diffraction correction spectra. For the particular transducer used in this work, such an offset between diffraction correction spectra is negligible and limited by the small propagation distances within the artery wall. AIBS measurements in a region of fibrotic plaque (dense collagen) and thickened intima (loose collagen) obtained at equal depths beneath the artery surface were -43 2 2 dB and -54 + 2 dB, respectively. Although no conclusions can be made based on these isolated measurements, reported results (De Kroon et al. 1991; Shepard et al. 1992; Wickline et al. 1994) have shown that dense fibrous tissue scatters considerably more strongly than early intimal hyperplasia. The AIBS parameter may be useful for plaque discrimination, but comparison of values obtained using different gate lengths or depths in the sample cannot be made,
Volume 23, Number 2, 1997
nor can this parameter be used to identify regions of tissue beneath tissue for which the attenuating properties are unknown. Wear et al. (1995a) have demonstrated the feasibility of the compensation of backscatter for the attenuation in layers of overlying tissues in vivo. They compensated the echographic signal obtained in vivo from kidney and liver for the attenuation in the intervening tissues (skin, fat, muscle) using average tissue layer thicknesses and attenuation coefficient values obtained from separate measurements or from the literature. Eventually, AIBS could be compensated for overlying attenuation to obtain the backscatter coefficient (Hartman et al. 1991; Lizzi et al. 1988; Roberjot et al. 1996) by using local attenuation estimations such as those reported here, a model describing the scattering structure of the tissue and an algorithm that detects strong heterogeneities (such as interfaces) preventing the compensation by artifactual values measured at interfaces. This parameter would provide an additional, system- and depth-independent, index of material composition from which information related to the number density, scattering strength and size of the scatterers in the material could be extracted. The attenuation measurement method was tested in segments of media beneath normal intima and atherosclerotic plaque, and results from seven such segments are shown in Fig. 7. Although the medial tissue beneath an atherosclerotic plaque can be altered in a severe case and after a long-standing history of the disease, atherosclerosis primarily involves the intima (Stary et al. 1994). The normal state of medial segments and its homogeneity beneath both atherosclerotic plaque and normal thin intima was verified by histology. The homogeneity of these regions was also tested ultrasonically based upon the linear decay of the centroid, with depth as demonstrated in Fig. 6b. Measurement of the attenuation across normal carotid and femoral arteries obtained using a substitution (transducer-reflector) method has been reported by Lockwood et al. (1991) to be 40 2 12 dB/cm at 30 MHz and 90 t 28 dB /cm at 50 MHz (standard deviations were estimated from the curve in Fig. 11 of this reference). The measurements reported by Lockwood et al. ( 1991) and the measurements reported in this article, which were obtained in backscatter mode, are compared in Table 2. Comparison of the values in this table reveals that they are within 1 SD of each other throughout the common bandwidth, but that the values reported by Lockwood et al. ( 1991) are systematically lower than the present values by approximately 2%. Several potentially important differences between the two studies should be noted. First, it should be kept in mind that temperature has been shown to have a significant effect on ultrasonic attenuation, and that
Parametric (integrated backscatter and attenuation) images 0 S. L. BRIDAL et al.
Table 2. Average attenuation and its standard deviation in normal arterial media as a function of frequency. Frequency (MHz)
Attenuation multinarrow-band method (dB cm-‘)
25 31 38 4.4 50 56
45 rc_16 54% 13 73, 14 89 t 10 102 f 13 115 i 15
Attenuation substitution method+ (dB cm-‘) 34 43 59 76 90 102
” 14 -c 14 r 18 c 22 2 28 ” 31
The center column displays the values reported in this work measured in the media of human thoracic aorta in vitro at 24°C. These measurements were obtained from the signal backscattered from the internal structure of the media by applying a multinarrow-band signal analysis technique. The third column displays the values reported by Lockwood et al. (1991) measured in vitro across normal regions of carotid and femoral arteries at 37°C. The measurements of Lockwood et al. (1991) were obtained using a substitution (transducerreflector) method. ’ Values of the attenuation in arterial media reported by Lockwood et al. (1991) were estimated from the curve in Fig. 11 of this reference.
the values reported in this work at 24°C _+ 3°C may not be directly comparable with measurements such as those of Lockwood et al. (1991) obtained at 37°C. Second, for the work reported here, measurements were performed in elastic aortae, whereas the measurements of Lockwood et al. ( 1991) were conducted in femoral arteries (muscular) and carotid arteries (elastic). Lockwood et al. ( 1991) reported that intersample variation was large enough that statistically significant differences could not be found between femoral and carotid samples. The comparison of the results reported in this article with those reported by Lockwood et al. ( 199 1) appears to demonstrate another instance in which the attenuation does not vary significantly as a function of medial type (muscular vs. elastic). It is interesting to note that important differences in the AIBS measured from the media of muscular and elastic arteries have been reported in both humans and animals (De Kroon et al. 1991; Lockwood et al. 1991; Wickline et al. 1994). In light of this, the lack of a significant difference between the attenuation in these structures seems to warrant further investigation and may eventually provide insight into the relative importance of scattering and absorption attenuation mechanisms in arterial tissues. Because the multinarrow-band method is based on differences between consecutive spectra, measurements of both the attenuation and the slope of attenuation acquired with this technique should be comparable, regardless of region depth or differences in tissue intervening between the transducer and the region. No significant difference was observed between the attenuation measured in regions of media
227
beneath layers of atherosclerotic plaque and beneath normal intima (Fig. 7a), which is in agreement with the histological finding that all these regions consisted of normal media. The potential of slope of attenuation and attenuation measurements performed with the multinarrow-band method for the characterization of material in underlying layers should make it particularly useful for the characterization of the artery wall, but the resolution of these measurements requires improvement if accurate measurements in small regions or thin layers of arteries are to be achieved. Possible strategies for improving the axial resolution in the future could include the sacrifice of circumferential and longitudinal resolution to obtain more stable estimates of the average spectrum at a given depth of data. This will only be successful if the artery is homogeneous across this lateral volume of data. The introduction of higher frequency transducers could allow higher resolution measurements within the uppermost layers of the artery, at the expense of depth of exploration. Autoregressive modeling may also potentially aid in resolution enhancement by permitting shorter time segments than fast-Fourier transform methods (Baldeweck et al. 1995; Wear et al. 1995b). One possible source of bias in local parameter estimation can arise from the convolution of the scattered signal with a short time gate. This convolution will reduce the spectral resolution to approximately the width of the gate in the frequency domain, estimated as l/7, where 7 is the length of the gate in the time domain. Thus, the 64-sample gate digitized at 400 MHz used in this work results in a spectral smoothing of approximately 6 MHz. The bandwidth of measurements was strongly limited to reduce the potential for the introduction of noise outside the bandwidth due to this spectral smoothing. Because the attenuation acts as a low-pass filter, the bandwidth was limited more strongly on the high-frequency end, leading to a bandwidth corresponding to the -17 and -4 dB down points of the reference spectra from a stainless steel plate at the focus. These limits correspond to 25 and 56 MHz, respectively. A second concern was the small amount of spatial averaging in any one measurement volume (64-sample segments of 18 rf lines). To validate the method of measurement using these small, local volumes, results from ROIs on the parametric images were compared with measurements from the corresponding block of rf data treated as a whole. The values measured using the two methods are not expected to be identical. The measurement from an ROI on the parametric image includes many smaller measurement volumes with significant axial overlap; thus, the scattered signal at the axial center of the ROI is included in several measurement volumes while the
228
Ultrasound in Medicine and Biology
scattered signal at the axial beginning or end of the ROI is included in fewer of the volumes. Measurements from the block of the rf treated as a whole give equal weight to each axial segment of the rf line. This difference in spatial averaging may explain why estimates made using the two methods agreed less well for ROI 4, which consisted of a dense fibrous plaque. Histologically, ROI 4 consisted of bands of dense collagen mixed with areas of looser collagen. This mixture of different collagen patterns results from the evolution of the plaque. Overall, results obtained using the two approaches agreed well, indicating that the small number of lines of rf used to calculate parameters for the image do not distort the final results as long as sufficient spatial averaging is obtained across the image and heterogeneous regions are avoided. To fully establish the sensitivity of these parameters to plaque composition requires the examination of more specimens with plaque extent great enough (at least 360 pm in depth using the methods described in this article) that homogeneous measurement sites can be found. Once methods have been clearly established in vitro, further investigations will be required to test their usefulness in viva . In certain respects, intravascular ultrasound offers a novel opportunity for quantitative tissue characterization, since the only intervening material between the ultrasonic probe and the tissue is the circulating blood, for which the attenuating properties have been measured between 30 and 60 MHz for a variety of hematocrits and normal flow (Lockwood et al. 1991). However, the field of potential applications for these methods extends well beyond arterial characterization. Because most organs are accessible in echographic mode, these techniques may be adapted for many cases of clinical interest. The representation of ultrasonic parameters in images facilitates their interpretation, in particular, in the evaluation of strongly heterogeneous structures. Thus, quantitative ultrasonic imaging could add a new, more objective view to ultrasonic imaging. REFERENCES Baldeweck T, Laugier P, Herment A, Berger G. Application of autoregressive spectral analysis for ultrasound attenuation estimation: Interest in highly attenuating medium. IEEE Tram UFFC 1995;42:99-110. Bamber JC. Theoretical modelling of the acoustic scattering structure of human liver. Acoustics Lett 1979;3:114-119. Barzilai B, Saffitz JE, Miller JG, Sobel BE. Quantitative ultrasonic characterization of the nature of atherosclerotic plaques in human aorta. Circ Res 1987;60:459-463. Berger G, Laugier P, Fink M, Perrin J. Optimal precision in ultrasound attenuation estimation and application to the detection of duchenne muscular dystrophy carriers. Ultrason Imaging 1987; 9:1-17. Davies MJ, Thomas AC. Plaque fissuring-The cause of acute myocardial infarction, sudden ischaemic death, and crescendo angina. Br Heart J 1985;53:363-373.
Volume 23, Number 2, 1997 De Kroon MGM, van der Wal LF, Gussenhoven WJ, Rijsterborgh H, Born N. Backscatter directivity and integrated backscatter power of arterial tissue. Int J Card Imaging 1991;6:265-275. Faran JJ Jr. Sound scattering by solid cylinders and spheres. J Acoust Sot Am 1951;23:405-418. Fink M, Cardoso JF. Diffraction effects in pulse echo measurement. IEEE Tram Sonics Ultrason 1984;SU-31:313-329. Garra BS, Insana MF, Sesterhenn IA, et al. Quantitative ultrasonic detection of parenchymal structural changes in diffuse renal disease. Invest Radio1 1994;29:134-143. Greenleaf JF. Duck FA. Samavoa < WF. Johnson SA. Ultrasonic data acquisition and processing system for atherosclerotic tissue characterization. IEEE Ultrason Symp 1974; 74CH0896-ISU: 738-743. Gussenhoven ET, Essed CE, Lancee CT, et al. Arterial wall characterization determined by intravascular ultrasound imaging: An in vitro study. J Am Co11 Card 1989; 14:947-952. Hartman PC, Oosterveld BJ, Thijssen JM, Rosenbusch GJE. Variability of quantitative echographic parameters of the liver: Intraand interindividual spread, temporal- and age-related effects. Ultrasound Med Biol 1991; 171857-867. Insana MF, Hall TJ. Parametric ultrasound imaging from backscatter coefficient measurements: Image formation and interpretation. Ultrason Imaging 1990;12:245-267. Insana MF, Hall TJ, Fishback JL. Identifying acoustic scattering sources in normal renal parenchyma from the anisotropy in acoustic vroverties. Ultrasound Med Biol 1991: 17:613-626. Laugier P, Berger G, Fink M, Perrin J. Specular reflector noise: Effect and correction for in vlvo attenuation estimation. Ultrason Imaging 1985;7:277-292. Lee RT, Richardson SG, Loree HM, et al. Prediction of mechanical properties of human atherosclerotic tissue by high-frequency intravascular ultrasound imaging: An in vitro study. Arteriosclerosis Thrombosis 1992; 12:1-5. Lizzi FL, King DL, Rorke MC, et al. Comparison of theoretical scattering results and ultrasonic data from clinical liver examinations. Ultrasound Med Biol 1988; 14:377-385. Lockwood GR, Ryan LK, Hunt JW, Foster FS. Measurement of the ultrasonic properties of vascular tissues and blood from 35-65 MHz. Ultrasound Med Biol 1991; 17:653-666. Nassiri DK, Hill CR. The differential and total bulk acoustic scattering cross sections of some human and animal tissues. J Acoust Sot Am 1986a;79:2034-2047. Nassiri DK, Hill CR. The use of angular acoustic scattering measurements to estimate structural parameters of human and animal tissues. J Acoust Sot Am 1986b;79:2048-2054. O’Donnell M, Mimbs JW, Miller JG. The relationship between collagen and ultrasonic backscatter in myocardial tissue. J Acoust Sot Am 1981;69:580-588. Glerud JE, O’Brien WD Jr, Riederer-Henderson MA, et al. Correlation of tissue constituents with the acoustic properties of skin and wound. Ultrasound Med Biol 1990; 1655-64. Picano E, Landini L, Distante A, et al. Different degrees of atherosclerosis detected by backscattered ultrasound: An in vitro study on fixed human aortic walls. J Clin Ultrasound 1983; 11:375379. Picano E, Landini L, Distante A, et al. Fibrosis, lipids, and calcium in human atherosclerotic plaque: In vitro differentiation from normal aortic walls by ultrasonic attenuation. Circ Res 1985; 56:556-562. Roberjot V, Bridal SL, Laugier P, Berger G. Absolute backscatter coefficient over a wide range of frequencies in a tissue-mimicking phantom containing two populations of scatterers. IEEE Trans UFFC 1996;43:970-978. Rooney JA, Gammell PM, Hestenes JD, Chin HP, Blankenhorn DH. Velocity and attenuation of sound in arterial tissues. J Acoust Sot Am 1982;71:462-466. Shepard RK, Miller JG, Wickline SA. Quantification of atherosclerotic plaque composition in cholesterol-fed rabbits with 50-MHz acoustic microscopy. Arteriosclerosis Thrombosis 1992; 12: 1227-1234. Silverman RH, Rondeau MJ, Lizzi FL, Coleman DJ. Three-dimen-
Parametric (integrated backscatter and attenuation) images 0 S. L. BRIDAL er al. sional high-frequency ultrasonic parameter imaging of anterior segment pathology. Ophthalmology 1995; 102:837-843. Stary HC, Blankenhom DH, Chandler AB, et al. A definition of the intima of human arteries and of its atherosclerosis-prone regions. Circulation 1992; 85:391-405. Stary HC, Chandler AB, Glagov S, et al. A definition of initial fatty streak and intermediate lesions of atherosclerosis. Arteriosclerosis Thrombosis 1994; 14:840-856. Urbani MP, Picano E, Giuliano P, et al. In vivo radiofrequency-based ultrasonic tissue characterization of the atherosclerotic plaque. Stroke 1993;24:1507-1512. Wear KA, Milunski MR, Wickline SA, et al. Differentiation between acutely ischemic myocardium and zones of completed infarction in dogs on the basis of frequency dependent backscatter. J Acoust Sot Am 1989;85:2634-2641. Wear KA, Garra BS, Hall TJ. Measurements of ultrasonic backscat-
229
ter coefficients in human liver and kidney in viva. J Acoust Sot Am 1995a;98:1852-1857. Wear KA, Wagner RF, Garra BS. A comparision of autoregressive spectral estimation algorithims and order determination methods in ultrasonic tissue characterization. IEEE Trans UPFC 1995b; 42:709-716. Wickhne SA, Shepard RK, Daugherty A. Quantitative ultrasonic characterization of lesion composition and evolution in atherosclerotic rabbit aorta. Arteriosclerosis Thrombosis 1993; 13:1543-1550. Wickline SA, Miller JG, Recchia D, et al. Beyond intravascular imaging: Quantitative ultrasonic tissue characterization of vascular pathology. JEEE Ultrason Symp 1994;3:1589-1597. Wilson LS, Neale ML, Talhami HE, Appleberg M. Preliminary rcsults from attenuation-slope mapping of plaque using intravascular ultrasound. Ultrasound Med Biol 1994; 20:529-542.