Annals of Biomedical Engineering, Vol. 41, No. 5, May 2013 (Ó 2013) pp. 939–951 DOI: 10.1007/s10439-012-0737-7

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia SLEIMAN R. GHORAYEB1,2 and DAVID M. ROONEY1 1

School of Engineering and Applied Sciences, Ultrasound Research Laboratory, Hofstra University, Hempstead, NY 11549, USA; and 2Orthopaedics Research Lab – FIMR, North Shore Hospital, Manhasset, NY, USA (Received 13 September 2012; accepted 26 December 2012; published online 9 January 2013) Associate Editor Mona Kamal Marei oversaw the review of this article.

INTRODUCTION

Abstract—Several imaging modalities have traditionally been utilized to assess bone health. However, none of these standards is capable of providing a clear rendition or display of the damaged bone layers caused, for instance, by osteoporosis. This study examines the use of ultrasound for non-invasive monitoring of bone quality in bone samples with various degrees of porosity. A user-defined region of interest (ROI) in the iliac portion of extracted human cadaver coxal bones is monitored and quantified. Raster Cscan images of the ROI were acquired and compared to basic physical measurements, and to bone scans using dual energy X-ray absorptiometry (DXA). A quantitative measure of the superficial sub-surface composite matrix (ScM) content was analyzed using linear regression with all physical and DXA measures. The trend in the degree of percent bone loss (PBL) measured by ultrasound (US) was found to be closely paralleled with that measured by DXA (R2 = 0.82, p < .0005). Also, the trend in which PBL (US) correlated with bone mineral density (BMD) (R2 = 0.62, p < .01) was found to exhibit a similar behavior when the latter was compared to dry mass density (DmD) of the bone samples (R2 = 0.63, p < .01). However, when PBL (DXA) was compared to DmD, it did reveal a better linearity (R2 = 0.69, p < .005) than the one obtained when PBL (US) was compared with the same DmD (R2 = 0.45, p < .05). A similar outcome was observed when PBL (US) was compared with percent porosity (R2 = 0.51, p < .05), as opposed to the better linearity exhibited between PBL (DXA) and porosity (R2 = 0.86, p < .0005). Despite these slight variations, further analyses on the statistical significance between these correlations suggest that ultrasound can be an effective imaging technique in assessing the degree of bone damage, and can be used to assess the structural integrity of bones.

Osteoporosis is one of the most prominent causes of loss of bone mass. Unfortunately, this disease can develop without any symptoms and can cause serious health problems or injuries if left undiscovered. Osteoporosis weakens the structural integrity of the trabecular bone while making the cortical bone more porous, which makes the bone more vulnerable to fractures. Researchers estimate that about 1 out of 5 American women over the age of 50 have osteoporosis. About half of all women over the age of 50 and men over the age of 70 have a higher risk for osteoporosis and will suffer a fracture of the hip, wrist, or vertebra.11 Therefore, understanding the underlying mechanisms of bone failure is crucial to early detection and prevention of this silent disease. Currently, there are several methods used to detect and determine the level of bone loss in many areas of the body, including the hip, spine, and heel. However, most of these methods require ionizing radiation, which may be harmful to the body. Some of the methods include single energy X-ray absorptiometry (SXA),24 dual energy X-ray absorptiometry (DXA),24 synchrotron radiation,4 and peripheral quantitative computed tomography (pQCT).28,8 Diagnostic ultrasound for osteoporosis detection and evaluation is well documented.7,9,13,15–17,19–23,29–32,34 Unlike its X-ray based counterpart tools, ultrasoundbased techniques offer clear advantages due to their non-ionizing nature, low cost, portability, and ease of use. Ultrasonic characterization of backscatter from human cancellous bone with Renyi entropy was shown to have a correlation with X-ray BMD in the study done by Wallace et al.29 This work assessed material properties of cancellous bone as compared with pQCT. Ultrasonic C-scan imaging is a two dimensional presentation of data displayed as a top or planar view of a test piece, similar in its graphic perspective to an X-ray

Keywords—Noninvasive ultrasound, Bone loss, Bone quality, Bone mineral density, Osteoporosis.

Address correspondence to Sleiman R. Ghorayeb, School of Engineering and Applied Sciences, Ultrasound Research Laboratory, Hofstra University, Hempstead, NY 11549, USA. Electronic mail: [email protected]

939 0090-6964/13/0500-0939/0

Ó 2013 Biomedical Engineering Society

940

S. R. GHORAYEB

image, where color represents the gated signal amplitude or depth at each point in the test piece mapped to its position. Planar images can be generated on flat parts by tracking data to X–Y position, or on cylindrical parts by tracking axial and angular position (Olympus-Panametrics, Waltham, MA). Petculescu et al.23 presented an analysis of C-scan images in amplitude and time of flight (TOF) using a pulse echo setup to characterize the microstructure of human bone. Three samples (a 2000 year old, a young, and a medium age, but with no mention of the exact age or type of bones used) were interrogated by pulse–echo technique, in immersion. Image analysis in amplitude and in TOF led to distribution maps of bone density for the three samples. However, information reported about the health of these samples was very limited. Ultrasonic testing revealed larger inhomogeneity and lower density in the old bone when compared to the young and medium age ones, but was determined to be non-predictive of physical alterations with respect to bone classification and abnormalities. No correlation studies between trabecular matrix echostructure measured with ultrasound and other modalities were reported. Quantitative ultrasonic assessment of osteoporosis from the tibia shaft was studied by Chen et al.7 In this paper, a dual-transducer ultrasound technique was employed to measure the mean ultrasound Speed of Sound (SOS) in the cortical layer as well as the cancellous layer in the tibia shaft. Encouraging results from 18 outpatients showed a high correlation between measurements of BMD and those from DXA. Lochmu¨ller et al.19 compared ultrasonic measurements in the human calcaneus using mechanical failure loads. Using a commercial transmission scanner with two (25 mm) piezoelectric transducers with fixed separation, a correlation between the calcaneal ultrasonic speed of sound and experimentally determined failure loads of the proximal femur and lumbar spine were measured. Ultrasonic backscatter and its relationship to BMD measurements in human calcaneus were also investigated by Wear and Armstrong.32 They showed that backscattering is a useful way of detecting bone fractures caused by osteoporosis. Laugier16 reviewed the current state of development in quantitative ultrasound (QUS) measurements to assess skeletal status and outlined a few promising developments. He highlighted that QUS techniques have, for a while, focused on acquiring measurements at easily accessible peripheral sites—i.e., heel, finger, wrist or tibia—where the impact of a thin layer of surrounding soft tissue is less of an issue. However, the skeleton is heterogeneous and the pattern of bone loss may vary at different sites, depending on age or on the underlying disorder. In addition, several bone properties,5 including bone mineral density,

AND

D. M. ROONEY

micro-architecture and tissue elasticity contribute independently to bone strength. Therefore, a site-specific or a parameter-specific measuring technique only provides a partial view of skeletal status. As a result, the paper reviewed considerations that guided investigations towards innovative multisite (i.e., combining ultrasonic parameters obtained at multiple skeletal sites), or multiparametric (i.e., combining different ultrasonic parameters reflecting independent bone properties) QUS techniques. Lasaygues et al.15 presented an implementation scheme for bone ultrasonic imaging by Ultrasound Tomography (UT). This work focused on the issue of strong contrast exhibited by bones. For this kind of medium, ultrasound reflection tomography (URT) associated with signal processing can be used to characterize the shape and size of bone, while ultrasound transmission tomography (UTT), with compensation procedure, provides quantitative values of the sound velocity. They suggested that both URT and UTT may be combined to perform the correction method by acquisition lines. In order to substantiate what has already been reported in this field as ‘‘Gold Standard’’, the goal of this study is to present a new ultrasound parameter, percent bone loss (PBL) in the superficial sub-surface composite matrix (ScM) content, and evaluate its performance in the human cadaver iliac bone. We propose a method that measures image brightness, and then quantify the correlation of the latter to the overall content of the matrix. Basic physical measurements of dry mass density of bone samples and bone mineral density generated by DXA are used to validate the viability of the proposed method. Given that the main issue at hand is to correlate image brightness to ScM, one should note that the brightness is mainly produced by the amount of bone present in the ROI. Cardoso et al.6 performed experimental tests on trabecular canine bone samples that were either immobilized (disuse-induced osteoporosis), or immobilized and treated with antiresorptive therapy (biphosphonates), and compared with non-immobilized control bones. The goal of the study was to integrate a poroelastic media approach into the analysis of ultrasound wave propagation in porous anisotropic bone, in order to distinguish the relative contribution of the solid bone fraction and anisotropy from the acoustic wave measurements during bone loss. Therefore, as a confirmation of Biot’s theory of wave propagation in porous media and to describe the dynamics of a poroelastic solid saturated by a viscous fluid, their study showed that ultrasonic wave displacements generated in the solid (bone) were different from those in the fluid (pores). As a result, the immobilized bones showed significant losses in

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia

mechanical properties, causing dramatic changes in the (an)isotropy elastic coefficient pattern. This effect provoked subsequent quasi-edema of the bones, thus disrupting the anisotropic microarchitechture structure. More recently, a new study35 adopted a similar approach to propose a new simple imaging technique based on the two-wave phenomenon for investigating the anisotropic structure of cancellous bone. A cylindrical specimen of cancellous bone obtained from a bovine femur was used in the study. A conventional ultrasonic receiver was rotated around the specimen to investigate the ultrasonic fields after propagation within the specimen. The ultrasonic propagation image clearly showed the effect of the bone structure. The fast wave showed apparent refraction, whereas the slow wave did not, confirming the notion that fast-wave propagation imaging is convenient for the interpretation of the anisotropic structure of trabeculae during in vivo measurements. It is important then to frame these phenomena in a hypothesis. That is, as bones undergo certain degradation, the fluid content in the bone mesh increases, resulting in less scatter per volume, and thus less brightness. On the other hand, increased bone density becomes prominent with increased healing, and thus more brightness in the scan. For this reason, a palette was selected that would make the difference between bone and non-bone zones the clearest. This palette translated the original areas nearing either end of the spectrum to white, representing the densest areas of bone, and those in the middle of the original spectrum to black, representing the fluidic content and other proteins in the matrix structure, with a grayscale gradient separating them depending on the acoustic impedances of those elements.

941

established for correlation purposes when ultrasonic and DXA results are compared later in the study. Each sample was weighed using a precision electronic scale. The length and width of each bone was also recorded, where they were measured from the uppermost iliac crest (A in Fig. 1) to the lowermost pubis articular surface (B in Fig. 1), and from the anterior superior iliac spine (C in Fig. 1) to the posterior inferior iliac spine (D in Fig. 1), respectively. Note that the numbering of the samples was simply random with no significance to the study at hand. The dry volume of each of the bones was obtained by degassing and shrink-wrapping each sample in a very thin, high quality adhesive plastic wrap and measuring the volumetric displacement as it was submerged underwater in a scaled beaker. Thus, the dry volume represents the total volume of the bone including the pores. In order to mimic a real situation in a ‘‘non-cadaveric’’ sample, the saturated volume was determined by measuring the volumetric displacement after removing the shrink-wraps and immersing each sample underwater (phosphate buffered saline solution) for 24 h, to ensure that no air was entrapped within the sample. Thus, the saturated volume represents the volume of bone in the sample. For these volumetric and mass measurements, the dry (DmD) and saturated (SmD) densities (Table 2) for each sample could be calculated using the following: DmD = Dry Mass/Dry Volume, and SmD = (Dry

MATERIALS AND METHODS Basic Physical Measurements Coxal bones were used for this project because they contain both cortical and trabecular matrices, and are anatomically close to the surface of the skin. Ten randomly chosen human samples were used in this study. All of the specimens belong to female cadavers, but age and race were unknown. The iliac crest portion of the bone was used for this research because of its anatomical shape (i.e., flatness), which makes it ideal from the standpoint of ultrasonic wave transmission. The region of interest (ROI) that was ultrasonically interrogated in this study is shown in Fig. 1. More discussion regarding the choice of the ROI will be given later. First, several basic parameters were measured (as shown in Table 1) so that a physical baseline is

FIGURE 1. Bone sample with superimposed ROI showing the Iliac Crest (A), the pubis articular surface (B), the anterior superior iliac spine (C), and the posterior inferior iliac spine (D).

942

S. R. GHORAYEB

AND

D. M. ROONEY

TABLE 1. Ilium sample measurements recorded at room temperature. Sample # 10 11 12 15 44 48 49 52 54 63

Dry mass (g)

Length (cm)

Width (cm)

Dry volume (cm3)

Saturated volume (cm3)

173.68 89.91 134.64 71.83 90.10 66.58 131.43 90.45 84.01 145.16

20.16 20.32 22.54 17.30 18.26 17.78 19.05 17.15 20.64 17.94

14.61 14.61 15.56 12.86 12.70 12.38 13.97 13.02 14.61 13.33

260 260 375 210 250 200 290 210 280 270

175 130 195 100 125 100 180 125 110 175

TABLE 2. Measured ilium densities. Sample # 10 11 12 15 44 48 49 52 54 63

TABLE 3. Calculated porosity.

DmD (g/cm3)

SmD (g/cm3)

0.67 0.35 0.36 0.34 0.36 0.33 0.45 0.43 0.30 0.54

0.99 0.85 0.84 0.87 0.86 0.83 0.83 0.84 0.91 0.89

Mass + Density of Water * Volume of Water)/Dry Volume, where Density of Water = 1 g/cm3 and Volume of Water = Dry Volume 2 Saturated Volume. The porosity (=(Dry Volume 2 Saturated Volume)/Dry Volume 9 100%) was also calculated from the measured values for each of the 10 samples (Table 3). Ultrasonic Experimental Set-Up In order to assess the degree of ScM degradation, it was necessary first to determine a method of identifying the solid zones within sonographic images. In examining the ultrasonic propagation properties of the superficial sub-surface mesh structure, it has been observed that the latter presents itself in images as either dark amorphous areas or light areas with quasifibrillar inhomogeneous patterns based on its morphologic characteristics, similar to what is depicted by Petculescu et al.,23 Wear,30 Droin et al.,9 and Jenson et al.13 Although the experiments in these latter studies are very different from the work presented here, insofar as the operating frequency (ranging from 200 kHz to 10 MHz combined, vs. 75 MHz in this study) and size of samples are concerned, they are yet comparable in hypothesis.

Sample # 10 11 12 15 44 48 49 52 54 63

Porosity (%) 32.7 50.0 48.0 52.4 50.0 50.0 37.9 40.5 60.7 35.2

Imaging was performed using a scanning acoustic microscope consisting of an ultrasonic pulser/receiver (Model 5900PR, Olympus-Panametrics, Waltham, MA), an immersion transducer (Model V3320, 75 MHz, 0.5 in. focal length, Olympus-Panametrics, Waltham, MA), and a FlexSCAN-CÒ ultrasonic C-Scan system (Sonix Inc., Springfield, VA) with a tank containing phosphate buffered saline (PBS) solution as the couplant, and computer software system with a front-wall-follower gating capability. The images were created using successive raw RF A-scan signals in space, that are laterally positioned in a raster fashion relative to one another as in a C-mode scan. In the imaging of the bone samples, it is important to note that the bones were scanned with the transducer perpendicular to the bone surface. Also, the reason why such a high operating frequency (75 MHz) was used is twofold: (a) the scanning occurred directly on the bone samples with no intermediate layers of soft tissues, and (b) the area of interest was the sub-surface superficial ScM content of the samples. Since the ultimate goal of this study is to show the feasibility of this technique in non-invasive clinical situations in vivo, in which the measurements are conducted through the skin and other soft tissues, the operating frequency will be lowered to account for propagation in these layers.

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia

The transducer is designed for negative spike excitation. The maximum spike excitation voltage is 300 V with a fast rise time and short duration. Although negative spike excitation is recommended, continuous wave or tone burst excitations may be used. It is helpful to note that the number of cycles per tone burst excitation depends on the duty cycle, which itself depends on several other factors such as impedance, total power, voltage, and phase of the transducer. The spatial length of the pulse for the transducer used in this experiment has been determined to be about 0.2 ls, which is small compared to the echoes obtained from the bone layers (calcanae and trabeculae). The pulser/receiver was set to send an ultrasonic pulse with an energy setting of 32 lJ. In the C-scan imaging mode, the transducer scans across a 12.7 mm 9 12.7 mm (0.5 in 9 0.5 in) region of interest (ROI) as shown in Fig. 1. Each of the samples was placed in the water tank in the focal plane within the transducer’s cone of insonification. The FlexSCAN-CÒ system was used by setting the gate to examine the sub-surface composite matrix (ScM) within the ROI. As an example, Fig. 2 shows typical RF A-mode scans (amplitude vs. time) from two different bone samples. The gated areas in these A-scans, used for calculating the ScM matrix, are clearly indicated in these figures. The first yellow gate (shown across the first largest peak) is referred to as the ‘‘front surface follower’’ (FSF) and indicates the portion of the sample that needs to be excluded from the final C-scan image. The FSF will ‘‘slave’’ the next gate position to the first crossing of the FSF threshold. This ‘‘slave’’ capability keeps the next gate in the proper location, accounting for uneven or tilted surfaces. The second red gate shown is the ‘‘sub-surface follower’’ (SSF) which allows the tracking of the sub-surface matrix (ScM) and that’s what is included in the final image. Furthermore, once the gates are set, the system’s software uses a peak detector that determines the maximum signal strength (absolute peak) of the A-mode signal within each gate at each data point. This peak value is displayed in the final C-scan image. The C-scan images were recorded for each of the 10 samples. DXA Measurements The DXA measurements were made with a Hologic 4500 (Hologic, Bedford, MA). Note that these measurements usually include both bone and surrounding soft tissue during actual clinical assessments. However in this study, the air surrounding the sample was interpreted by the machine as soft tissue. In brief, two ROI’s are defined (Fig. 3): (1) operator specific as shown by the ‘‘L4’’ rectangular box, and (2) a bone dependent sub-ROI specified by the system to produce

943

the most statistically viable results, as shown by the ‘‘zig-zagged’’ tracing of the bone. Note that the ROI where subsequent ultrasonic interrogations were performed was intentionally placed within that latter sub-ROI for an adequate comparison. Each DXA test yielded an array of performance results including bone mineral content (BMC), bone mineral density (BMD), T-score, and percent available bone (PR). It is important to note that the T-score leads to the PR value, and that the latter leads to PBL. In brief, the T-score = (BMD 2 YN)/SD, where YN = young normal BMD reference of lumbar spine in women at age 35 (when bone mass is at its peak and fracture risk is at its lowest), and SD = standard deviation. For example, a T-score of 21 corresponds to a PR value of 90% and therefore a PBL of 10%, a T-score of 24.1 corresponds to a PR value of 59% (PBL = 41%), and so on. The PBL values measured by DXA can then be compared with those obtained by the subsequent ultrasound tests. Image Post-Processing The measurements of the percentage of solid (bone) structure content in the ScM within the scan area in the ultrasound images were obtained using a MatlabTM (MathworksÒ, Natick, MA) algorithm that implements the Floyd-Steinberg error-diffusion dithering technique.1 This technique simply converts an image to grayscale and rounds the intensity of each pixel to its nearest extreme, black (0’s indicating pores) or white (1’s indicating solid bone). It then calculates the percentage of white pixels with respect to the total number of pixels in the ROI. The scan area (ROI) was positioned in the vicinity of the iliac crest, and did not change from sample to sample but was randomly selected (and remained the same across all samples) in order to confirm repeatability of the percentage values in the ScM. As a result, the total number of pixels in the ROI did not vary from one sample to the other. This was determined to be 405,130 pixels (636.5 9 636.5 pixels). The sector profiles obtained from the scans allowed observation of the effects of ultrasonic wave interaction with the composite layers in the samples, as shown for sample #48 in Fig. 4a. The grayscale image contains 256 levels of color across a linear range from black to white representing the C-scan amplitude. Together, these data points constitute a distribution map of bone density regions. The closer the pixel color level is to white, the stronger the bone density is in that area. To analyze the percent bone loss in these images, they must first be converted from grayscale to binary as described above. Each grayscale intensity value was converted to a corresponding density of white pixels,

944

S. R. GHORAYEB

AND

D. M. ROONEY

FIGURE 2. Typical RF A-mode scans (amplitude vs. time) from two different bone samples showing front surface follower (FSF) and sub-surface follower (SSF) gates used to generate the C-scan images of the sub-surface composite matrix (ScM).

pixels to either black or white, the program can easily count a ratio between them, or percentage of bone loss measured in pixels, over the ROI.

RESULTS

FIGURE 3. ROI as measured by DXA.

or bone, and black pixels, or pores as depicted in Fig. 4b. Thus, pores are mapped to black and bone is mapped to white. In gray regions, the area is divided into a percentage of white and black pixels corresponding to that gray level intensity. By converting all

The composite porous appearance of the ScM was clearly observed in the C-scans. The bone samples exhibiting low and high porosity are shown in Fig. 5. These images correspond to the ROI described above. As expected, samples with higher porosity revealed greater hypoechogenicity. The exact locations of these hypoechogenic regions varied with each sample, but as can be seen for example in Figs. 5b, 5d and 5i, the samples exhibited similar features with the increase in the amount of dark regions signifying degradation in the sub-surface matrix. It is important to note however that several ROI’s were placed at various locations in order to confirm the position accuracy and therefore the assessment of ScM content in the appropriate region. A point worth mentioning is that the chosen ROIs are not always representative of the entire bone, and that degeneration of the ScM (if any) may not be uniformly distributed as seen (quad arrows), for example, in Figs. 5c, 5e and 5g. As a confirmation of the hypothesis stated earlier, the degree of hypoechogenicity was less prominent in other samples, indicating a normal amount of calcium

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia

FIGURE 4. (a) Raw image of ilium sample #48 used as input to the MatlabTM program. (b) after dithering.

and phosphate depositions within the ScM of the bone specimens. The dithering program used the binary version of the images such as the one shown in Fig. 4b to calculate the percent bone loss (PBL (US)) as referenced to an image filled only with white pixels, or pure bone. Figure 6 shows an overall bar scale of the results obtained for the percent bone loss for both ultrasound and DXA across all 10 samples, with calculated percent error (as shown by the bars) relative to values obtained by DXA. As seen in this figure, there is a strong correlation between the PBL measured by ultrasound and DXA. As the PBL measured by ultrasound decreases, so does the PBL measured by DXA. Results show an overall absolute average percent error of about 3.5% when

945

compared to DXA. Also, at first glance, a strong correlation exists between PBL (US) and the physical bone parameters measured earlier (Tables 1, 2, and 3). Ultrasonic PBL increases with decreased bone mass, decreased density, and increased porosity. This is reflected in the strong correlation between the two methods as shown in Fig. 6. In order to determine the relationship between the ScM concentration and all other DXA and physical measures, regression analysis was used. The coefficient of determination, R2, and the p value for the significance of the regression models were calculated. In general, almost all relationships between the superficial sub-surface composite matrix (ScM) content and all physical and DXA measures were found to be statistically significant (p < .05). Expressing Fig. 6 differently using a scatter plot of the two PBL factors (ultrasound and DXA) with respect to each other yielded a linear regression with (R2ud = 0.82, p < .0005), indicating a significant correlation, as can be seen in Fig. 7a. However, when PBL (DXA) was compared with percent porosity (Fig. 7b), it did reveal a better linearity (R2dpor = 0.86, p < .0005) than the one obtained when PBL (US) was compared with the same percent porosity (R2upor = 0.51, p < .05) (Fig. 7c). A similar outcome was observed when PBL (US) was compared to DmD (R2udmd = 0.45, p < .05), as opposed to the better linearity exhibited when PBL (DXA) was correlated with the same DmD (R2ddmd = 0.69, p < .005). Table 4 shows a summary of R2 and p values of all other correlations between physical and measured ultrasound and DXA properties. Note that it is entirely possible for two regression lines to have similar slopes (same regression) but the data to be tightly clustered around one regression line (high correlation) and significantly less tightly clustered around the second line (different correlations). In these situations, it is desirable to be able to make statistical comparisons between correlation coefficients measured on the same common sample.27 Using the Fisher10,27 r-to-z transform, further calculations were made based on the statistical significance of the difference between each set of two correlation coefficients; namely, the correlations between PBL (US) and porosity (R2upor) and PBL (DXA) and porosity (R2dpor), and the correlations between PBL (US) and DmD (R2udmd) and PBL (DXA) and DmD (R2ddmd). Noting that, in this case, when the p value is less than 0.05, the conclusion is that the two coefficients are significantly different. In our study, when R2upor (0.51, n = 10) was compared with R2dpor (0.86, n = 10), the resulting z-statistic was –1.36, which is associated with a p value of 0.174. Since this p value is greater than 0.05, it is concluded that the two correlation coefficients

946

S. R. GHORAYEB

AND

D. M. ROONEY

FIGURE 5. Ultrasonic images within the ROI for ilium (a) #10, (b) #11, (c) #12, (d) #15, (e) #44, (f) #48, (g) #49, (h) #52, (i) #54, and (j) #63, respectively, clearly showing the superficial sub-surface structure.

(R2upor and R2dpor) are statistically comparable. On the other hand, when R2udmd (0.45, n = 10) was compared with R2ddmd (0.69, n = 10), the resulting z-statistic was 20.67, which is associated with a p value of 0.503. Again, since this p value is greater than 0.05, it is

further concluded that the latter two correlation coefficients (R2udmd and R2ddmd) are also statistically comparable. This confirms the initial strong correlation (R2ud = 0.82) exhibited between the two methods (ultrasound and DXA).

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia

947

FIGURE 5. continued.

70

Percent Bone Loss

60 50 40

US

30

DXA 20 10 0

10

11

12

15

44

48

49

52

54

63

SAMPLE #

FIGURE 6. Plot of percent bone loss values for each of the 10 samples for both ultrasound and DXA methods showing relative error between the two methods.

948

S. R. GHORAYEB

AND

D. M. ROONEY

PBL (US) vs. PBL (DXA)

(a) 65

y = 0.9474x + 2.5498 R² = 0.8154

60

PBL (US)

55 50 45 40 35 30 40

45

50

55

60

PBL (DXA)

PBL (US) vs. Porosity

Porosity (%)

(b) 70.0

y = 1.1573x - 11.731 R² = 0.5064

65.0 60.0 55.0 50.0 45.0 40.0 35.0 30.0 25.0 20.0 40

45

50

55

60

PBL (US)

y = 1.4355x - 25.459 R² = 0.8576

PBL (DXA) vs. Porosity

(c) 65.0 60.0

Porosity (%)

55.0 50.0 45.0 40.0 35.0 30.0 25.0 20.0 40

45

50

55

60

PBL (DXA) FIGURE 7. Linear regressions showing correlation between (a) ultrasound PBL data and PBL by DXA, (b) ultrasound PBL data and percent porosity, (c) PBL by DXA data and percent porosity. TABLE 4. Results of regression analysis between ultrasound, DXA, and physical tests. Test Bone mineral density (DXA) vs. Dry mass density Percent bone loss (DXA) vs. Percent bone loss (US) Percent bone loss (DXA) vs. Dry mass density Percent bone loss (US) vs. Dry mass density Percent bone loss (US) vs. Bone mineral density (DXA) Percent bone loss (U/S) vs. Porosity (%) Percent bone loss (DXA) vs. Porosity (%)

R2

p value

0.63 0.82 0.69 0.45 0.62 0.51 0.86

0.00632 0.00034 0.00294 0.03270 0.00671 0.02100 0.00012

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia

DISCUSSION The goal of this investigation was to show a proofof-concept method to see whether ultrasound can be used to monitor the level of bone quality in the superficial sub-surface matrix of coxal bones, by correlating ultrasonic echo intensity (quasi-static C-scan image brightness) to DXA images and to physically measured bone properties. Although the measurements in this study are made in cadaver bone samples (in vitro) using a high operating frequency, which does not support a realistic clinical situation where measurements are always conducted in vivo through the skin and other soft tissues, the results were in fact very promising. The PBL trend in the degree of ScM content in the excised cadaver coxal bones scanned by ultrasound was found to be most closely paralleled with PBL evaluations obtained with post-ultrasound DXA tests. Also, the trend in which PBL (US) correlated with bone mineral density (BMD) was found to exhibit a similar behavior when the latter was compared to dry mass density (DmD) of the bone samples. The linear behavior which DXA and physical measurements exhibited was significant in that they followed a similar distribution as the ultrasound data. The relative 3.5% error between ultrasound and DXA corresponded proportionally to the similar trends in which PBL (US) correlated with bone mineral density (BMD) and when the latter was compared to dry mass density (DmD) of the bone samples. However, less successful outcomes were evident when PBL (DXA) compared to DmD revealed a better linearity than that obtained when PBL (US) was compared with the same DmD, and when PBL (DXA) was compared with percent porosity revealing also a better linearity than the one obtained when PBL (US) was compared with the same percent porosity, although PBL (US) and PBL (DXA) measurements paralleled. This discrepancy can be attributed to a couple of reasons. As mentioned earlier, ultrasonic testing interrogated only a small ROI as compared with DXA which covered a lager ROI more representative of the dry mass density of the bone as determined by physical measurements. On the other hand, the dry densities (Table 2) for samples 10 and 63 are much higher than for all other samples, which may be cause for this slight variation. Further analyses however, made on the statistical significance of the difference between the correlations between PBL (US) and porosity and PBL (DXA) and porosity, and those between PBL (US) and DmD and PBL (DXA) and DmD, revealed that each set of correlations were statistically comparable. It should be noted that the relationships reported in this study, that correlate C-scan echo brightness to the sub-surface composite matrix, are not without limita-

949

tion. Our ultrasound set-up used a holder that freely gripped the extracted coxal bones (without posing any tension) from the iliac crest to the pubis articular surface ends, exposing the flat surface of the ilium in between to the ultrasound cone of insonification. The lack of pre-tensioning limits the comparison of these findings to some other studies3 where pre-compression of the bone was performed prior to ultrasound analysis. Another important aspect that needs to be highlighted is that bone tissue is a hierarchical composite at many levels.2 A composite of mineralized collagen fibrils ranging in size from 0.1 to 10 lm is next to another porous layer (5–10 lm) associated with bone cells. Adjacent to this is a layer of cortical bone and a highly porous trabecular bone (200–300 lm thick) with large interstitial spaces.14 Also, there exist wide ranges of reported material property values for bone and bonelike substances.33 It was determined that in the frequency range from 300 to 700 kHz, the speed of sound for trabecular bone ranged from 2500 to 7000 m/s.33 Later, another study25 determined an average sound velocity (3680 m/s) in trabecular bone via computer simulation as a function of bone density, elasticity, and porosity. Assuming that the latter average velocity data is applicable to our subject coxal bones, then using our present ultrasonic frequency of operation (75 MHz), features with diameter greater than about 49 lm (=average sound velocity/frequency of operation) can be detected with the present system sensitivity. Our objective has been to demonstrate a proof-of-concept echographic modality and its potential for clinical applications.

CONCLUSION The results of this study suggest that brightness in ultrasound imaging correlates to certain material properties of healthy and damaged bone tissue. In a clinical environment, this technique may prove useful in assessing bone porosity, and potentially indicating changes in porosity over time, reflecting the advance or reversal of osteoporosis in a patient. The ability to monitor the degree of porosity present in the superficial sub-surface composite matrix could also prove useful in determining the effectiveness of physical therapy treatments.

ACKNOWLEDGMENTS The authors wish to thank Dr. Nick Solounias from the New York College of Osteopathic Medicine of the New York Institute of Technology, Old Westbury,

S. R. GHORAYEB

950

NY, for providing the 10 bone samples for this study; and Dr. Richard J. Puerzer from the School of Engineering and Applied Sciences, Hofstra University, Hempstead, NY, for his valuable comments on statistical analyses. Also, a special thank you goes to Ms. Sharon Sprintz, BSRT, CBDT, of the Bone Mineral Research Center at Winthrop University Hospital, Mineola, NY, for obtaining all DXA scans of the bone sample; and to Ms. Crystal Wagner for setting up the ultrasound experiments and collecting the data while working on her senior design component at Hofstra University.

REFERENCES 1

Bankman, I. Handbook of Medical Imaging: Processing and Analysis Management. New York: Academic Press, 2000. 2 Bartel, D. L., D. T. Davy, and T. M. Keaveny. Orthopaedic Biomechanics: Mechanics and Design in Musculoskeletal Systems. Prentice Hall, NJ: Pearson, 2006. 3 Binkowski, M., Z. Wrobel, and A. Dyszkiewicz. The mineral density and mechanical strength of trabecular bone tissue in densitometry test and compression test. Artif. Intel Methods 145–146, 2003. 4 Bousson, V., F. Peyrin, C. Bergot, M. Hausard, A. Sautet, and J. D. Laredo. Cortical bone in the human femoral neck: three-dimensional appearance and porosity using synchrotron radiation. J. Bone Miner. Res. 19:794–701, 2004. 5 Brighton, C. T., J. Black, and S. R. Pollack. Electrical Properties of Bone and Cartilage. New York: Grune & Stratton Inc., 1979. 6 Cardoso, L., Y. Vendrenyuk, M. B. Schaffler, and S. C. Cowin. Ultrasound wave propagation in disuse-induced osteoporosis. In: Poro-Mechanics IV, edited by H. I. Ling, A. Smyth, and R. Betti. Lancaster, PA: DEStech Publications Inc., 2009. 7 Chen, T., P. Chen, C. Fung, C. Lin, and W. Yao. Quantitative assessment of osteoporosis from the tibia shaft by ultrasound techniques. Med. Eng. Phys. 26:141–145, 2004. 8 Crawford, R., W. Rosenberg, and T. Keaveny. Quantitative computer tomography-based finite element models of the human lumbar vertebral body: effect of element size on stiffness, damage, and fracture strength predictions. Trans. ASME 125:434–438, 2003. 9 Droin, P., G. Berger, and P. Laugier. Velocity dispersion of acoustic waves in cancellous bone. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 45:581–591, 1998. 10 Fisher, R. A. On the probable error of a coefficient of correlation deduced from a small sample. Metron 1:1–32, 1972. 11 From Osteoporosis: Thin Bones, http://www.ncbi.nlm.nih. gov/pubmedhealth/PMH0001400/, the National Center for Biotechnology Information, U.S. National Library of Medicine 8600 Rockville Pike, Bethesda MD, 20894 USA. 12 Hancox, N. M. Biology of Bone. London: Cambridge University Press, 1972. 13 Jenson, F., F. Padilla, and P. Laugier. Estimation of the mean trabecular thickness from backscatter measurements in cancellous bone. Journe´es Os – Ultrasons 1–2, 2002.

AND

D. M. ROONEY 14

Keaveny, T. M., E. F. Morgan, G. L. Niebur, and O. C. Yeh. Biomechanics of trabecular bone. Annu. Rev. Biomed. 3:307–333, 2001. 15 Lasaygues, P., E. Ouedraogo, J. P. Lefebvre, M. Gindre, M. Talmant, and P. Laugier. Bone imaging by ultrasound tomography. Journe´es Os – Ultrasons 1–2, 2002. 16 Laugier, P. In vivo ultrasound assessment of skeletal status: principles and techniques. Journe´es Os – Ultrasons 1–2, 2002. 17 Lefebvre, F., G. Berger, and P. Laugier. Automatic detection of the boundary of the calcaneus from ultrasound parametric images using an active contour model: clinical assessment. IEEE Trans. Med. Imag. 17:45–51, 1998. 18 Lin, J. C., M. Amling, D. C. Newitt, K. Selby, S. K. Srivastav, G. Delling, H. K. Genant, and S. Majumdar. Heterogeneity of trabecular bone structure in the calcaneus using magnetic resonance imaging. Osteoporos. Int. 8:16– 24, 1998. 19 Lochmu¨ller, E., F. Eckstein, J.-B. Zeller, R. Steldinger, and R. Putz. Comparison of quantitative ultrasound in the human calcaneus with mechanical failure loads of the hip and spine. Ultrasound Obstet. Gynecol. 14:125–133, 1999. 20 Mohamed, M., L. Shaat, and A. N. Mahmoud. Propagation of ultrasonic waves through demineralized cancellous bone. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 50:279–288, 2003. 21 Nassiri, D. K. Ultrasound bone densitometry—real or imaginary. Ultrasound Obstet. Gynecol. 14:87–91, 1991. 22 Nowicki, A., J. Litniewski, W. Secomski, P. Lewin, and I. Trots. Estimation of ultrasonic attenuation in a bone using coded excitation. Ultrasonics 41:615–621, 2003. 23 Petculescu, P., R. Zagan, and G. Prodan. Ultrasonic Cscan imaging for the bone sample. J. Optoelectron. Adv. Mater. 8(1):225–229, 2006. 24 Rey, P., S. Morillo, I. Laborde, C. Errecalde, M. P. Alessandrini, S. Colombo, H. Plotkin and J. R. Zanchetta. Discriminatory ability of bone mass measurements (SXA and DXA) for vertebral fractures in postmenopausal women. Osteoporos. Int. 6(1), 1996. 25 Saade´, R. G., G. Tsoukas, and J. Caminis. Understanding velocity of sound in trabecular bone via computer simulations. Comput. Biol. Med. 36(5):439–447, 2006. 26 Shung, K., M. Smith, and B. Tsui. Principles of Medical Imaging. New York: Academic Press, 1992. 27 Steiger, J. H. Tests for comparing elements of a correlation matrix. Psychol. Bull. 87(2):245–251, 1980. 28 Tysarczyk-Niemeyer, G. New noninvasive pQCT devices to determine bone structures. J. Jpn. Soc. Bone Morphom. 7:97–105, 1997. 29 Wallace, K., B. Hoffmeister, L. Thomas IV, S. Kaste, G. Lanza, and S. Wickline. Ultrasonic characterization of backscatter from human cancellous bone with a Renyi entropy metric: correlation with X-ray bone mineral density. IEEE International Ultrasonics Symposium Proceedings, pp. 542–545, 2009. 30 Wear, K. Ultrasonic attenuation in human calcaneus from 0.2 to 1.7 MHz. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 48:602–607, 2001. 31 Wear, K. Characterization of trabecular bone using the backscattered spectral centroid shift. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 50:402–407, 2003. 32 Wear, K., and D. Armstrong. The relationship between ultrasonic backscatter and bone mineral density in human calcaneus. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 47:777–779, 2000.

Ultrasonic Evaluation of Bone Quality in Cadaver Ilia 33

Wear, K. The effect of trabecular material properties on the frequency dependence of backscatter from cancellous bone (L). J. Acoust. Soc. Am. 114(1). 2003. 34 Wear, K., and A. Laib. The dependence of ultrasonic backscatter on trabecular thickness in human calcaneus: theoretical and experimental results. IEEE

951

Trans. Ultrason. Ferroelectr. Freq. Control 50:979–984, 2003. 35 Yamashita, K., F. Fujita, K. Mizuno, I. Mano, and M. Matsukawa. Two-wave propagation imaging to evaluate the structure of cancellous bone. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 59(6):1160–1166, 2012.