- Email: [email protected]
Quantitative Ultrasound Measurements at the Heel: Improvement of Short- and Mid-Term Speed of Sound Precision
Ultrasound in Med. & Biol., Vol. 41, No. 3, pp. 858–870, 2015 Copyright Ó 2015 World Federation for Ultrasound in Medicine & Biology Printed in the USA. All rights reserved 0301-5629/$ - see front matter
http://dx.doi.org/10.1016/j.ultrasmedbio.2014.10.013
d
Original Contribution QUANTITATIVE ULTRASOUND MEASUREMENTS AT THE HEEL: IMPROVEMENT OF SHORT- AND MID-TERM SPEED OF SOUND PRECISION MELANIE DAUGSCHIES,* KIM BRIXEN,y PERNILLE HERMANN,yz KERSTIN ROHDE,* €ER,* and REINHARD BARKMANN* CLAUS-CHRISTIAN GLU * Sektion Biomedizinische Bildgebung, Klinik f€ur Radiologie und Neuroradiologie, Universit€atsklinikum Schleswig-Holstein, Kiel, Germany; y Department of Medical Endocrinology, Odense University Hospital, Odense, Denmark; and z Department of Internal Medicine, Kolding Hospital, Kolding, Denmark (Received 8 April 2014; revised 22 October 2014; in final form 23 October 2014)
Abstract—Calcaneal quantitative ultrasound can be used to predict osteoporotic fracture risk, but its ability to monitor therapy is unclear possibly because of its limited precision. We developed a quantitative ultrasound device (foot ultrasound scanner) that measures the speed of sound at the heel with the aim of minimizing common error sources like the position and penetration angle of the ultrasound beam, as well as the soft tissue temperature. To achieve these objectives, we used a receiver array, mechanics to adjust the beam direction and a foot temperature sensor. In a group of 60 volunteers, short-term precision was evaluated for the foot ultrasound scanner and a commercial device (Achilles Insight, GE Medical, Fairfield, CT, USA). In a subgroup of 20 subjects, mid-term precision (1-mo follow-up) was obtained. Compared with measurement of the speed of sound with the Achilles Insight, measurement with the foot ultrasound scanner reduced precision errors by half (p , 0.05). The study indicates that improvement of the precision of calcaneal quantitative ultrasound measurements is feasible. (E-mail: [email protected]) Ó 2015 World Federation for Ultrasound in Medicine & Biology. Key Words: Quantitative ultrasound, Calcaneus, Mid-term precision, Osteoporosis.
Dual X-ray absorptiometry of the spine or hip and quantitative ultrasound (QUS) measurements at the heel permit estimation of osteoporotic fracture risk with comparable performance (Bauer et al. 2007; Hans et al. 1996; Pinheiro et al. 2006). Therapy monitoring is also relevant, but here the performance of QUS methods remains controversial (Blake and Fogelman 2007; Krieg et al. 2008). DXA monitoring of the spine (Faulkner 1998) is more sensitive compared with that of the hip, as the vertebrae consist mainly of cancellous bone, which is more responsive to changes in bone metabolism because of the larger surface compared with cortical bone. However, degenerative changes of the vertebrae—that is, calcifications, especially those on or near the outer surface of the cortex of the vertebrae—are a major source of error, which limits the potential to use spinal DXA for monitoring (Guglielmi et al. 2005). In addition, drugs such as bisphosphonate minimize fracture risk even if no or only minimal changes in aBMD are measured (Chapurlat et al. 2005; Watts et al. 2004). In contrast to DXA, ultrasound, as a mechanical wave, is influenced by other aspects and is related to bone microstructure and stiffness (Goossens et al. 2008; Hodgskinson et al.
INTRODUCTION The increasing life span in industrial countries leads to an aging population, and thus, disorders of old age become more relevant. Osteoporosis is the most prevalent metabolic bone disease and one of the most frequent diseases in the elderly population (Jones et al. 1994; Warming et al. 2002). It is characterized by bone loss and, hence, increased risk of fractures, which can cause immobilization (Hall et al. 1999) and increased mortality (Center et al. 1999). Because a large proportion of the population (about 39% of the women older than 50 in Germany [H€aussler et al. 2007]) have osteoporosis, high-quality diagnosis, prognosis and therapy monitoring are essential. Today, the clinical gold standard for these tasks is the assessment of areal bone mineral density (aBMD) of the hip and spine by dual X-ray absorptiometry (DXA).
Address correspondence to: Melanie Daugschies, Sektion Biomedizinische Bildgebung, Klinik f€ur Radiologie und Neuroradiologie, Universit€atsklinikum Schleswig-Holstein, Am Botanischen Garten 14, Kiel 24118, Germany. E-mail: [email protected] 858
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
1997; Nicholson et al. 1998). Nevertheless, it is unclear if ultrasound measurements can yield more information about the development of fracture risk during therapy. Some longitudinal studies of different therapies did not prove the feasibility of using QUS for monitoring (Frost et al. 2001; Gonnelli et al. 2002, 2006; Sahota et al. 2000). One contributing factor might be the poor (longterm) precision of current QUS devices as criticized by the International Society for Clinical Densitometry (ISCD) (Krieg et al. 2008). Furthermore, adequate studies reporting the ability of QUS to monitor therapy are lacking, although the ISCD recognized some evidence of this potential. QUS at the heel seems to be better suited for this task than QUS assessments at other sites (e.g., radius and phalanges) because of a stronger response to antiresorptive treatment of the QUS parameters at this site (Krieg et al. 2008). The common QUS parameters measured by commercial devices at the heel include the apparent speed of sound (SOS) and broadband ultrasound attenuation (BUA). Gonnelli et al. (2002) reported a five times higher monitoring time interval for BUA compared with SOS in monitoring bisphosphonate treatment. Therefore, in this study we focused on SOS as the more sensitive parameter of changes in bone status. The sensitivity to detect changes is related to responsiveness (i.e., changes in the reading of the given method over a specific interval) and the long-term precision error (Gl€ uer 1999). Thus, increases in sensitivity may be obtained by decreasing errors in precision. Common error sources in QUS include repositioning of the foot along with the placement of the region of interest (ROI) and the temperature of the coupling medium as well as of the soft tissue (Njeh et al. 1999). The device developed in our lab (foot ultrasound scanner [FUS]) was constructed using innovative design features with the aim of minimizing the impact of these issues by using an ultrasound array to generate an image, mechanics for adjusting the ultrasound incident beam angle, temperature stabilization of the coupling medium and a sensor to measure the temperature of the foot. The FUS was built and tested to assess whether achievement of substantial improvements in SOS mid-term precision compared with commercially available devices is feasible and to estimate the impact of the design features introduced on the precision of calcaneal QUS. METHODS Principle underlying the measurement For QUS measurements of the calcaneus, the foot is placed between two transducers, one acting as emitter and the other as receiver (Fig. 1). In both the Achilles Insight (GE Medical, Fairfield, CT, USA) and the FUS, membranes filled with a liquid acoustical coupling
859
Fig. 1. Arrangement of the transducers at the heel and adjustment of the incident beam angle. Two membranes filled with a coupling liquid can be inflated and, moistened with ethanol, provide the coupling to the skin. The ultrasound wave is excited by the emitter, travels through coupling medium, soft tissue and the calcaneus and arrives at the receiver. The arrows indicate the rotation around a vertical axis and tilting movement around a horizontal axis to adjust the incident beam angle.
medium are used between the foot and transducers. The emitter excites an ultrasound wave that passes through the coupling medium, the soft tissue of the foot and the calcaneus. For SOS measurements, the time between excitation and arrival of the ultrasound wave at the receiver, called time of flight (TOF), is evaluated. TOF is influenced mainly by the properties of the bone in the ultrasound path, but also by the temperature-dependent ultrasound velocities of the coupling medium and the overlying soft tissue. Devices An overview of technical aspects of the two devices can be found in Appendix A. Achilles Insight. In the commercial ultrasound heel scanner Achilles Insight, two quarter wave-matched broadband elements are used as transducers with a center frequency of 0.5 MHz. Their relative position is fixed, providing only one ultrasound beam angle for the measurement. For coupling with the skin, two flexible membranes are filled with 33 C water. The Achilles comprises an ultrasound array as receiver and a single transducer as emitter. It produces an image for controlling the correct positioning and coupling of the skin with the membranes. Only cells in a circular subarea with a diameter of
860
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 2. (a) Receiver array of the foot ultrasound scanner. The array comprises 156 cells with an edge length of 6 mm and a distance of 0.1 mm. The 100 cells within the black square are evaluated. (b) Unfocused emitter consists of one transducer cell 100 mm in diameter.
25.4 mm in the center of the array are analyzed (i.e., resulting in a fixed ROI). The signals of all of these cells are averaged, and the averaged signal is used to calculate the QUS parameters SOS, BUA and stiffness index (SI). For SOS, the first zero crossing with a negative slope is evaluated, and the thickness of the heel is assumed to be 40 mm (Njeh et al. 1999). The BUA is calculated as described by Langton et al. (1984) but not normalized for heel width. The equation used to calculate SI from SOS and BUA, adapted from Njeh et al. (1999), is SI 5 0:673BUA db MHz21 10:283SOS ms21 2420
controlling the temperatures and motors and for excitation and receipt of ultrasonic signals were developed in our laboratory. The emitter is excited by a 100-ns-long, 15-V pulse using a high-speed driver (TC4451, Microchip
(1) Foot ultrasound scanner. As illustrated in Figure 2a, the circular unfocused ultrasound array used as receiver in the FUS has a diameter of 100 mm and consists of 156 square-shaped cells (edge length 5 6 mm) separated by 0.1-mm-wide ditches (only 100 cells in the center are evaluated); opposite it is the unfocused emitter, one transducer cell 100 mm in diameter (Fig. 2b). The center frequency of the transducers is 0.5 MHz with a –3-dB bandwidth of 10%. The distance between emitter and receiver is 120 mm. Both are mounted on a C-arm, which can be rotated around a vertical axis and tilted around a horizontal axis (Fig. 1). The whole setup is housed. At present, the FUS is suited only for measurements of the right foot, which is placed on a foothold between the emitter and receiver (Fig. 3). The lateral side of the foot has to be aligned with the right-hand edges of the foothold. Skin temperature is measured with the device’s temperature sensor, which can be attached to the foot near the heel, but outside of the ultrasonic beam. Coupling of the ultrasound wave to the skin is provided by oil-filled membranes (castor oil), which together with the air inside the housing are maintained at 32.5 C. The electronics for
Fig. 3. Positioning of the right foot in the foot ultrasound scanner. The heel is fixed by a heel positioner below and behind the foot; the lateral side of the foot has to be neatly aligned to the lateral foot margin. Oil-filled membranes (here shown deflated) provide coupling to the skin. The arrows indicate the rotation movement in positive and negative directions.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
Technology, Chandler, AZ, USA). The signals received are multiplexed, amplified (AD8331, Analog Devices, Norwood, MA, USA; gain ranging between 7.5 and 55.5 dB) and digitized with a sampling rate of 40 MHz and a resolution of 14 bit (AD9244, Analog Devices). Remote control and evaluation of the measurements are performed with special software routines developed in our lab and based on LabVIEW (National Instruments, Austin, TX, USA). The emitter is excited 100 times because the signals of all 100 receiver cells are recorded sequentially. One data set, therefore, contains 100 ultrasound signals, each from a different array cell. When data sets are averaged, for example, five times, the ultrasound signal of each array cell is measured five times and these signals are averaged to form a new signal for the given array cell. Study design A total of 60 volunteers participated in this study. All gave written and verbal informed consent. The volunteers (55.4 6 17.3 y; 30 females, 30 males) were measured in Odense in the context of collaboration within an Interreg IVa project of the European Community (‘‘cross boundary improvement of the situation of osteoporosis patients’’). A subgroup of 20 participants (51.7 6 20.0 y; 12 females, 8 males) was measured again after 1 mo to obtain mid-term precision. Short-term precision was calculated for all volunteers. Half of the subjects were measured first with the FUS, and the other half were measured first with the Achilles in alternating sequence. The study was approved by the Ethics Committee of Southern Denmark. Measurement procedure Achilles Insight. The default setting of the Achilles device is a mode that reports only the QUS stiffness index. To access the SOS values, another measurement mode was selected (BUA/SOS mode). The values for SOS, BUA and SI were used to determine the precision error. For each volunteer, two subsequent measurements with repositioning were performed. Before the foot was placed in the device, the skin was wet with ethanol spray. To stir the water for improved temperature control, a quality control measurement was performed before each set of measurements. This was a pure water measurement where the membranes were inflated to touch each other, with no foot in the device. Foot ultrasound scanner. Ultrasound transmission through the right foot was measured with beam directions varied across a range of different tilting and rotation angles. The tilting angles were identical for each measurement and ranged from 18 to –8 considering the horizontal axis as zero position. Measurements were per-
861
formed at five tilting angles with a step size of 4 . The rotation angles were in a range of 30 with a step size of 6 , resulting in measurements of six angles. The start angle for the rotation was adjusted to the medial side of the foot for each tilting angle and, therefore, depended on the size and shape of the foot. The varied tilting and rotation angles resulted in up to 30 different incident beam directions. A temperature sensor was attached to the right foot close to a vein near the ankle on the lateral side. A piece of neoprene was placed over the temperature sensor to isolate it against the warm air inside the device. Afterward, the foot was inserted into the device with the lateral side of the foot and the heel closely aligned with the foot holder as depicted in Figure 3. For each tilting angle, the starting rotation angle was determined individually so as to ensure that for this angle the emitter (always on the medial side, see Fig. 3) was as close to the medial side of the foot as possible. Afterward, the foot was wet with ethanol spray and the membranes were inflated to a pressure of 45 mbar above atmospheric pressure to enable contact with the skin of the foot. The rotation and tilting took approximately 2 min in total. At each incident beam angle, five data sets were recorded. For each cell, five ultrasound signals were averaged. After the membranes were drained, the foot was removed from the device and inserted and wet with ethanol spray again. Then the next measurement was performed. During measurements, the temperatures of the air, oil and foot were logged, as was oil pressure. Two measurements were performed on each volunteer. Before the measurements, a complete scanning routine was performed (inflation of the membranes, rotation and tilting, deflation of the membranes) to homogenize the oil and to allow the foot to adapt to the air and oil temperatures. Analysis Achilles Insight. The values of BUA, SI and SOS were internally calculated by the device in the standard manner. These data were then used to estimate the precision errors of SOS, BUA and SI. Foot ultrasound scanner. Calculation of the SOS. The TOF of the first-arriving signal was defined as the interval between the time at which the pulse was sent and the time of the first zero crossing with a negative slope (Fig. 4). Apparent SOS in the bone was calculated using the formula (Laugier 2008) SOS 5
d d 1TOFb 2TOFref SOSref
(2)
where d is the assumed thickness of the bone (30 mm), SOSref is the SOS in castor oil (1448 m/s at 33 C,
862
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 4. A typical ultrasound signal as received with the array in the region of interest. To calculate the speed of sound, we used TOF, which is the time of flight between excitation at the emitter and the first zero crossing with a negative slope in the signal received. The amplitude evaluated is the amplitude of the first oscillation as indicated in the figure and is used for defining the region of interest. The period length and the durations of the first (positive) and second (negative) halfwaves determine whether the signal is valid or not.
technical information from Karl Deutsch, Wuppertal, Germany), TOFref is the calculated reference TOF in castor oil (82.3 ms for a distance of 120 mm) and TOFb is the measured TOF with the foot placed in the ultrasound pathway. Evaluation of the measurement. Signal analysis requires the application of several measures which are first outlined and then described in detail. An overview of the steps is provided in Appendix B. Signals of a fixed rectangular region in the center of the calcaneus containing 12 cells (24 mm 3 18 mm) were evaluated. Only signals that met preset criteria were considered to be valid. Data sets with valid signals from fewer than two cells were excluded from the analysis. Because this was the case for a majority of the measurements within a specific range of the angle pairs of rotation and tilting, we excluded these angle combinations completely from the evaluation, leaving 15 angle pairs as valid. The angles excluded were 18 and –8 for tilting and the angle farthest from the start angle for rotation. Some features of the preset criteria were adjusted to generate an individual filter for each measurement. Therefore, averaged features of the signals of all valid cells across all valid angles were calculated for a second step of filtering. Then, the cell with the highest weighted signal amplitude of the first oscillation (see Fig. 4, later referred to as ‘‘amplitude’’), weighted by its count of valid signals, was selected for the calculation of SOS in all angles. The final SOS value is the lowest of these values.
In a QUS image of the calcaneus measured in the mediolateral direction, the signal amplitude is always higher in the center of the bone than at its fringes. Figure 5 illustrates an example of such an amplitude image. A predefined fixed region in the array comprising 12 cells (see Fig. 5, black dashed rectangle, and Appendix B, step 2) containing signals transmitted through the central bone region was further analyzed to identify the ROI (comprising a single cell). A filter for analyzing the signal features (see Appendix C) was applied to this region, excluding cells in which signal properties differ strongly from the properties of a typical signal in the valid bone area (see Appendix B, step 3). The latter was defined as a signal with features within a predefined range. Features of the filter included SOS, the period length of the first oscillation, the duration of the first (positive) half-wave and the difference between the durations of the first (positive) and second (negative) half-waves, all visualized in Figure 4. Their values are listed in Appendix C. Signals matching these filter specifications were denoted ‘‘common valid.’’ Because of the variability in properties of the heel bones between subjects, new filter settings were adjusted individually for each subject in a second step. 1. For the 15 valid pairs of angles, the average period length of all common valid signals was calculated, leading to new filter settings for ‘‘period length’’ and ‘‘difference between the durations of first and second half-waves’’ (see Appendix B, steps 4–6). Signals
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
863
Fig. 6. Speed of sound (SOS) values of the region of interest plotted over the rotation and tilting angles, depicted as isocontour lines. A brighter gray indicates a lower SOS. The interval between the iso-contour lines is 2 m/s. Fig. 5. Iso-contour line image (0.2-V difference between lines) illustrating the distribution of signal amplitudes across the ultrasound array. The higher the amplitude, the brighter is the color. The shape of the heel is outlined by the white line. The predefined region is used by the automated algorithm which searches for the region of interest, here outlined by a black dashed square.
matching these filter specifications were denoted ‘‘specific valid.’’ 2. The algorithm calculated the mean amplitude for each cell in the predefined region of all specific valid signals for the 15 valid pairs of angles (see Appendix B, steps 7–9). For each cell, the number of valid signals was used to calculate a weighted value of the mean amplitude. The cells and their weighted mean amplitude defined the output of the algorithm (see Appendix B, step 10). 3. For the cell with the highest weighted amplitude (preliminary ROI), SOS values were calculated for each valid angle (see Appendix B, step 11). The lowest SOS value within the range of the angles applied was taken as the result of the measurements (see Appendix B, step 12). Figure 6 depicts an example of iso-lines of SOS results plotted over the angles with the rotation angles on the x-axis and the tilting angles on the y-axis clearly showing the minimum in SOS. If the signal of the SOS minimum was not specifically valid, the SOS values of the cell with the second highest weighted average amplitude were used for evaluation, and so on. The ROI was saved for each volunteer and used for all follow-up measurements. If the SOS minimum of the ROI in the follow-up measurements was not common valid, the algorithm was applied to the measurements to obtain a new ROI.
Estimation of the influence of rotation and tilting on precision. In the subgroup of 20 subjects remeasured after 1 mo, the influence of rotation and tilting movements on short- and mid-term precision was estimated. In addition to the measurement and evaluation procedure described, the signal at the angle most similar to the Achilles beam angle, later referred to as ‘‘fixed angle,’’ was evaluated as well. The ROI was determined as described above but only for the one fixed angle. In this ROI, SOS was calculated according to formula (2). Short- and mid-term precision data were also calculated for this specific angle. Estimation of the influence of oil temperature and pressure. To verify if variations in oil temperature and pressure had an effect on the measurements, both values were included in a multivariate model to estimate the influence on SOS. Oil temperature was measured during inflation and draining of the membranes. The oil pressure was logged during the measurement for each beam angle. Determining the influence of the foot temperature. The impact of foot temperature on the SOS values was calculated using a multivariate model (multiple linear regression), with SOS as the dependent variable and foot temperature and the IDs of the volunteers as independent variables. The foot temperature coefficient for SOS (kfoot in m/s/ C) of the model was used to calculate a corrected SOS (SOScorr in m/s) according to the formula SOScorr 5 SOS1 wmean 2wfoot kfoot (3) using the original SOS (in m/s), the mean foot temperature of the population ymean (in C) and the actual foot
864
Ultrasound in Medicine and Biology
Table 1. Age-dependent slope per year and short- and mid-term precision errors of the foot ultrasound scanner and the Achilles Insight in absolute values Achilles Insight
Foot ultrasound scanner
SOS BUA (m/s) (dB/MHz) SI
SOS SOScorr* (m/s) (m/s)
Age-dependent slope (y21) –1.9 RMSE short-term (n 5 100) 3.9 RMSE mid-term (n 5 20) 5.9
–0.9 3.0 2.1
–1.1 –2.6 1.7 2.5 2.3 7.1
–2.7 2.8 4.6
BUA 5 broadband ultrasound attenuation; RMSE 5 root mean square error; SOS 5 speed of sound. * SOScorr is corrected for the impact of foot temperature.
temperature yfoot (in C). For these new SOS values, short- and mid-term precision errors were calculated as well. Statistical calculations were performed using JMP (SAS Institute, Cary, NC, USA) and Excel (Microsoft, Redmond, WA, USA). Both devices. Determination of short- and midterm precision. The precision error (root mean square error [RMSE]) was defined as the root mean square of the standard deviations (SD) of repeated measurements. For short-term precision, the SD was determined for two successively performed measurements of each volunteer obtained on a single day. For mid-term precision, the SD was derived from the mean values of the two successive measurements for a given volunteer obtained on different days. Normalization and comparison. In comparison of the precision errors of two different techniques or devices, an appropriate method for their normalization is crucial. Absolute precision errors or those relative to the absolute value are not comparable because they do not consider responsiveness and, thus, are not appropriate as a measure of the longitudinal sensitivity (i.e., ability to detect small changes) of the device or technique, which is the relevant measure for characterizing longitudinal precision (Gl€ uer et al. 1995). To adjust for differences in responsiveness, Gl€ uer (1999) suggested use of the response ratio rr (A vs. R) for standardization purposes. The response ratio is calculated from the response rates of the two parameters to be compared:
Volume 41, Number 3, 2015
Our normalization method is based on the response rate of the linear age-dependent decrease in the measured parameters (SOS, BUA and SI) of the females older than 49. The reference response rate R is the slope of the regression line of the Achilles SOS versus age (slSOS_Ach). The response rate A is the slope of the age-dependent decrease in the parameter to be standardized (slMP). The ratio of these two slopes rr (A vs. R) was then used to calculate the normalized precision errors (RMSEnorm) scaled to the responsiveness of the Achilles SOS: RMSEnorm 5 RMSE rrðA vs: RÞ 5 RMSE
The average measured values of the first two subsequent measurements were used to calculate the slopes. Because precision errors do not follow a Gaussian distribution (tested with a goodness-of-fit test), the nonparametric van der Waerden test was applied to test the significance of the differences between precision values of the FUS (SOS) and Achilles (SOS, BUA, SI). Whether improvements in SOS precision by rotation and tilting were significant was also tested with the van der Waerden test. RESULTS Achilles Insight One volunteer had to be excluded from the midterm follow-up because the Achilles Insight was not able to measure two SOS values with a difference less than 100 m/s or aborted the measurements. The volunteer could be measured with the FUS without problems. The absolute precision errors are listed in Table 1, and the normalized precision errors of BUA and SI are in Table 2. SI had the smallest precision error when normalized (3.0 m/s for short-term and 4.0 m/s Table 2. Short- and mid-term precision errors in SOS measured with the foot ultrasound scanner and parameters of the Achilles Insight normalized to SOS of the Achilles in meters per second RMSEnorm (m/s) Achilles Insight
rrðA vs: RÞ 5 response rate R=response rate A
(4)
To compare the SOS precision errors of the two devices and to investigate how BUA and SI performed compared with SOS, we also expressed the errors of our device and the errors of BUA and SI of the Achilles in recalculated units of the Achilles SOS precision error.
slSOS Ach slMP (5)
Short-term (n 5 100) Mid-term (n 5 20)
Foot ultrasound scanner
SOS
From BUA
From SI
From SOS
From SOScorr
3.9 5.9
6.3 4.4
2.9 4.0
1.8 5.2
2.0 3.2
BUA 5 broadband ultrasound attenuation; RMSE 5 root mean square error; SOS 5 speed of sound.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
Fig. 7. Decrease in speed of sound (SOS) values with age for women older than 49. Illustrated here are the SOS values of the Achilles (p , 0.01) and the SOScorr values of the foot ultrasound scanner (p , 0.002). FUS 5 foot ultrasound scanner, SOS 5 speed of sound.
for mid-term precision); these values are lower than those of SOS (3.9 m/s for short-term and 5.9 m/s for mid-term precision) and normalized BUA (6.3 m/s for short-term and 4.4 m/s for mid-term precision). Slopes of age dependence (Table 1) were –1.9 6 0.6 m/s/y (r2 5 0.27, p , 0.01) for SOS (Fig. 7), –0.9 6 0.4 dB/MHz/y (r2 5 0.18, p , 0.04) for BUA and 1.1 6 0.4/y (r2 5 0.23, p , 0.02) for SI. Foot ultrasound scanner One volunteer had no common valid SOS minimum for the mid-term follow-up measurement. The ROI had to be re-defined as described above. Two other volunteers had no specific valid SOS minimum at the baseline measurement. The cell with the next highest weighted average amplitude was chosen as the ROI. Absolute and normalized precision errors of SOS and SOScorr are listed in Tables 1 and 2. Oil temperature (stable at 32.4 6 0.4 C) and pressure had no significant effect on the SOS values, but foot temperature did influence SOS, with kfoot 5 – 4.9 6 0.6 m/s/ C (p , 0.0001, r2 5 0.43). The shortterm precision error increased slightly when the foot temperature correction was applied (2.8 m/s vs. 2.5 m/s), but the mid-term precision error was reduced from 7.1 to 4.6 m/s (p , 0.1, van der Waerden test). The age dependences were almost the same for the SOS and SOScorr values, with slopes of –2.6 6 0.7 m/s/y (r2 5 0.34, p , 0.003) and –2.7 6 0.7 m/s/y (r2 5 0.38, p , 0.002) (see Fig. 7), respectively. Comparison As illustrated in Figure 8, the SOS values of the FUS correlated well with the values of the Achilles Insight
865
(r2 5 0.74 and p , 0.0001 for SOS; r2 5 0.73 and p , 0.0001 for SOScorr). The linear age-related decrease in Achilles SOS and FUS SOScorr in females older than 49 y (n 5 25) is illustrated in Figure 7. The normalized precision errors reflecting the differences in age dependence are listed in Table 2. For short-term precision, the normalized precision error of FUS for SOScorr (2.0 m/s) was significantly lower than the Achilles SOS precision error (3.9 m/s) at p , 0.005 and its normalized BUA and SI precision errors (6.3 and 2.9 m/s) at p , 0.0001 and p , 0.03, respectively. For mid-term precision, the normalized precision error for FUS SOScorr (3.2 m/s) was significantly decreased (p , 0.05) compared with the Achilles SOS precision error (5.9 m/s) and a trend (0.09 , p , 0.14) compared with normalized BUA and SI precision errors (4.4 m/s respectively 4.0 m/s). The FUS SOScorr shortand mid-term precision errors were almost halved compared with the Achilles SOS precision errors (2.0 m/s vs. 3.9 m/s for short-term and 3.2 m/s vs. 5.9 m/s for mid-term). Influence of rotation and tilting on precision errors The short- and mid-term SOS precision errors of the FUS for four different methods are illustrated in Table 3. On the right-hand side are the precision errors for the procedure in which the lowest SOS was determined across all angles (variable angle). On the left-hand side are the precision errors for the fixed angle, that is, the angle close to the Achilles incident beam angle (angle fixed). For both methods, errors are given without and with foot temperature correction. The influence of foot temperature on the
Fig. 8. Correlation between the SOS values of the Achilles and the SOScorr values (SOS corrected for foot temperature) of the FUS (p , 0.0001). FUS 5 foot ultrasound scanner; RMS 5 root mean square; SOS 5 speed of sound.
866
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Table 3. Short- and mid-term precision errors in SOS measured with the foot ultrasound scanner for two different angles (n 5 20) Standard deviation (m/s) Angle fixed
Short-term Mid-term
Angle variable
Uncorrected for foot temperature
Corrected for foot temperature
Uncorrected for foot temperature
Corrected for foot temperature
4.4 9.0
4.5 6.8
2.7* 7.1
3.0 4.6*,y,z
* Trend (p , 0.1), improvement by rotation/tilting movement. y Trend (p , 0.1), improvement by consideration of foot temperature. z Significant (p , 0.05), improvement by considering both foot temperature and rotation/tilting movement.
SOS values was kfoot_var 5 –5.1 6 0.7 m/s/ C (p , 0.0001) for the variable angle method and kfoot_fix 5 –5.8 6 1.0 m/s/ C (p , 0.0001) for the fixed angle method. The foot temperature correction alone yielded a borderline significant improvement for midterm precision of the variable angle method by 35% (p , 0.09). Rotation and tilting improved results with borderline significance: short-term precision by 39% (p , 0.1) when applied without temperature correction and mid-term precision by 32% (p , 0.12) when foot temperature correction was applied. When the influence of both foot temperature and rotation/tilting were considered, we achieved a not significant improvement for the short-term precision error of 32% (p , 0.17) and a significant improvement of 49% (p , 0.02) for the mid-term precision error. All other combinations not mentioned were not significant. DISCUSSION Using an advanced method including patientspecific definition of a ROI in the ultrasound image, adaption of the beam incidence angle and adjustment for foot temperature, we were able to improve the SOS precision of calcaneal ultrasound measurements. The method included an automated algorithm to identify the ROI (cell with highest weighted transmission amplitude) at the baseline measurement, keeping this ROI constant across visits provided it exhibited valid signals and searching for an SOS minimum over the rotation and tilting angles. The correlation with the Achilles was quite good (r2 5 0.73); however, it was not good enough to be used for standardization purposes. Instead, we normalized the precision values to those of the Achilles SOS by using the ratio of the age-dependent slopes. By use of this standardization method, precision errors of the FUS were almost halved compared with the Achilles SOS precision errors (2.0 m/s vs. 3.9 m/s for short-term and 3.2 vs. 5.9 m/s for mid-term precision). The regression coefficients of the age-dependent slopes found in
this study were in the same order as those reported by other studies. Hans et al. (1995) reported a regression coefficient of r 5 –0.55 resulting in r2 5 0.3, whereas Schott et al. (1993) noted r 5 –0.63 for postmenopausal women resulting in r2 5 0.4. The SOS, BUA and SI values of the Achilles had smaller regression coefficients of r2 ranging between 0.18 and 0.27; the SOS of the FUS had a regression coefficient comparable to those reported, with r2 5 0.34 and r2 5 0.38 when corrected for foot temperature. Age dependence was significant with 0.002 , p , 0.04, even though the group was relatively small with only 25 subjects. Interestingly, in women older than 49 y, age-dependent decreases in the SOS of the FUS (RMSEnorm 5 22–23 m/s) scattered less than all normalized variables of the Achilles (RMSEnorm 5 27–34 m/s). This might indicate that the FUS SOS is less affected by error sources than the Achilles parameters. Whether this affects the sensitivity of the device, however, would have to be determined in a longitudinal study. One source of error not accounted for by any of the devices was the true thickness of the bone measured. For the calculation of SOS, the thickness is assumed to be fixed: For the FUS at 30 mm, and for the Achilles at 40 mm (Njeh et al. 1999). In principle, the bone width may be estimated using signals reflected from the bone surface. However, the spatial accuracy is limited by the large wavelength of the ultrasound wave of about 3 mm in soft tissue. Miller et al. (1993) reported that there is no improvement in SOS precision error when a true instead of an apparent SOS is calculated. Moreover, correct estimation of the thickness and calculation of a true SOS value are not crucial in monitoring therapy, because only a precise assessment of changes between measurements at different points in time is required and the width of the bone can be expected to remain constant. The coefficient of the foot temperature was of the expected order of magnitude. Rajagopalan et al. (1979) reported a temperature coefficient for fat (human female breast) of –3.1 m/s/ C in the temperature range 22 C–37 C. In our study, foot temperatures were in the range 26.8 C–34.6 C. The coefficient determined with
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
our model ranged between –4.9 and –5.8 m/s/ C. Because the ultrasound pathway comprises subcutaneous fat and bone marrow and because temperature was measured at the skin and not inside the heel, no better agreement can be expected. An extended temperature measurement (e.g., at different locations on the foot) and monitoring of temperature dynamics might help to further improve the adjustment for temperature variations beyond the reduction in mid-term precision errors of 35% achieved with our method. Keeping the ROI constant across baseline and follow-up measurements is suboptimal but seems to be the best method for the current FUS device. Through use of this approach, repositioning errors will not be compensated if the ROI exhibits valid signals. On the other hand, with our method, possible anisotropic changes in bone structure which might lead to undesired changes in ROI positioning do not distort our measurements because the region is kept constant. One way of obtaining a constant ROI in the bone in the future might be to use an ultrasound array with a higher spatial resolution and to determine the ROI via the outline of the bone and not on the basis of transmission amplitude. The added feature of rotating and tilting the incident beam angle had positive effects on short-term and midterm precision, but with only borderline significance. This is not surprising given that only 20 subjects were included in this analysis. Foot temperature correction and tilting/rotation contributed almost equally to the precision errors. The foot temperature correction resulted in a 35% decrease in the mid-term precision error, and tilting/rotation decreased short- and mid-term precision errors between 32% and 39%. The combination of both had a significant effect on the mid-term precision error, which was reduced by 49%. Several improvements regarding the mechanical design of our scanner are conceivable. The mechanics to adjust the rotation angle were not optimal. Tolerance was relatively high at 0.7 considering the step size of 6 . Nevertheless, this should be a minor problem near the SOS minimum, where only slight changes in SOS (Fig. 6) were observed. For 52% of the subjects, the SOS minimum found was at the limit of the chosen range of angles. In 76% of them, it was close to the starting angle for rotation. The size of the transducers and their housings did not permit larger rotations. Thus, transducers and their housings should be kept small to enable a better starting angle and a higher probability of finding a SOS minimum. Mahrt et al. (1991) reported that with sufficient effort, a short-term SOS precision error of 0.1 m/s can be achieved. They measured the heel of a volunteer in a water bath until the foot reached thermal equilibrium and searched for a SOS minimum by scanning five
867
coordinates. The cited precision of 0.1 m/s is comparable to a value of 0.45 m/s taking into account a different calculation formula used in the Achilles Insight and the FUS. Chappard et al. (1999) reported a decrease in SOS of around 7.4 m/s during the first 7 min of immersion in a tempered water bath for postmenopausal women. This effect was smaller for premenopausal women (around 3.0 m/s) and men (increase of 1.5 m/s). We aimed to minimize the drifting effect by performing a complete ‘‘dummy’’ measurement before the first measurement, which, together with adjustment of the starting rotation angles, added about 7 min to the examination time. However, monitoring foot temperature dynamics might lead to further improvement. In this study, we focused on an optimized assessment of SOS rather than BUA or a combined index like SI. We used an array with a small bandwidth of 10%, resulting in a frequency range of 0.45–0.55 MHz, which made it difficult to calculate BUA. The typical frequency range for which BUA is calculated is 0.2–0.6 MHz (Langton and Njeh 2008). Therefore, the BUA of the FUS may not exploit the full potential with respect to precision and, thus, we did not evaluate its performance. The focus on SOS was motivated by reports in the literature. Here, standardized precision errors of SOS measurements are generally smaller than those of BUA measurements. The FUS performed numerically better than BUA of the Achilles for all types of precision errors assessed. As mentioned in the Introduction, Gonnelli et al. (2002) reported a monitoring time interval for BUA that was five times higher compared with that for SOS, documenting the superior performance of SOS in monitoring bisphosphonate treatment. The parameter SI, which combines SOS and BUA, offers the advantage of more robust performance compared with a single measure, but because it is calculated from weighted averages of BUA and SOS, the poorer performance of BUA precision will have a negative impact on SI precision. Therefore, we decided to focus on SOS in this study on precision. The precision errors of the FUS were numerically smaller than the normalized SI precision errors of the Achilles Insight (Table 2). In a previous study, Gonnelli et al. (2002) reported that a change of 1% aBMD measured with DXA at the lumbar spine (which is a typical level for DXA precision errors) was associated with a change of 2.4 m/s in SOS measured at the calcaneus with an Achilles Plus device. Our device had a normalized mid-term precision of 3.2 m/s, which corresponds to a DXA error of 1.3% and, thus, would be close to the precision error of lumbar spine aBMD. In summary, the new QUS calcaneus device tested here exhibited significantly improved precision compared with the commercially distributed Achilles
868
Ultrasound in Medicine and Biology
Insight device. This improvement was achieved combining use of an array transducer as the receiver with consistent ROI definition, the selection of optimized transmission angles and an adjustment for variability in foot temperature. As a result, the normalized precision error was close to the level of precision achieved with DXA of the spine. Further improvements with more advanced technology, for example, transducers with higher spatial resolution and a better evaluation algorithm, appear to be feasible. Testing of the performance of our improved method for monitoring response to treatment appears warranted. Acknowledgments—We thank Jane Nielsen and Elizabeth Hanmann for recruiting the volunteers and performing the Achilles measurements at Odense University Hospital.—This work was partly funded by the Federal Ministry for Economic Affairs and Technology (grant number 03VWP0042) (Program ‘‘Signo’’) and the European Union in the framework of Interreg IVa for cross-border collaborations (Project 04-1.5-08).
REFERENCES Bauer DC, Ewing SK, Cauley JA, Ensrud KE, Cummings SR, Orwoll ES. Quantitative ultrasound predicts hip and non-spine fracture in men: The MrOS study. Osteoporos Int 2007;18:771–777. Blake GM, Fogelman I. The role of DXA bone density scans in the diagnosis and treatment of osteoporosis. Postgrad Med J 2007;83: 509–517. Center JR, Nguyen TV, Schneider D, Sambrook PN, Eisman JA. Mortality after all major types of osteoporotic fracture in men and women: An observational study. Lancet 1999;353:878–882. Chappard C, Berger G, Roux C, Laugier P. Ultrasound measurement on the calcaneus: Influence of immersion time and rotation of the foot. Osteoporos Int 1999;9:318–326. Chapurlat RD, Palermo L, Ramsay P, Cummings SR. Risk of fracture among women who lose bone density during treatment with alendronate. The Fracture Intervention Trial. Osteoporos Int 2005;16: 842–848. Faulkner KG. Bone densitometry; Choosing the proper skeletal site to measure. J Clin Densitom 1998;1:279–285. Frost ML, Blake GM, Fogelman I. Changes in QUS and BMD measurements with antiresorptive therapy: A two-year longitudinal study. Calcif Tissue Int 2001;69:138–146. Gl€uer C, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate assessment of precision errors: How to measure the reproducibility of bone densitometry techniques. Osteoporos Int 1995;5:262–270. Gl€ uer CC. Monitoring skeletal changes by radiologic techniques. J Bone Miner Res 1999;14:1952–1962. Gonnelli S, Cepollaro C, Montagnani A, Martini S, Gennari L, Mangeri M, Gennari C. Heel ultrasonography in monitoring alendronate therapy: A four-year longitudinal study. Osteoporos Int 2002; 13:415–421. Gonnelli S, Martini G, Caffarelli C, Salvadori S, Cadirni A, Montagnani A, Nuti R. Teriparatide’s effects on quantitative ultrasound parameters and bone density in women with established osteoporosis. Osteoporos Int 2006;17:1524–1531. Goossens L, Vanderoost J, Jaecques S, Boonen S, D’hooge J, Lauriks W, Van der Perre G. The correlation between the SOS in trabecular bone and stiffness and density studied by finite-element analysis. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1234–1242. Guglielmi G, Floriani I, Torri V, Li J, van Kuijk C, Genant HK, Lang TF. Effect of spinal degenerative changes on volumetric bone mineral density of the central skeleton as measured by quantitative computed tomography. Acta Radiol 2005;46:269–275. Hall S, Criddle R, Comito T, Prince R. A case–control study of quality of life and functional impairment in women with long–standing vertebral osteoporotic fracture. Osteoporos Int 1999;9:508–515.
Volume 41, Number 3, 2015 Hans D, Dargent-Molina P, Schott AM, Sebert JL, Cormier C, Kotzki PO, Delmas PD, Pouilles JM, Breart G, Meunier PJ. Ultrasonographic heel measurements to predict hip fracture in elderly women: The EPIDOS prospective study. Lancet 1996;348:511–514. Hans D, Schott A, Arlot M. Influence of anthropometric parameters on ultrasound measurements of Os calcis. Osteoporos Int 1995;5: 371–376. H€aussler B, Gothe H, G€ol D, Glaeske G, Pientka L, Felsenberg D. Epidemiology, treatment and costs of osteoporosis in Germany—The BoneEVA Study. Osteoporos Int 2007;18:77–84. Hodgskinson R, Njeh CF, Currey JD, Langton CM. The ability of ultrasound velocity to predict the stiffness of cancellous bone in vitro. Bone 1997;21:183–190. Jones G, Nguyen T, Sambrook P, Kelly PJ, Eisman JA. Progressive loss of bone in the femoral neck in elderly people: Longitudinal findings from the Dubbo osteoporosis epidemiology study. BMJ 1994;309: 691–695. Krieg MA, Barkmann R, Gonnelli S. Quantitative ultrasound in the management of osteoporosis: The 2007 ISCD Official Positions. J Clin Densitom 2008;11:163–187. Langton CM, Njeh CF. The measurement of broadband ultrasonic attenuation in cancellous bone—A review of the science and technology. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1546–1554. Langton CM, Palmer SB, Porter RW. The measurement of broadband ultrasonic attenuation in cancellous bone. Eng Med 1984;13:89–91. Laugier P. Instrumentation for in vivo ultrasonic characterization of bone strength. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1179–1196. Mahrt KH, Barkmann R, Niedermayer W, Kroebel W. Extremely sensitive ultrasound scanning system for non-invasive in vivo detection of mineralisation changes in human heel bones. Proc IEEE Ultrason Symp 1991;1119–1122. Miller C, Herd R, Ramalingam T. Ultrasonic velocity measurements through the calcaneus: Which velocity should be measured? Osteoporos Int 1993;3:31–35. Nicholson PH, M€uller R, Lowet G, Cheng XG, Hildebrand T, R€uegsegger P, van der Perre G, Dequeker J, Boonen S. Do quantitative ultrasound measurements reflect structure independently of density in human vertebral cancellous bone? Bone 1998;23:425–431. Njeh CF, Hans D, Fuerst T, Gl€uer CC, Genant HK. Quantitative ultrasound: Assessment of osteoporosis and bone status. London: Dunitz, Martin; 1999. Pinheiro MM, Castro CM, Szejnfeld VL. Low femoral bone mineral density and quantitative ultrasound are risk factors for new osteoporotic fracture and total and cardiovascular mortality: A 5-year population-based study of Brazilian elderly women. J Gerontol A Biol Sci Med Sci 2006;61:196–203. Rajagopalan B, Greenleaf JF, Thomas PJ, Johnson SA, Bahn RC. Variation of acoustic speed with temperature in various excised human tissues studied by ultrasound computerized tomography. In: Ultrasonic tissue characterization II: A collection of reviewed papers based on talks presented at the Second International Symposium on Ultrasonic Tissue Characterization. Washington, DC: U.S. Government Printing Office; 1979. p. 227–233. Sahota O, San P, Cawte S. A comparison of the longitudinal changes in quantitative ultrasound with dual-energy X-ray absorptiometry: The four-year effects of hormone replacement therapy. Osteoporos Int 2000;11:52–58. Schott A, Hans D, Sornay-Rendu E, Delmas P, Meunier P. Ultrasound measurements on os calcis: Precision and age-related changes in a normal female population. Top Geriatr Rehabil 1993;8:7. Warming L, Hassager C, Christiansen C. Changes in bone mineral density with age in men and women: A longitudinal study. Osteoporos Int 2002;13:105–112. Watts NB, Cooper C, Lindsay R, Eastell R, Manhart MD, Barton IP, van Staa TP, Adachi JD. Relationship between changes in bone mineral density and vertebral fracture risk associated with risedronate: Greater increases in bone mineral density do not relate to greater decreases in fracture risk. J Clin Densitom 2004;7:255–261.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
869
APPENDIX A Appendix A. Overview of the Achilles Insight and foot ultrasound scanner Achilles Insight Emitter Receiver
1 Array
Center frequency Bandwidth Coupling medium Ultrasound velocity at 33 C Tempered Parameters measured Beam angle Region of interest
0.5 MHz Quarter wave-matched broadband elements Water 1516 m/s 33 C SOS, BUA, SI (combination of SOS and BUA) Fixed Fixed, 25-mm diameter
Duration of measurement Foot to measure Extras
30–60 s Both —
BUA 5 broadband ultrasound attenuation; SI 5 stiffness index; SOS 5 speed of sound.
Foot ultrasound scanner 1, diameter 5 100 mm Array (100-mm diameter, 100 cells evaluated, edge length 5 6 mm, 0.1-mm distance between cells) 0.5 MHz 10% Castor oil 1448 m/s 32.5 C SOS 6 8 tilting, 30 rotation (adjusted to foot) 1 array cell (6 3 6 mm), manually determined for each volunteer 2 min Right Foot temperature measurement used for adjusting SOS
870
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Appendix B. Algorithm for identifying the region of interest*
ROI 5 region of interest; SOS 5 speed of sound. * This schematic drawing illustrates the principle underlying the algorithm for identifying the region of interest. The algorithm is described in the text (see Evaluation of the Measurement in the Methods section). Briefly, a filter with various parameters is applied to a fixed region in the array. For all valid signals in the region for previously defined angles, an average period length is calculated that determines the parameters for a second filter applied to the same region. The signals fitting the filter conditions are used to calculate a weighted amplitude for each cell within the region. The highest weighted amplitude determines the region of interest for which all speed of sound values are plotted over the angles. The smallest speed of sound value is the measurement result.
Appendix C. Filter parameters for automatic algorithm Speed of sound (m/s) Period length of first oscillation (ms) Duration of first positive half-wave (ms)
Difference between first positive and negative half-waves (ms)
1450–1800 1450–1800
1 4
2.1–3.5 PLall 6 0.5
1.0–2.0 1.0–2.0
,0.7 PLall
http://dx.doi.org/10.1016/j.ultrasmedbio.2014.10.013
d
Original Contribution QUANTITATIVE ULTRASOUND MEASUREMENTS AT THE HEEL: IMPROVEMENT OF SHORT- AND MID-TERM SPEED OF SOUND PRECISION MELANIE DAUGSCHIES,* KIM BRIXEN,y PERNILLE HERMANN,yz KERSTIN ROHDE,* €ER,* and REINHARD BARKMANN* CLAUS-CHRISTIAN GLU * Sektion Biomedizinische Bildgebung, Klinik f€ur Radiologie und Neuroradiologie, Universit€atsklinikum Schleswig-Holstein, Kiel, Germany; y Department of Medical Endocrinology, Odense University Hospital, Odense, Denmark; and z Department of Internal Medicine, Kolding Hospital, Kolding, Denmark (Received 8 April 2014; revised 22 October 2014; in final form 23 October 2014)
Abstract—Calcaneal quantitative ultrasound can be used to predict osteoporotic fracture risk, but its ability to monitor therapy is unclear possibly because of its limited precision. We developed a quantitative ultrasound device (foot ultrasound scanner) that measures the speed of sound at the heel with the aim of minimizing common error sources like the position and penetration angle of the ultrasound beam, as well as the soft tissue temperature. To achieve these objectives, we used a receiver array, mechanics to adjust the beam direction and a foot temperature sensor. In a group of 60 volunteers, short-term precision was evaluated for the foot ultrasound scanner and a commercial device (Achilles Insight, GE Medical, Fairfield, CT, USA). In a subgroup of 20 subjects, mid-term precision (1-mo follow-up) was obtained. Compared with measurement of the speed of sound with the Achilles Insight, measurement with the foot ultrasound scanner reduced precision errors by half (p , 0.05). The study indicates that improvement of the precision of calcaneal quantitative ultrasound measurements is feasible. (E-mail: [email protected]) Ó 2015 World Federation for Ultrasound in Medicine & Biology. Key Words: Quantitative ultrasound, Calcaneus, Mid-term precision, Osteoporosis.
Dual X-ray absorptiometry of the spine or hip and quantitative ultrasound (QUS) measurements at the heel permit estimation of osteoporotic fracture risk with comparable performance (Bauer et al. 2007; Hans et al. 1996; Pinheiro et al. 2006). Therapy monitoring is also relevant, but here the performance of QUS methods remains controversial (Blake and Fogelman 2007; Krieg et al. 2008). DXA monitoring of the spine (Faulkner 1998) is more sensitive compared with that of the hip, as the vertebrae consist mainly of cancellous bone, which is more responsive to changes in bone metabolism because of the larger surface compared with cortical bone. However, degenerative changes of the vertebrae—that is, calcifications, especially those on or near the outer surface of the cortex of the vertebrae—are a major source of error, which limits the potential to use spinal DXA for monitoring (Guglielmi et al. 2005). In addition, drugs such as bisphosphonate minimize fracture risk even if no or only minimal changes in aBMD are measured (Chapurlat et al. 2005; Watts et al. 2004). In contrast to DXA, ultrasound, as a mechanical wave, is influenced by other aspects and is related to bone microstructure and stiffness (Goossens et al. 2008; Hodgskinson et al.
INTRODUCTION The increasing life span in industrial countries leads to an aging population, and thus, disorders of old age become more relevant. Osteoporosis is the most prevalent metabolic bone disease and one of the most frequent diseases in the elderly population (Jones et al. 1994; Warming et al. 2002). It is characterized by bone loss and, hence, increased risk of fractures, which can cause immobilization (Hall et al. 1999) and increased mortality (Center et al. 1999). Because a large proportion of the population (about 39% of the women older than 50 in Germany [H€aussler et al. 2007]) have osteoporosis, high-quality diagnosis, prognosis and therapy monitoring are essential. Today, the clinical gold standard for these tasks is the assessment of areal bone mineral density (aBMD) of the hip and spine by dual X-ray absorptiometry (DXA).
Address correspondence to: Melanie Daugschies, Sektion Biomedizinische Bildgebung, Klinik f€ur Radiologie und Neuroradiologie, Universit€atsklinikum Schleswig-Holstein, Am Botanischen Garten 14, Kiel 24118, Germany. E-mail: [email protected] 858
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
1997; Nicholson et al. 1998). Nevertheless, it is unclear if ultrasound measurements can yield more information about the development of fracture risk during therapy. Some longitudinal studies of different therapies did not prove the feasibility of using QUS for monitoring (Frost et al. 2001; Gonnelli et al. 2002, 2006; Sahota et al. 2000). One contributing factor might be the poor (longterm) precision of current QUS devices as criticized by the International Society for Clinical Densitometry (ISCD) (Krieg et al. 2008). Furthermore, adequate studies reporting the ability of QUS to monitor therapy are lacking, although the ISCD recognized some evidence of this potential. QUS at the heel seems to be better suited for this task than QUS assessments at other sites (e.g., radius and phalanges) because of a stronger response to antiresorptive treatment of the QUS parameters at this site (Krieg et al. 2008). The common QUS parameters measured by commercial devices at the heel include the apparent speed of sound (SOS) and broadband ultrasound attenuation (BUA). Gonnelli et al. (2002) reported a five times higher monitoring time interval for BUA compared with SOS in monitoring bisphosphonate treatment. Therefore, in this study we focused on SOS as the more sensitive parameter of changes in bone status. The sensitivity to detect changes is related to responsiveness (i.e., changes in the reading of the given method over a specific interval) and the long-term precision error (Gl€ uer 1999). Thus, increases in sensitivity may be obtained by decreasing errors in precision. Common error sources in QUS include repositioning of the foot along with the placement of the region of interest (ROI) and the temperature of the coupling medium as well as of the soft tissue (Njeh et al. 1999). The device developed in our lab (foot ultrasound scanner [FUS]) was constructed using innovative design features with the aim of minimizing the impact of these issues by using an ultrasound array to generate an image, mechanics for adjusting the ultrasound incident beam angle, temperature stabilization of the coupling medium and a sensor to measure the temperature of the foot. The FUS was built and tested to assess whether achievement of substantial improvements in SOS mid-term precision compared with commercially available devices is feasible and to estimate the impact of the design features introduced on the precision of calcaneal QUS. METHODS Principle underlying the measurement For QUS measurements of the calcaneus, the foot is placed between two transducers, one acting as emitter and the other as receiver (Fig. 1). In both the Achilles Insight (GE Medical, Fairfield, CT, USA) and the FUS, membranes filled with a liquid acoustical coupling
859
Fig. 1. Arrangement of the transducers at the heel and adjustment of the incident beam angle. Two membranes filled with a coupling liquid can be inflated and, moistened with ethanol, provide the coupling to the skin. The ultrasound wave is excited by the emitter, travels through coupling medium, soft tissue and the calcaneus and arrives at the receiver. The arrows indicate the rotation around a vertical axis and tilting movement around a horizontal axis to adjust the incident beam angle.
medium are used between the foot and transducers. The emitter excites an ultrasound wave that passes through the coupling medium, the soft tissue of the foot and the calcaneus. For SOS measurements, the time between excitation and arrival of the ultrasound wave at the receiver, called time of flight (TOF), is evaluated. TOF is influenced mainly by the properties of the bone in the ultrasound path, but also by the temperature-dependent ultrasound velocities of the coupling medium and the overlying soft tissue. Devices An overview of technical aspects of the two devices can be found in Appendix A. Achilles Insight. In the commercial ultrasound heel scanner Achilles Insight, two quarter wave-matched broadband elements are used as transducers with a center frequency of 0.5 MHz. Their relative position is fixed, providing only one ultrasound beam angle for the measurement. For coupling with the skin, two flexible membranes are filled with 33 C water. The Achilles comprises an ultrasound array as receiver and a single transducer as emitter. It produces an image for controlling the correct positioning and coupling of the skin with the membranes. Only cells in a circular subarea with a diameter of
860
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 2. (a) Receiver array of the foot ultrasound scanner. The array comprises 156 cells with an edge length of 6 mm and a distance of 0.1 mm. The 100 cells within the black square are evaluated. (b) Unfocused emitter consists of one transducer cell 100 mm in diameter.
25.4 mm in the center of the array are analyzed (i.e., resulting in a fixed ROI). The signals of all of these cells are averaged, and the averaged signal is used to calculate the QUS parameters SOS, BUA and stiffness index (SI). For SOS, the first zero crossing with a negative slope is evaluated, and the thickness of the heel is assumed to be 40 mm (Njeh et al. 1999). The BUA is calculated as described by Langton et al. (1984) but not normalized for heel width. The equation used to calculate SI from SOS and BUA, adapted from Njeh et al. (1999), is SI 5 0:673BUA db MHz21 10:283SOS ms21 2420
controlling the temperatures and motors and for excitation and receipt of ultrasonic signals were developed in our laboratory. The emitter is excited by a 100-ns-long, 15-V pulse using a high-speed driver (TC4451, Microchip
(1) Foot ultrasound scanner. As illustrated in Figure 2a, the circular unfocused ultrasound array used as receiver in the FUS has a diameter of 100 mm and consists of 156 square-shaped cells (edge length 5 6 mm) separated by 0.1-mm-wide ditches (only 100 cells in the center are evaluated); opposite it is the unfocused emitter, one transducer cell 100 mm in diameter (Fig. 2b). The center frequency of the transducers is 0.5 MHz with a –3-dB bandwidth of 10%. The distance between emitter and receiver is 120 mm. Both are mounted on a C-arm, which can be rotated around a vertical axis and tilted around a horizontal axis (Fig. 1). The whole setup is housed. At present, the FUS is suited only for measurements of the right foot, which is placed on a foothold between the emitter and receiver (Fig. 3). The lateral side of the foot has to be aligned with the right-hand edges of the foothold. Skin temperature is measured with the device’s temperature sensor, which can be attached to the foot near the heel, but outside of the ultrasonic beam. Coupling of the ultrasound wave to the skin is provided by oil-filled membranes (castor oil), which together with the air inside the housing are maintained at 32.5 C. The electronics for
Fig. 3. Positioning of the right foot in the foot ultrasound scanner. The heel is fixed by a heel positioner below and behind the foot; the lateral side of the foot has to be neatly aligned to the lateral foot margin. Oil-filled membranes (here shown deflated) provide coupling to the skin. The arrows indicate the rotation movement in positive and negative directions.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
Technology, Chandler, AZ, USA). The signals received are multiplexed, amplified (AD8331, Analog Devices, Norwood, MA, USA; gain ranging between 7.5 and 55.5 dB) and digitized with a sampling rate of 40 MHz and a resolution of 14 bit (AD9244, Analog Devices). Remote control and evaluation of the measurements are performed with special software routines developed in our lab and based on LabVIEW (National Instruments, Austin, TX, USA). The emitter is excited 100 times because the signals of all 100 receiver cells are recorded sequentially. One data set, therefore, contains 100 ultrasound signals, each from a different array cell. When data sets are averaged, for example, five times, the ultrasound signal of each array cell is measured five times and these signals are averaged to form a new signal for the given array cell. Study design A total of 60 volunteers participated in this study. All gave written and verbal informed consent. The volunteers (55.4 6 17.3 y; 30 females, 30 males) were measured in Odense in the context of collaboration within an Interreg IVa project of the European Community (‘‘cross boundary improvement of the situation of osteoporosis patients’’). A subgroup of 20 participants (51.7 6 20.0 y; 12 females, 8 males) was measured again after 1 mo to obtain mid-term precision. Short-term precision was calculated for all volunteers. Half of the subjects were measured first with the FUS, and the other half were measured first with the Achilles in alternating sequence. The study was approved by the Ethics Committee of Southern Denmark. Measurement procedure Achilles Insight. The default setting of the Achilles device is a mode that reports only the QUS stiffness index. To access the SOS values, another measurement mode was selected (BUA/SOS mode). The values for SOS, BUA and SI were used to determine the precision error. For each volunteer, two subsequent measurements with repositioning were performed. Before the foot was placed in the device, the skin was wet with ethanol spray. To stir the water for improved temperature control, a quality control measurement was performed before each set of measurements. This was a pure water measurement where the membranes were inflated to touch each other, with no foot in the device. Foot ultrasound scanner. Ultrasound transmission through the right foot was measured with beam directions varied across a range of different tilting and rotation angles. The tilting angles were identical for each measurement and ranged from 18 to –8 considering the horizontal axis as zero position. Measurements were per-
861
formed at five tilting angles with a step size of 4 . The rotation angles were in a range of 30 with a step size of 6 , resulting in measurements of six angles. The start angle for the rotation was adjusted to the medial side of the foot for each tilting angle and, therefore, depended on the size and shape of the foot. The varied tilting and rotation angles resulted in up to 30 different incident beam directions. A temperature sensor was attached to the right foot close to a vein near the ankle on the lateral side. A piece of neoprene was placed over the temperature sensor to isolate it against the warm air inside the device. Afterward, the foot was inserted into the device with the lateral side of the foot and the heel closely aligned with the foot holder as depicted in Figure 3. For each tilting angle, the starting rotation angle was determined individually so as to ensure that for this angle the emitter (always on the medial side, see Fig. 3) was as close to the medial side of the foot as possible. Afterward, the foot was wet with ethanol spray and the membranes were inflated to a pressure of 45 mbar above atmospheric pressure to enable contact with the skin of the foot. The rotation and tilting took approximately 2 min in total. At each incident beam angle, five data sets were recorded. For each cell, five ultrasound signals were averaged. After the membranes were drained, the foot was removed from the device and inserted and wet with ethanol spray again. Then the next measurement was performed. During measurements, the temperatures of the air, oil and foot were logged, as was oil pressure. Two measurements were performed on each volunteer. Before the measurements, a complete scanning routine was performed (inflation of the membranes, rotation and tilting, deflation of the membranes) to homogenize the oil and to allow the foot to adapt to the air and oil temperatures. Analysis Achilles Insight. The values of BUA, SI and SOS were internally calculated by the device in the standard manner. These data were then used to estimate the precision errors of SOS, BUA and SI. Foot ultrasound scanner. Calculation of the SOS. The TOF of the first-arriving signal was defined as the interval between the time at which the pulse was sent and the time of the first zero crossing with a negative slope (Fig. 4). Apparent SOS in the bone was calculated using the formula (Laugier 2008) SOS 5
d d 1TOFb 2TOFref SOSref
(2)
where d is the assumed thickness of the bone (30 mm), SOSref is the SOS in castor oil (1448 m/s at 33 C,
862
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 4. A typical ultrasound signal as received with the array in the region of interest. To calculate the speed of sound, we used TOF, which is the time of flight between excitation at the emitter and the first zero crossing with a negative slope in the signal received. The amplitude evaluated is the amplitude of the first oscillation as indicated in the figure and is used for defining the region of interest. The period length and the durations of the first (positive) and second (negative) halfwaves determine whether the signal is valid or not.
technical information from Karl Deutsch, Wuppertal, Germany), TOFref is the calculated reference TOF in castor oil (82.3 ms for a distance of 120 mm) and TOFb is the measured TOF with the foot placed in the ultrasound pathway. Evaluation of the measurement. Signal analysis requires the application of several measures which are first outlined and then described in detail. An overview of the steps is provided in Appendix B. Signals of a fixed rectangular region in the center of the calcaneus containing 12 cells (24 mm 3 18 mm) were evaluated. Only signals that met preset criteria were considered to be valid. Data sets with valid signals from fewer than two cells were excluded from the analysis. Because this was the case for a majority of the measurements within a specific range of the angle pairs of rotation and tilting, we excluded these angle combinations completely from the evaluation, leaving 15 angle pairs as valid. The angles excluded were 18 and –8 for tilting and the angle farthest from the start angle for rotation. Some features of the preset criteria were adjusted to generate an individual filter for each measurement. Therefore, averaged features of the signals of all valid cells across all valid angles were calculated for a second step of filtering. Then, the cell with the highest weighted signal amplitude of the first oscillation (see Fig. 4, later referred to as ‘‘amplitude’’), weighted by its count of valid signals, was selected for the calculation of SOS in all angles. The final SOS value is the lowest of these values.
In a QUS image of the calcaneus measured in the mediolateral direction, the signal amplitude is always higher in the center of the bone than at its fringes. Figure 5 illustrates an example of such an amplitude image. A predefined fixed region in the array comprising 12 cells (see Fig. 5, black dashed rectangle, and Appendix B, step 2) containing signals transmitted through the central bone region was further analyzed to identify the ROI (comprising a single cell). A filter for analyzing the signal features (see Appendix C) was applied to this region, excluding cells in which signal properties differ strongly from the properties of a typical signal in the valid bone area (see Appendix B, step 3). The latter was defined as a signal with features within a predefined range. Features of the filter included SOS, the period length of the first oscillation, the duration of the first (positive) half-wave and the difference between the durations of the first (positive) and second (negative) half-waves, all visualized in Figure 4. Their values are listed in Appendix C. Signals matching these filter specifications were denoted ‘‘common valid.’’ Because of the variability in properties of the heel bones between subjects, new filter settings were adjusted individually for each subject in a second step. 1. For the 15 valid pairs of angles, the average period length of all common valid signals was calculated, leading to new filter settings for ‘‘period length’’ and ‘‘difference between the durations of first and second half-waves’’ (see Appendix B, steps 4–6). Signals
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
863
Fig. 6. Speed of sound (SOS) values of the region of interest plotted over the rotation and tilting angles, depicted as isocontour lines. A brighter gray indicates a lower SOS. The interval between the iso-contour lines is 2 m/s. Fig. 5. Iso-contour line image (0.2-V difference between lines) illustrating the distribution of signal amplitudes across the ultrasound array. The higher the amplitude, the brighter is the color. The shape of the heel is outlined by the white line. The predefined region is used by the automated algorithm which searches for the region of interest, here outlined by a black dashed square.
matching these filter specifications were denoted ‘‘specific valid.’’ 2. The algorithm calculated the mean amplitude for each cell in the predefined region of all specific valid signals for the 15 valid pairs of angles (see Appendix B, steps 7–9). For each cell, the number of valid signals was used to calculate a weighted value of the mean amplitude. The cells and their weighted mean amplitude defined the output of the algorithm (see Appendix B, step 10). 3. For the cell with the highest weighted amplitude (preliminary ROI), SOS values were calculated for each valid angle (see Appendix B, step 11). The lowest SOS value within the range of the angles applied was taken as the result of the measurements (see Appendix B, step 12). Figure 6 depicts an example of iso-lines of SOS results plotted over the angles with the rotation angles on the x-axis and the tilting angles on the y-axis clearly showing the minimum in SOS. If the signal of the SOS minimum was not specifically valid, the SOS values of the cell with the second highest weighted average amplitude were used for evaluation, and so on. The ROI was saved for each volunteer and used for all follow-up measurements. If the SOS minimum of the ROI in the follow-up measurements was not common valid, the algorithm was applied to the measurements to obtain a new ROI.
Estimation of the influence of rotation and tilting on precision. In the subgroup of 20 subjects remeasured after 1 mo, the influence of rotation and tilting movements on short- and mid-term precision was estimated. In addition to the measurement and evaluation procedure described, the signal at the angle most similar to the Achilles beam angle, later referred to as ‘‘fixed angle,’’ was evaluated as well. The ROI was determined as described above but only for the one fixed angle. In this ROI, SOS was calculated according to formula (2). Short- and mid-term precision data were also calculated for this specific angle. Estimation of the influence of oil temperature and pressure. To verify if variations in oil temperature and pressure had an effect on the measurements, both values were included in a multivariate model to estimate the influence on SOS. Oil temperature was measured during inflation and draining of the membranes. The oil pressure was logged during the measurement for each beam angle. Determining the influence of the foot temperature. The impact of foot temperature on the SOS values was calculated using a multivariate model (multiple linear regression), with SOS as the dependent variable and foot temperature and the IDs of the volunteers as independent variables. The foot temperature coefficient for SOS (kfoot in m/s/ C) of the model was used to calculate a corrected SOS (SOScorr in m/s) according to the formula SOScorr 5 SOS1 wmean 2wfoot kfoot (3) using the original SOS (in m/s), the mean foot temperature of the population ymean (in C) and the actual foot
864
Ultrasound in Medicine and Biology
Table 1. Age-dependent slope per year and short- and mid-term precision errors of the foot ultrasound scanner and the Achilles Insight in absolute values Achilles Insight
Foot ultrasound scanner
SOS BUA (m/s) (dB/MHz) SI
SOS SOScorr* (m/s) (m/s)
Age-dependent slope (y21) –1.9 RMSE short-term (n 5 100) 3.9 RMSE mid-term (n 5 20) 5.9
–0.9 3.0 2.1
–1.1 –2.6 1.7 2.5 2.3 7.1
–2.7 2.8 4.6
BUA 5 broadband ultrasound attenuation; RMSE 5 root mean square error; SOS 5 speed of sound. * SOScorr is corrected for the impact of foot temperature.
temperature yfoot (in C). For these new SOS values, short- and mid-term precision errors were calculated as well. Statistical calculations were performed using JMP (SAS Institute, Cary, NC, USA) and Excel (Microsoft, Redmond, WA, USA). Both devices. Determination of short- and midterm precision. The precision error (root mean square error [RMSE]) was defined as the root mean square of the standard deviations (SD) of repeated measurements. For short-term precision, the SD was determined for two successively performed measurements of each volunteer obtained on a single day. For mid-term precision, the SD was derived from the mean values of the two successive measurements for a given volunteer obtained on different days. Normalization and comparison. In comparison of the precision errors of two different techniques or devices, an appropriate method for their normalization is crucial. Absolute precision errors or those relative to the absolute value are not comparable because they do not consider responsiveness and, thus, are not appropriate as a measure of the longitudinal sensitivity (i.e., ability to detect small changes) of the device or technique, which is the relevant measure for characterizing longitudinal precision (Gl€ uer et al. 1995). To adjust for differences in responsiveness, Gl€ uer (1999) suggested use of the response ratio rr (A vs. R) for standardization purposes. The response ratio is calculated from the response rates of the two parameters to be compared:
Volume 41, Number 3, 2015
Our normalization method is based on the response rate of the linear age-dependent decrease in the measured parameters (SOS, BUA and SI) of the females older than 49. The reference response rate R is the slope of the regression line of the Achilles SOS versus age (slSOS_Ach). The response rate A is the slope of the age-dependent decrease in the parameter to be standardized (slMP). The ratio of these two slopes rr (A vs. R) was then used to calculate the normalized precision errors (RMSEnorm) scaled to the responsiveness of the Achilles SOS: RMSEnorm 5 RMSE rrðA vs: RÞ 5 RMSE
The average measured values of the first two subsequent measurements were used to calculate the slopes. Because precision errors do not follow a Gaussian distribution (tested with a goodness-of-fit test), the nonparametric van der Waerden test was applied to test the significance of the differences between precision values of the FUS (SOS) and Achilles (SOS, BUA, SI). Whether improvements in SOS precision by rotation and tilting were significant was also tested with the van der Waerden test. RESULTS Achilles Insight One volunteer had to be excluded from the midterm follow-up because the Achilles Insight was not able to measure two SOS values with a difference less than 100 m/s or aborted the measurements. The volunteer could be measured with the FUS without problems. The absolute precision errors are listed in Table 1, and the normalized precision errors of BUA and SI are in Table 2. SI had the smallest precision error when normalized (3.0 m/s for short-term and 4.0 m/s Table 2. Short- and mid-term precision errors in SOS measured with the foot ultrasound scanner and parameters of the Achilles Insight normalized to SOS of the Achilles in meters per second RMSEnorm (m/s) Achilles Insight
rrðA vs: RÞ 5 response rate R=response rate A
(4)
To compare the SOS precision errors of the two devices and to investigate how BUA and SI performed compared with SOS, we also expressed the errors of our device and the errors of BUA and SI of the Achilles in recalculated units of the Achilles SOS precision error.
slSOS Ach slMP (5)
Short-term (n 5 100) Mid-term (n 5 20)
Foot ultrasound scanner
SOS
From BUA
From SI
From SOS
From SOScorr
3.9 5.9
6.3 4.4
2.9 4.0
1.8 5.2
2.0 3.2
BUA 5 broadband ultrasound attenuation; RMSE 5 root mean square error; SOS 5 speed of sound.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
Fig. 7. Decrease in speed of sound (SOS) values with age for women older than 49. Illustrated here are the SOS values of the Achilles (p , 0.01) and the SOScorr values of the foot ultrasound scanner (p , 0.002). FUS 5 foot ultrasound scanner, SOS 5 speed of sound.
for mid-term precision); these values are lower than those of SOS (3.9 m/s for short-term and 5.9 m/s for mid-term precision) and normalized BUA (6.3 m/s for short-term and 4.4 m/s for mid-term precision). Slopes of age dependence (Table 1) were –1.9 6 0.6 m/s/y (r2 5 0.27, p , 0.01) for SOS (Fig. 7), –0.9 6 0.4 dB/MHz/y (r2 5 0.18, p , 0.04) for BUA and 1.1 6 0.4/y (r2 5 0.23, p , 0.02) for SI. Foot ultrasound scanner One volunteer had no common valid SOS minimum for the mid-term follow-up measurement. The ROI had to be re-defined as described above. Two other volunteers had no specific valid SOS minimum at the baseline measurement. The cell with the next highest weighted average amplitude was chosen as the ROI. Absolute and normalized precision errors of SOS and SOScorr are listed in Tables 1 and 2. Oil temperature (stable at 32.4 6 0.4 C) and pressure had no significant effect on the SOS values, but foot temperature did influence SOS, with kfoot 5 – 4.9 6 0.6 m/s/ C (p , 0.0001, r2 5 0.43). The shortterm precision error increased slightly when the foot temperature correction was applied (2.8 m/s vs. 2.5 m/s), but the mid-term precision error was reduced from 7.1 to 4.6 m/s (p , 0.1, van der Waerden test). The age dependences were almost the same for the SOS and SOScorr values, with slopes of –2.6 6 0.7 m/s/y (r2 5 0.34, p , 0.003) and –2.7 6 0.7 m/s/y (r2 5 0.38, p , 0.002) (see Fig. 7), respectively. Comparison As illustrated in Figure 8, the SOS values of the FUS correlated well with the values of the Achilles Insight
865
(r2 5 0.74 and p , 0.0001 for SOS; r2 5 0.73 and p , 0.0001 for SOScorr). The linear age-related decrease in Achilles SOS and FUS SOScorr in females older than 49 y (n 5 25) is illustrated in Figure 7. The normalized precision errors reflecting the differences in age dependence are listed in Table 2. For short-term precision, the normalized precision error of FUS for SOScorr (2.0 m/s) was significantly lower than the Achilles SOS precision error (3.9 m/s) at p , 0.005 and its normalized BUA and SI precision errors (6.3 and 2.9 m/s) at p , 0.0001 and p , 0.03, respectively. For mid-term precision, the normalized precision error for FUS SOScorr (3.2 m/s) was significantly decreased (p , 0.05) compared with the Achilles SOS precision error (5.9 m/s) and a trend (0.09 , p , 0.14) compared with normalized BUA and SI precision errors (4.4 m/s respectively 4.0 m/s). The FUS SOScorr shortand mid-term precision errors were almost halved compared with the Achilles SOS precision errors (2.0 m/s vs. 3.9 m/s for short-term and 3.2 m/s vs. 5.9 m/s for mid-term). Influence of rotation and tilting on precision errors The short- and mid-term SOS precision errors of the FUS for four different methods are illustrated in Table 3. On the right-hand side are the precision errors for the procedure in which the lowest SOS was determined across all angles (variable angle). On the left-hand side are the precision errors for the fixed angle, that is, the angle close to the Achilles incident beam angle (angle fixed). For both methods, errors are given without and with foot temperature correction. The influence of foot temperature on the
Fig. 8. Correlation between the SOS values of the Achilles and the SOScorr values (SOS corrected for foot temperature) of the FUS (p , 0.0001). FUS 5 foot ultrasound scanner; RMS 5 root mean square; SOS 5 speed of sound.
866
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Table 3. Short- and mid-term precision errors in SOS measured with the foot ultrasound scanner for two different angles (n 5 20) Standard deviation (m/s) Angle fixed
Short-term Mid-term
Angle variable
Uncorrected for foot temperature
Corrected for foot temperature
Uncorrected for foot temperature
Corrected for foot temperature
4.4 9.0
4.5 6.8
2.7* 7.1
3.0 4.6*,y,z
* Trend (p , 0.1), improvement by rotation/tilting movement. y Trend (p , 0.1), improvement by consideration of foot temperature. z Significant (p , 0.05), improvement by considering both foot temperature and rotation/tilting movement.
SOS values was kfoot_var 5 –5.1 6 0.7 m/s/ C (p , 0.0001) for the variable angle method and kfoot_fix 5 –5.8 6 1.0 m/s/ C (p , 0.0001) for the fixed angle method. The foot temperature correction alone yielded a borderline significant improvement for midterm precision of the variable angle method by 35% (p , 0.09). Rotation and tilting improved results with borderline significance: short-term precision by 39% (p , 0.1) when applied without temperature correction and mid-term precision by 32% (p , 0.12) when foot temperature correction was applied. When the influence of both foot temperature and rotation/tilting were considered, we achieved a not significant improvement for the short-term precision error of 32% (p , 0.17) and a significant improvement of 49% (p , 0.02) for the mid-term precision error. All other combinations not mentioned were not significant. DISCUSSION Using an advanced method including patientspecific definition of a ROI in the ultrasound image, adaption of the beam incidence angle and adjustment for foot temperature, we were able to improve the SOS precision of calcaneal ultrasound measurements. The method included an automated algorithm to identify the ROI (cell with highest weighted transmission amplitude) at the baseline measurement, keeping this ROI constant across visits provided it exhibited valid signals and searching for an SOS minimum over the rotation and tilting angles. The correlation with the Achilles was quite good (r2 5 0.73); however, it was not good enough to be used for standardization purposes. Instead, we normalized the precision values to those of the Achilles SOS by using the ratio of the age-dependent slopes. By use of this standardization method, precision errors of the FUS were almost halved compared with the Achilles SOS precision errors (2.0 m/s vs. 3.9 m/s for short-term and 3.2 vs. 5.9 m/s for mid-term precision). The regression coefficients of the age-dependent slopes found in
this study were in the same order as those reported by other studies. Hans et al. (1995) reported a regression coefficient of r 5 –0.55 resulting in r2 5 0.3, whereas Schott et al. (1993) noted r 5 –0.63 for postmenopausal women resulting in r2 5 0.4. The SOS, BUA and SI values of the Achilles had smaller regression coefficients of r2 ranging between 0.18 and 0.27; the SOS of the FUS had a regression coefficient comparable to those reported, with r2 5 0.34 and r2 5 0.38 when corrected for foot temperature. Age dependence was significant with 0.002 , p , 0.04, even though the group was relatively small with only 25 subjects. Interestingly, in women older than 49 y, age-dependent decreases in the SOS of the FUS (RMSEnorm 5 22–23 m/s) scattered less than all normalized variables of the Achilles (RMSEnorm 5 27–34 m/s). This might indicate that the FUS SOS is less affected by error sources than the Achilles parameters. Whether this affects the sensitivity of the device, however, would have to be determined in a longitudinal study. One source of error not accounted for by any of the devices was the true thickness of the bone measured. For the calculation of SOS, the thickness is assumed to be fixed: For the FUS at 30 mm, and for the Achilles at 40 mm (Njeh et al. 1999). In principle, the bone width may be estimated using signals reflected from the bone surface. However, the spatial accuracy is limited by the large wavelength of the ultrasound wave of about 3 mm in soft tissue. Miller et al. (1993) reported that there is no improvement in SOS precision error when a true instead of an apparent SOS is calculated. Moreover, correct estimation of the thickness and calculation of a true SOS value are not crucial in monitoring therapy, because only a precise assessment of changes between measurements at different points in time is required and the width of the bone can be expected to remain constant. The coefficient of the foot temperature was of the expected order of magnitude. Rajagopalan et al. (1979) reported a temperature coefficient for fat (human female breast) of –3.1 m/s/ C in the temperature range 22 C–37 C. In our study, foot temperatures were in the range 26.8 C–34.6 C. The coefficient determined with
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
our model ranged between –4.9 and –5.8 m/s/ C. Because the ultrasound pathway comprises subcutaneous fat and bone marrow and because temperature was measured at the skin and not inside the heel, no better agreement can be expected. An extended temperature measurement (e.g., at different locations on the foot) and monitoring of temperature dynamics might help to further improve the adjustment for temperature variations beyond the reduction in mid-term precision errors of 35% achieved with our method. Keeping the ROI constant across baseline and follow-up measurements is suboptimal but seems to be the best method for the current FUS device. Through use of this approach, repositioning errors will not be compensated if the ROI exhibits valid signals. On the other hand, with our method, possible anisotropic changes in bone structure which might lead to undesired changes in ROI positioning do not distort our measurements because the region is kept constant. One way of obtaining a constant ROI in the bone in the future might be to use an ultrasound array with a higher spatial resolution and to determine the ROI via the outline of the bone and not on the basis of transmission amplitude. The added feature of rotating and tilting the incident beam angle had positive effects on short-term and midterm precision, but with only borderline significance. This is not surprising given that only 20 subjects were included in this analysis. Foot temperature correction and tilting/rotation contributed almost equally to the precision errors. The foot temperature correction resulted in a 35% decrease in the mid-term precision error, and tilting/rotation decreased short- and mid-term precision errors between 32% and 39%. The combination of both had a significant effect on the mid-term precision error, which was reduced by 49%. Several improvements regarding the mechanical design of our scanner are conceivable. The mechanics to adjust the rotation angle were not optimal. Tolerance was relatively high at 0.7 considering the step size of 6 . Nevertheless, this should be a minor problem near the SOS minimum, where only slight changes in SOS (Fig. 6) were observed. For 52% of the subjects, the SOS minimum found was at the limit of the chosen range of angles. In 76% of them, it was close to the starting angle for rotation. The size of the transducers and their housings did not permit larger rotations. Thus, transducers and their housings should be kept small to enable a better starting angle and a higher probability of finding a SOS minimum. Mahrt et al. (1991) reported that with sufficient effort, a short-term SOS precision error of 0.1 m/s can be achieved. They measured the heel of a volunteer in a water bath until the foot reached thermal equilibrium and searched for a SOS minimum by scanning five
867
coordinates. The cited precision of 0.1 m/s is comparable to a value of 0.45 m/s taking into account a different calculation formula used in the Achilles Insight and the FUS. Chappard et al. (1999) reported a decrease in SOS of around 7.4 m/s during the first 7 min of immersion in a tempered water bath for postmenopausal women. This effect was smaller for premenopausal women (around 3.0 m/s) and men (increase of 1.5 m/s). We aimed to minimize the drifting effect by performing a complete ‘‘dummy’’ measurement before the first measurement, which, together with adjustment of the starting rotation angles, added about 7 min to the examination time. However, monitoring foot temperature dynamics might lead to further improvement. In this study, we focused on an optimized assessment of SOS rather than BUA or a combined index like SI. We used an array with a small bandwidth of 10%, resulting in a frequency range of 0.45–0.55 MHz, which made it difficult to calculate BUA. The typical frequency range for which BUA is calculated is 0.2–0.6 MHz (Langton and Njeh 2008). Therefore, the BUA of the FUS may not exploit the full potential with respect to precision and, thus, we did not evaluate its performance. The focus on SOS was motivated by reports in the literature. Here, standardized precision errors of SOS measurements are generally smaller than those of BUA measurements. The FUS performed numerically better than BUA of the Achilles for all types of precision errors assessed. As mentioned in the Introduction, Gonnelli et al. (2002) reported a monitoring time interval for BUA that was five times higher compared with that for SOS, documenting the superior performance of SOS in monitoring bisphosphonate treatment. The parameter SI, which combines SOS and BUA, offers the advantage of more robust performance compared with a single measure, but because it is calculated from weighted averages of BUA and SOS, the poorer performance of BUA precision will have a negative impact on SI precision. Therefore, we decided to focus on SOS in this study on precision. The precision errors of the FUS were numerically smaller than the normalized SI precision errors of the Achilles Insight (Table 2). In a previous study, Gonnelli et al. (2002) reported that a change of 1% aBMD measured with DXA at the lumbar spine (which is a typical level for DXA precision errors) was associated with a change of 2.4 m/s in SOS measured at the calcaneus with an Achilles Plus device. Our device had a normalized mid-term precision of 3.2 m/s, which corresponds to a DXA error of 1.3% and, thus, would be close to the precision error of lumbar spine aBMD. In summary, the new QUS calcaneus device tested here exhibited significantly improved precision compared with the commercially distributed Achilles
868
Ultrasound in Medicine and Biology
Insight device. This improvement was achieved combining use of an array transducer as the receiver with consistent ROI definition, the selection of optimized transmission angles and an adjustment for variability in foot temperature. As a result, the normalized precision error was close to the level of precision achieved with DXA of the spine. Further improvements with more advanced technology, for example, transducers with higher spatial resolution and a better evaluation algorithm, appear to be feasible. Testing of the performance of our improved method for monitoring response to treatment appears warranted. Acknowledgments—We thank Jane Nielsen and Elizabeth Hanmann for recruiting the volunteers and performing the Achilles measurements at Odense University Hospital.—This work was partly funded by the Federal Ministry for Economic Affairs and Technology (grant number 03VWP0042) (Program ‘‘Signo’’) and the European Union in the framework of Interreg IVa for cross-border collaborations (Project 04-1.5-08).
REFERENCES Bauer DC, Ewing SK, Cauley JA, Ensrud KE, Cummings SR, Orwoll ES. Quantitative ultrasound predicts hip and non-spine fracture in men: The MrOS study. Osteoporos Int 2007;18:771–777. Blake GM, Fogelman I. The role of DXA bone density scans in the diagnosis and treatment of osteoporosis. Postgrad Med J 2007;83: 509–517. Center JR, Nguyen TV, Schneider D, Sambrook PN, Eisman JA. Mortality after all major types of osteoporotic fracture in men and women: An observational study. Lancet 1999;353:878–882. Chappard C, Berger G, Roux C, Laugier P. Ultrasound measurement on the calcaneus: Influence of immersion time and rotation of the foot. Osteoporos Int 1999;9:318–326. Chapurlat RD, Palermo L, Ramsay P, Cummings SR. Risk of fracture among women who lose bone density during treatment with alendronate. The Fracture Intervention Trial. Osteoporos Int 2005;16: 842–848. Faulkner KG. Bone densitometry; Choosing the proper skeletal site to measure. J Clin Densitom 1998;1:279–285. Frost ML, Blake GM, Fogelman I. Changes in QUS and BMD measurements with antiresorptive therapy: A two-year longitudinal study. Calcif Tissue Int 2001;69:138–146. Gl€uer C, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate assessment of precision errors: How to measure the reproducibility of bone densitometry techniques. Osteoporos Int 1995;5:262–270. Gl€ uer CC. Monitoring skeletal changes by radiologic techniques. J Bone Miner Res 1999;14:1952–1962. Gonnelli S, Cepollaro C, Montagnani A, Martini S, Gennari L, Mangeri M, Gennari C. Heel ultrasonography in monitoring alendronate therapy: A four-year longitudinal study. Osteoporos Int 2002; 13:415–421. Gonnelli S, Martini G, Caffarelli C, Salvadori S, Cadirni A, Montagnani A, Nuti R. Teriparatide’s effects on quantitative ultrasound parameters and bone density in women with established osteoporosis. Osteoporos Int 2006;17:1524–1531. Goossens L, Vanderoost J, Jaecques S, Boonen S, D’hooge J, Lauriks W, Van der Perre G. The correlation between the SOS in trabecular bone and stiffness and density studied by finite-element analysis. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1234–1242. Guglielmi G, Floriani I, Torri V, Li J, van Kuijk C, Genant HK, Lang TF. Effect of spinal degenerative changes on volumetric bone mineral density of the central skeleton as measured by quantitative computed tomography. Acta Radiol 2005;46:269–275. Hall S, Criddle R, Comito T, Prince R. A case–control study of quality of life and functional impairment in women with long–standing vertebral osteoporotic fracture. Osteoporos Int 1999;9:508–515.
Volume 41, Number 3, 2015 Hans D, Dargent-Molina P, Schott AM, Sebert JL, Cormier C, Kotzki PO, Delmas PD, Pouilles JM, Breart G, Meunier PJ. Ultrasonographic heel measurements to predict hip fracture in elderly women: The EPIDOS prospective study. Lancet 1996;348:511–514. Hans D, Schott A, Arlot M. Influence of anthropometric parameters on ultrasound measurements of Os calcis. Osteoporos Int 1995;5: 371–376. H€aussler B, Gothe H, G€ol D, Glaeske G, Pientka L, Felsenberg D. Epidemiology, treatment and costs of osteoporosis in Germany—The BoneEVA Study. Osteoporos Int 2007;18:77–84. Hodgskinson R, Njeh CF, Currey JD, Langton CM. The ability of ultrasound velocity to predict the stiffness of cancellous bone in vitro. Bone 1997;21:183–190. Jones G, Nguyen T, Sambrook P, Kelly PJ, Eisman JA. Progressive loss of bone in the femoral neck in elderly people: Longitudinal findings from the Dubbo osteoporosis epidemiology study. BMJ 1994;309: 691–695. Krieg MA, Barkmann R, Gonnelli S. Quantitative ultrasound in the management of osteoporosis: The 2007 ISCD Official Positions. J Clin Densitom 2008;11:163–187. Langton CM, Njeh CF. The measurement of broadband ultrasonic attenuation in cancellous bone—A review of the science and technology. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1546–1554. Langton CM, Palmer SB, Porter RW. The measurement of broadband ultrasonic attenuation in cancellous bone. Eng Med 1984;13:89–91. Laugier P. Instrumentation for in vivo ultrasonic characterization of bone strength. IEEE Trans Ultrason Ferroelectr Freq Control 2008;55:1179–1196. Mahrt KH, Barkmann R, Niedermayer W, Kroebel W. Extremely sensitive ultrasound scanning system for non-invasive in vivo detection of mineralisation changes in human heel bones. Proc IEEE Ultrason Symp 1991;1119–1122. Miller C, Herd R, Ramalingam T. Ultrasonic velocity measurements through the calcaneus: Which velocity should be measured? Osteoporos Int 1993;3:31–35. Nicholson PH, M€uller R, Lowet G, Cheng XG, Hildebrand T, R€uegsegger P, van der Perre G, Dequeker J, Boonen S. Do quantitative ultrasound measurements reflect structure independently of density in human vertebral cancellous bone? Bone 1998;23:425–431. Njeh CF, Hans D, Fuerst T, Gl€uer CC, Genant HK. Quantitative ultrasound: Assessment of osteoporosis and bone status. London: Dunitz, Martin; 1999. Pinheiro MM, Castro CM, Szejnfeld VL. Low femoral bone mineral density and quantitative ultrasound are risk factors for new osteoporotic fracture and total and cardiovascular mortality: A 5-year population-based study of Brazilian elderly women. J Gerontol A Biol Sci Med Sci 2006;61:196–203. Rajagopalan B, Greenleaf JF, Thomas PJ, Johnson SA, Bahn RC. Variation of acoustic speed with temperature in various excised human tissues studied by ultrasound computerized tomography. In: Ultrasonic tissue characterization II: A collection of reviewed papers based on talks presented at the Second International Symposium on Ultrasonic Tissue Characterization. Washington, DC: U.S. Government Printing Office; 1979. p. 227–233. Sahota O, San P, Cawte S. A comparison of the longitudinal changes in quantitative ultrasound with dual-energy X-ray absorptiometry: The four-year effects of hormone replacement therapy. Osteoporos Int 2000;11:52–58. Schott A, Hans D, Sornay-Rendu E, Delmas P, Meunier P. Ultrasound measurements on os calcis: Precision and age-related changes in a normal female population. Top Geriatr Rehabil 1993;8:7. Warming L, Hassager C, Christiansen C. Changes in bone mineral density with age in men and women: A longitudinal study. Osteoporos Int 2002;13:105–112. Watts NB, Cooper C, Lindsay R, Eastell R, Manhart MD, Barton IP, van Staa TP, Adachi JD. Relationship between changes in bone mineral density and vertebral fracture risk associated with risedronate: Greater increases in bone mineral density do not relate to greater decreases in fracture risk. J Clin Densitom 2004;7:255–261.
Improved precision of QUS heel measurements d M. DAUGSCHIES et al.
869
APPENDIX A Appendix A. Overview of the Achilles Insight and foot ultrasound scanner Achilles Insight Emitter Receiver
1 Array
Center frequency Bandwidth Coupling medium Ultrasound velocity at 33 C Tempered Parameters measured Beam angle Region of interest
0.5 MHz Quarter wave-matched broadband elements Water 1516 m/s 33 C SOS, BUA, SI (combination of SOS and BUA) Fixed Fixed, 25-mm diameter
Duration of measurement Foot to measure Extras
30–60 s Both —
BUA 5 broadband ultrasound attenuation; SI 5 stiffness index; SOS 5 speed of sound.
Foot ultrasound scanner 1, diameter 5 100 mm Array (100-mm diameter, 100 cells evaluated, edge length 5 6 mm, 0.1-mm distance between cells) 0.5 MHz 10% Castor oil 1448 m/s 32.5 C SOS 6 8 tilting, 30 rotation (adjusted to foot) 1 array cell (6 3 6 mm), manually determined for each volunteer 2 min Right Foot temperature measurement used for adjusting SOS
870
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Appendix B. Algorithm for identifying the region of interest*
ROI 5 region of interest; SOS 5 speed of sound. * This schematic drawing illustrates the principle underlying the algorithm for identifying the region of interest. The algorithm is described in the text (see Evaluation of the Measurement in the Methods section). Briefly, a filter with various parameters is applied to a fixed region in the array. For all valid signals in the region for previously defined angles, an average period length is calculated that determines the parameters for a second filter applied to the same region. The signals fitting the filter conditions are used to calculate a weighted amplitude for each cell within the region. The highest weighted amplitude determines the region of interest for which all speed of sound values are plotted over the angles. The smallest speed of sound value is the measurement result.
Appendix C. Filter parameters for automatic algorithm Speed of sound (m/s) Period length of first oscillation (ms) Duration of first positive half-wave (ms)
Difference between first positive and negative half-waves (ms)
1450–1800 1450–1800
1 4
2.1–3.5 PLall 6 0.5
1.0–2.0 1.0–2.0
,0.7 PLall