Ultrasound in Med. & Biol., Vol. 41, No. 3, pp. 748–759, 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.2014.11.016
d
Original Contribution QUANTITATIVE SONOGRAPHY OF BASAL CELL CARCINOMA HANNA PIOTRZKOWSKA-WROBLEWSKA,* JERZY LITNIEWSKI,* ELZBIETA SZYMANSKA,y and ANDRZEJ NOWICKI* * Department of Ultrasound, Institute of Fundamental Technological Research, Warsaw, Poland; and y Dermatology Clinic, CSK MSWiA Hospital, Warsaw, Poland (Received 16 January 2014; revised 29 October 2014; in final form 23 November 2014)
Abstract—A 30-MHz ultrasonic scanner was used to collect B-scan images together with appropriate radiofrequency echoes from diseased and healthy skin regions of patients with diagnosed basal cell carcinoma and pre-cancerous lesions (actinic keratosis). Radiofrequency data were processed to obtain the attenuation coefficient and statistics of the backscattered echo signal determination (K-distribution and effective density of scatterers [EDS]). The attenuation coefficient was significantly higher for patients with basal cell carcinoma than for healthy patients. Also, the pre-cancerous skin lesions had increased attenuation. The averaged EDS values for cancer lesions were significantly lower than those for pre-cancerous lesions and healthy skin. The successful differentiation between the tissue groups examined suggests the potential value of the attenuation coefficient and EDS for carcinoma characterization. (E-mail: [email protected]) Ó 2015 World Federation for Ultrasound in Medicine & Biology. Key Words: Quantitative ultrasound, High frequency, Human skin, Skin lesions, K-distribution, Attenuation coefficient, Tissue characterization.
Foster et al. (2000) reported that although ultrasound (US) has the capability to image fine features in the skin, such as sweat gland ducts, hair follicles and veins, the diagnostic ability of skin images to reveal specific pathologies is limited. It is difficult to differentiate between benign and malignant lesions using B-scan images (Fornage et al. 1993) because both types of lesions appear hypo-echogenic compared with healthy skin. Another study indicated that both scar tissue and malignant melanoma could appear similar in US scans (Turnbull et al. 1995). Thus, biopsy is still the gold standard in final diagnosis of skin cancer; however, quantitative ultrasound can provide additional information that is potentially helpful in lesion assessment and screening tests. First, quantitative studies of skin lesions in vivo using ultrasound have mostly been limited to parameters that could be computed directly from images, recorded by using commercially available 20-MHz systems. For instance, changes in skin echogenicity (mean pixel amplitude, which is proportional to the mean backscatter amplitude) and skin thickness have been found to be related to photo-aging of the skin (Gniadecka and Jemec, 1998). The degree of acoustic shadowing, measured as the ratio of echogenicity of the retrocessional dermis to that of the perilesional dermis, was found to be
INTRODUCTION Basal cell carcinoma (BCC) is the most common cutaneous malignancy, representing 80% of all skin cancer cases. BCC arises from the basal cells of the epidermis. Although BCC is not associated with significant mortality, the associated morbidity and therapeutic costs are an increasing burden to the health care system. BCC cases are rather easily diagnosed, but a substantial number of cases require biopsy because of their similarity to other cutaneous neoplasms, both benign and malignant, making the BCC direct diagnosis ambiguous. Basal cell carcinoma can develop from precancerous growths like an actinic keratosis (AK), as well as from unchanged skin. According to Cham (2013), it is estimated that up to 50% of the population is affected by AK. The difficulty in predicting the evolution of this kind of lesion makes AK very dangerous. It may not be a threat to the health and life of the patient for a long period, but at some point it can vanish or transform into cancer. Address correspondence to: Hanna Piotrzkowska-Wroblewska, Department of Ultrasound, Institute of Fundamental Technological Research, ul. Pawinskiego 5 b, 02-106 Warsaw, Poland. E-mail: [email protected] 748
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
capable of differentiating between basal cell papilloma and malignant melanoma (Harland et al. 2000). Other parameters that require analysis of radiofrequency (RF) backscattered echoes, such as the attenuation coefficient (Guittet et al. 1999) and the apparent integrated backscatter (Fournier et al. 2001), have also been studied in healthy skin tissues in vivo. Although the statistics of ultrasonic signals have been extensively studied with respect to their potential to classify tissue types (Shankar et al. 1993, 2000; Tsui et al. 2010), such studies have not been widely carried out in skin tissue. Raju and Srinivasan (2001, 2002) studied in vivo the frequency-dependent attenuation and backscatter coefficient, as well as parameters related to echo statistics, such as the ratio of mean to standard deviation, denoted as signal-to-noise ratio (SNR), and the parameters of envelope probability density functions (PDFs) of healthy skin tissues. They found that the generalized gamma, K- and Weibull distributions modeled the envelope statistics well, whereas the Rayleigh distribution provided a rather poor fit. The use of quantitative ultrasound in the monitoring of skin tissue affected by basal cell carcinoma was proposed by Petrella et al. (2012). Their study was performed ex vivo using an ultrasound biomicroscope working at a frequency of 45 MHz. They examined SNR and shape parameters of Weibull and generalized gamma PDFs. The significant differences between lesions with various distribution patterns of tumor nests suggested that quantitative ultrasound has potential for use in carcinoma characterization. Similar results were presented in our previous study. We found that the K-distribution is appropriate for modeling statistics of the envelope of signals obtained from human skin in vivo, and the value of the shape parameter of the K-distribution can help in differentiation of healthy skin and skin affected by BCC (Piotrzkowska et al. 2012). The goal of this study was to find the quantitative measure of skin tissue backscattering properties for differentiating changes in tissue structure induced by BCC and AK. The new method allows non-invasive differentiation of the disease entities and, consequently, selection of the proper method of treatment. In the case of BCC, the most common method of treatment is surgical removal of lesions with the correspondingly large margin or, in some cases, cryotherapy or laser therapy. Actinic keratosis requires conservative treatment, locally (imiquimod or isotretinoin) or with cryotherapy or laser therapy. Because the recorded signals are dependent on both the tissue properties and the instrument used in recording the signals, methods for computing quantitative parameters, such as the attenuation coefficient and the shape parameter of the K-distribution, were investigated,
749
taking into account the compensation of the influence of the system-dependent effects. STRUCTURE OF HUMAN SKIN AND BBC LESIONS Skin is a multilayered organ, with the separate histologic layers epidermis, dermis and subcutaneous tissue. The epidermis is the thinnest layer of skin. The main cell populations of the epidermis are keratinocytes, melanocytes and Langerhans cells. The middle layer of skin is the dermis, which consists of connective tissue arranged in two layers: the more superficial papillary dermis and the deeper reticular dermis. Thin papillary dermis is made of loose connective tissue containing mostly collagen fibers. In the thicker reticular dermis, connective tissue is denser because of the coarse bundles of collagen fibers arranged in layers parallel to the skin surface. There are other structures in the dermis, such as blood vessels, lymphatic vessels, nerve fibers, portions of hair follicles and sweat glands; however; their volume percentage is much lower than that of collagen fibers. All dermal components are bound together by a gel-like ground substance made of various glyosaminoglycans and glycoproteins. The third skin layer is subcutaneous fat, further divided into lobules containing adipose cells and separated by thin fibrovascular septa. The septa consist of collagen and reticulin fibers, blood and lymphatic vessels and cutaneous nerves. Basal cell carcinoma tumors are typically characterized by a fibrous stroma surrounding clusters of tumor cells that resemble collagen fibers of the basal layer of the epidermis. Usually, these tumor cells are fairly regular with rounded nuclei and little cytoplasm. The level of skin changes (percentage of pathologic structures in the volume of tissue) reflects the progression of the lesion. In the initial phase of the disease, the number of cancerous cells is relatively low; however, in the advanced stage of cancer, the number and size of clusters increase, which reduces the proportion of collagen fibers in the dermis to about 30%. In 30-MHz ultrasonic images, all three layers of skin can be observed. BCC lesions are usually located in the dermis, and all results discussed in this article concern this layer. The echogenicity in images of the dermis is determined by a collagen component (Wortsman and Jemec 2013). Collagen fibers are 5–15 mm thick with an average density of 1250 kg/m3 and longitudinal wave sound speed of 1730 m/s (Weber et al. 1984) Thus, the resultant impedance is much higher than the impedance of BCC cells. We assume that the replacement of fibers by tumor cells leads to changes in tissue attenuation and backscatter statistics.
750
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
METHODS Acquisition system The experimental 30-MHz ultrasonic scanner and data acquisition procedure are described in detail elsewhere (Lewandowski and Nowicki 2008), and only a brief description is given here. Ultrasonic waves were generated by the 20-mm-thick spherical transducer made of the modified thick film piezoelectric lead zirconate titanate (PZT) 37 deposited on the PZT substrate (Ferroperm, Vedbæk, Denmark). Properties of the thick film transducer and the generated acoustic field are given in Table 1. The transverse pressure distribution of the probing pulse measured in focus using the needle hydrophone (Precision Acoustics, UK) is illustrated in Figure 1. The transducer was mounted in the mechanical, wobbler type-probe generating a sector scan with the 10-Hz frame rate. To increase the SNR, we used the 32bit Golay-encoded transmit sequences and compression of received echoes (Litniewski et al. 2007). Ultrasonic data were converted from analogue to digital using a 200-MHz, 12-bit resolution sampling system. The computer was used to control mechanical scanning and data acquisition. Received sequences were first compressed and then envelope detected and displayed. The compressed RF data were stored with a time gain control (TGC) curve that was used to increase the quality of skin images. One image corresponded to 200 RF echo lines, each 2048 samples long. Once the RF data were collected, further processing and analysis were done off-line on the computer using Mathcad (Mathsoft, Cambridge, MA, USA) software. Quantitative parameters Attenuation coefficient. The attenuation coefficient aðf Þ of the skin was determined using the spectral difference technique, which is based on comparison of the power spectra of backscattered signals recorded at two different depths and denoted as SAi ðf Þ and SBi ðf Þ, respectively, where i represents the number of echo lines considered, and f denotes frequency. Only the RF data from selected regions of interest (ROIs) were used. The procedure was as follows: First, from the selected part
Fig. 1. (a) Transducer pressure distribution in the focal plane. (b) Shape of the scanning pulse measured in the focus.
of the scattered echo signal in the focal zone, two segments, 0.18 ms long and separated by distance Dz 5 1 mm, were weighted using a Hanning window. Next, these segments were Fourier transformed, and the power spectra were calculated. The spectra from the same depth were averaged over all echo lines of the
Table 1. Properties of the transducer and generated acoustic field Center frequency
30 MHz
26-dB bandwidth F-number Diameter Focal length Depth of focus (26 dB) Axial resolution Lateral resolution in focus
14 MHz 2.9 3 mm 9 mm 2 mm 0.07 mm 0.14 mm
Fig. 2. Plots of the K-distribution probability density functions for various values of effective density of scatterers (EDS).
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
751
Fig. 3. B-Mode images of healthy human dermis (a), pre-cancerous skin (b) and basal cell carcinoma (c) within the indicated regions of interest.
ROI (not ,25 lines), yielding two averaged spectra SB ðf Þ and SA ðf Þ, and the attenuation aðf Þ was estimated from their log ratio. The longitudinal wave velocity of 1540 m/s was assumed (Wells 1977). The formula used to determine the attenuation was. B ðf ;zÞ 10 log SSAB ðfðf Þ$corr Þ$corrA ðf ;zÞ (1) aðf Þ 5 2 4Dz where P P SA ðf Þ5ð1=NÞ Ni5 1 SAi ðf Þ and SB ðf Þ5 ð1=NÞ Ni5 1 SBi ðf Þ. N is the number of echo lines in the ROI, and corrB ðf Þ and corrA ðf Þ denote diffraction correction coefficients dependent on frequency and distance from the focus (see Fig. 5).
The attenuation coefficient was then calculated from the slope of a least-squares linear fit of attenuation versus frequency dependence. Statistics of echo envelope. The statistics of echo signals backscattered by a medium depends on both medium structure and spatial resolution of the scanning system, which is defined by the size of the resolution cell. This size varies with pulse length and acoustic beam cross section. The soft tissue is often modeled as a collection of spatially distributed small scatterers (Wagner et al. 1983, 1987). It can be illustrated that for the large number of scatterers within the resolution cell, the signal amplitude statistics follow the Rayleigh distribution. But in biological tissue, the spatial distribution of scattering
Fig. 4. Normalized amplitudes of three selected frequency components (15, 25 and 35 MHz) of the reflected pulse power spectrum over a distance of 2 mm on either side of the focus.
752
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 5. Diffraction correction curves, corrB(f, z) and corrA(f, z) in formula (1), measured 60.5 mm from the focus.
elements is not uniform. Also, the scatterer density and size vary throughout the tissue. Thus, the effective number of scatterers in the resolution cell may not fulfill the conditions for Rayleigh statistics, particularly when the resolution cell is small. In this case non-Rayleigh statistics are considered (Goodman 2000). Jakeman and Tough (1987) pointed out that for the low effective density of scatterers, the PDF of the backscatter amplitude is given by the K-distribution. More recently, it was reported that the K-distribution describes well the statistics of the skin backscatter (Raju and Srinivasan 2002). The K-distribution is defined by pðAÞ 5 2
EDS EDS11 A b KEDS21 bA 2 GðEDSÞ
(2)
where A is amplitude, pðAÞ is probability density, b 5 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi p 4EDS=E½A2 , Kb ð$Þ is the modified Bessel function of the second kind of order b, Gð$Þ is the standard gamma function, E[ ] is the expectation operator, EDS 5 Nð11vÞ is the effective density of scatterers within the resolution cell, N is the number of scatterers and v is a constant depending on scatterer characteristics (Dutt 1995). Figure 2 illustrates the probability density function of the K-distribution plotted for various values of EDS. It is easy to see that the K-distribution covers the range of distributions, approaching the Rayleigh distribution as EDS becomes large and becoming log normal as EDS / 0. It can be seen that, in practice, the K-distribution for EDS . 10 converges to the Rayleigh distribution. The effective density of scatterers estimated from the K-distribution model depends on the actual number of scattering sites per resolution cell, as well as the uniformity of backscatter coefficient. The type and distribution of scatterers are intrinsically related to the type of tissue through which the ultrasound beam is passing. Therefore, the EDS parameter can be used to distinguish
between regions differing in spatial density of scatterer or between regions of varying scatterer cross section and, as a consequence, can be used as a parameter for tissue characterization. The EDS can be estimated from the fourth normalized moment (r4) of the K-distribution (Weng et al. 1991): E½A 2 where r4 5 ½E½A EDS 5 r4 22 2 2 4
(3)
The start and endpoints of the signal acquisition and the number of lines (ROI) were first selected from the image. The moments E½An were directly computed from the envelopes of RF signals enclosed in the ROI. Next, r4 was determined, and the EDS parameter was calculated using eqn (3). To verify whether the K-distribution is a good model to describe statistics of the amplitude of skin backscatter when using our scanner, the histograms of experimental data were calculated and fitted by the K- and Rayleigh distributions. The results obtained for the skin of healthy volunteers, discussed in the Results, reveal the superiority of the K-distribution to the Rayleigh distribution. Correct estimation of the effective density of scatterers requires a large number of data samples. To find the size of ROIs that gives the ROI size-independent EDS parameter, we used RF data corresponding to the homogenous part of the skin. A set of concentric rectangular ROIs of varying length was selected in the focal zone. Data from each ROI were analyzed for their respective EDS parameter. It was found that a 1 3 1-mm ROI box, corresponding to a 1.3-ms-long time window and 25 adjacent scan lines, is sufficient to obtain reliable results. Acquisition of ultrasonic data from the skin Skin tissues were examined in vivo. A group of patients from the Dermatology Clinic with diagnosed
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
BCC and pre-cancerous lesions (AK) participated in this study (21 and 14 cases, respectively, sex and age unknown). The approval of the ethical review board of the Medical University of Warsaw and informed consent for the study from all patients were obtained before measurements. Examinations were carried out by medical doctors. Three types of data—traditional B-mode image, set of RF echo lines and TGC curve—were stored for the skin regions where BCC was diagnosed and where precancerous lesions were localized. For all patients, reference data for the selected healthy fragments of the skin were also recorded. Healthy skin RF samples were taken from regions positioned as close to the lesion as possible. In general, the distance between lesion and healthy skin reference depends on the extension of the lesion, but in our experience, spacing from 3 to 5 cm is sufficient; however, it was always decided by the examiner. Examples of B-mode images of healthy skin and skin with BCC and pre-cancerous lesions (AK) are provided in Figure 3. Data pre-processing for attenuation estimation Attenuation in the examined tissue results in the decay of wave amplitude with depth. The latter is partially compensated by the TGC system of the receiving amplifier, which is useful for B-mode imaging, but interferes with estimation of quantitative parameters. The influence of the TGC system on RF echoes was compensated before the processing that was carried out to determine the attenuation coefficient of the skin and the statistical properties of the skin backscatter. It is important to compensate for the effects of diffraction when using the focused transducer to determine tissue attenuation from the backscatter. Focusing introduces variation of the beam amplitude with distance and shifts the mean frequency of the ultrasonic pulse in the focal region. When the acoustic pulse is approaching the focus, the amplitudes of higher-frequency components increase faster than those of lower-frequency components. For the diverging beam, the opposite effect occurs. Thus, diffraction affects the measurements of spectra and, consequently, the estimated attenuation coefficient. To compensate for the diffraction effects, several approaches have been proposed. Chen et al. (1997) developed theoretical diffraction correction curves based on the physics of acoustic wave propagation, including transducer geometry and frequency characteristics. The empirical approach was presented by Raju and Srinivasan (2002). Correction curves were determined from the acoustic data collected from the same tissue that was placed at different distances from the focus. The diffraction correction term proposed by Bigelow et al. (2008) is based on the assumption that the field
753
pattern along the focal zone can be approximated by a Gaussian function. The Gaussian depth of focus as a function of wavelength was measured by acquiring echoes from a rigid plane located at different locations along the beam axis. In our study, we compensated for focusing by processing the spectra similarly to the method proposed by Bigelow et al. (2008); however, the correction coefficients were determined directly from the amplitude spectra of echoes obtained from a rigid plane reflector located in water at various axial distances from the focus. Figure 4 illustrates the amplitudes of three selected frequency components of the reflected pulse power spectrum over a distance of 2 mm on either side of the focus. It can be seen that the 15-MHz frequency component has the larger width, which results from the weaker focusing at low frequency. For each spectral component, the correction coefficients compensating the amplitude changes induced by focusing were calculated. Correction coefficient-versus-frequency curves are provided in Figure 5 for two selected distances from the focus. They equalize the amplitude of the frequency components obtained in water, which is assumed to be a non-attenuating medium. These coefficients were used in formula (1) to compensate the power spectra collected from patients. Data selection and pre-processing for statistical analysis Focusing varies the beam cross-sectional area along the path of the beam, which changes the volume of the resolution cell. Thus, the number of scatterers contributing to the echo signal from the tissue varies, and consequently, the measured effective density of scatterers varies, too. It is important to determine the measurement conditions that reduce the influence of beam size. To this end, we used the needle hydrophone (40 mm in diameter) to measure the 26-dB width of the beam in the focal zone (62 mm). Results are illustrated in Figure 6. Pulse length remained constant in the focal zone; thus, only the variation in beam cross-sectional area was used to assess the changes in resolution cell volume. To assess the influence of resolution cell volume variation on the effective density of scatterers, we followed the technique proposed by Raju et al. (2003). We stored 20 RF images, each obtained for the scanner probe focused at different depths (step 5 0.25 mm) in the tissuemimicking phantom. For all images, the same part of the phantom (2 3 2 mm) was selected, and the appropriate RF data were used to calculate EDS values. Depending on the focus position, the same scatterers were located within the narrow beam (in focus) or the wider beam (out of focus) (Fig. 7). The results revealed that beam cross-sectional area increased maximally by 15% when the observation point
754
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 6. The 26-dB cross-sectional area of the beam emitted by the transducer used in the scanner, measured in a 62 mm focal zone (0 focus).
moved 61 mm from the focus. We also found that for this focal zone, the variation in EDS did not exceed 12% (Fig. 8). So, during analysis of skin RF data, the effective density of scatterers was calculated for ROIs that were localized in a 61-mm focal zone.
The acquired echoes, first corrected for TGC, were subsequently compensated for attenuation (Litniewski et al. 2011). For each measured RF line, the following attenuation compensating algorithm was applied. First, the spectrum of the recorded RF signal was calculated. The amplitude of each spectral component was individually compensated by increasing its value along with an increase in the pulse penetration and value of the previously determined frequency-dependent attenuation, a(f). Next, a compensated signal, RFc, was constructed on the basis of compensated spectral components. The process is described by the formula. RFc ðzi Þ 5
M X
FFTk expðaðfk Þzi Þexpð22pjfk ti Þ
(4)
k50
where k is the index of the spectral component, fk denotes the spectrum frequency component and FFT is a complex spectrum of backscattered signal. zi 5 ti $c, where c denotes velocity of the longitudinal wave in tissue (1540 m/s according to Wells 1977), andti 5 idt is the time where dt is sampling interval. The summation is carried over the whole range of frequency components of the transmitted signal (from 0 to M). The real part of RFc is desired backscattered signal compensated for attenuation. RESULTS
Fig. 7. Principle of determination of the influence of focusing on the value of the effective density of scatters (EDS). The data were collected from the same region of interest placed at different distances from the transducer.
Several authors (Raju and Srinivasan 2002; Lebertre et al. 2002) have pointed out that the K-distribution describes the statistics of the skin backscattered signal envelope much better than the Rayleigh distribution. But backscatter statistics depend greatly on the frequency and focusing conditions and, thus, should be determined individually for each scanner system used in experiments. Thus, the data for healthy skin
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
755
Fig. 8. Variation in the value of the effective density of scatters (EDS) as a function of the distance from the focus denoted as a ‘0’.
were pre-processed for statistical analysis (TGC and attenuation influence correction), and then histograms of the amplitudes were created. Also, theoretical probability density functions for the K- and Rayleigh distributions were calculated using the EDS parameters b 5 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 4EDS=E½A2 and s 5 E[A2]/2 for the K- and Rayleigh distributions, respectively. These parameters were determined from the envelopes of signals recorded from the skin. The experimental histograms were fitted to the Kand Rayleigh distributions (see Fig. 9), and the goodness of fit was assessed by calculation of the mean square error (MSE). For all cases, the MSE was several times smaller for the K-distribution fit. The mean MSEs for the K- and Rayleigh distribution fits were 0.012 and 0.083, respectively.
Attenuation coefficient and EDS were calculated including the compensation of the focusing effects, TGC and attenuation. The results obtained for BCC lesions, pre-cancerous lesions and healthy skin are illustrated in Figures 10 and 11. For all patients, attenuation coefficients calculated for healthy skin were significantly lower (see Table 2) than those for basal cell carcinoma, and ranged from 1.73 to 2.40 dB/(cm$MHz), with a mean of 2.05 dB/(cm$MHz) and standard deviation (SD) of 0.19 dB/(cm$MHz). From the data obtained for BCC lesions, attenuation coefficient values ranged from 2.60 to 3.72 dB/(cm$MHz), with a mean of 3.27 and SD of 0.42 dB/(cm$MHz). Also, the EDS values obtained for skin cancer for all patients were lower than those calculated for healthy skin.
Fig. 9. Example of histogram obtained for signals recorded from human healthy dermis together with the calculated probability density functions.
756
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 10. Values of the attenuation coefficient calculated for actinic keratosis (C) and healthy skin (:) (a) and basal cell carcinoma (C) and healthy skin (:) (b).
The EDS values of scatterers for healthy skin ranged from 1.61 to 2.31 (mean 51.88, SD 5 0.31), and those for skin lesions ranged from 0.82 to 1.28 (mean 5 1.03, SD 5 0.14). As in the case of BCC, attenuation coefficients for all pre-cancerous lesions were higher than coefficients measured for healthy skin. Mean (SD) attenuation coefficient values obtained for pre-cancerous lesions and healthy skin were 3.33 (0.44) and 2.12 (0.19) dB/ (cm$MHz), respectively. COMPARISON OF TISSUE GROUPS EXAMINED WITH RESPECT TO STATISTICAL DIFFERENCES Statistical analysis of the skin lesions examined was performed using Statistica Version 8 software (StatSoft, Tulsa, OK, USA). The analysis concerned the (i) attenuation coefficient (AC); (ii) EDS values determined in AK or BCC lesions and healthy skin; and (iii) differences in AC or EDS between the lesion and healthy skin for the
same patient (ACD, EDSD). Healthy skin had to be at least 3–5 cm from a lesion. When statistical analysis was conducted within a single diagnostic group (AK or BCC), the variables were considered dependent. In analyses conducted between diagnostic groups (comparison between AK and BCC), variables were treated as independent. In the AK group, one outlier was found and was not included in analysis, consequently reducing AK group size to 13 cases. The results of the statistical analysis are summarized in Table 2. Analysis of the attenuation coefficient Differences in attenuation coefficient between diseased and healthy skin were analyzed within each diagnostic group (AK and BCC). The analysis was based on Student’s t-test for dependent variables. It is quite understandable why we chose this parametric test. The statistical power of non-parametric tests is predominant in studies in which the number of data is small, as in our study.
Fig. 11. Values of EDS (effective density of scatterers) calculated for actinic keratosis (C) and healthy skin (:) (a) and basal cell carcinoma (C) and healthy skin (:) (b).
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
757
Table 2. Comparison of examined tissue groups with respect to statistical differences Group/parameter AK vs. healthy BCC vs. healthy AK vs. BCC
p
AC
EDS
ACD
EDSD
Statistically significant at p , 0.05 p Statistically significant at p , 0.05 p Statistically significant at p , 0.05 p
Yes ,0.0001 Yes ,0.0001 No .0.05
No .0.05 Yes ,0.0001 Yes ,0.0001
n/a
n/a
n/a
n/a
No .0.05
Yes ,0.0001
AC 5 attenuation coefficient; ACD 5 difference between lesion AC and healthy skin AC; AK 5 actinic keratosis; BCC 5 basal cell carcinoma; EDS 5 effective density of scatterers; EDSD 5 difference between lesion EDS and healthy skin EDS; n/a 5 not available.
The normality of the variables analyzed was tested with the Shapiro–Wilk test, and the Grubbs statistic was used to capture outliers. For hypothesis testing, the level of significance was a 5 0.05, and in cases where the choice was up to the researcher, the two-sided critical region was used. In both groups (AK, BCC) the mean attenuation coefficient for diseased skin was significantly higher than that for healthy skin (p , 0.0001). Differences between the AK and BCC groups caused by variation in ACs for diseased and healthy skin (ACD) were also analyzed. Next, Student’s t-test or the Cochran–Cox test (depending on the result of the F-test used to assess the equality of variances) was carried out to analyze the independent variables. There were no statistically significant differences between means. EDS analysis Analysis of EDS was based on the same method as analysis of AC. It was found that in the AK group, there was no need to reject the null hypothesis that the mean EDS values of diseased and healthy skin do not differ from each other. In the BCC group, such a hypothesis was rejected. The average EDS value for diseased skin was significantly lower than the mean EDS value for healthy skin (p , 0.0001). At the same time, the average EDS value for diseased skin in the BCC group was significantly lower than that in the AK group (p , 0.0001). The average EDS value for healthy skin in the BCC group did not differ significantly from the average EDS value for healthy skin in the AK group. Discriminant analysis Significant differences in EDS between the AK and BCC groups caused us to build a discriminatory model, the aim of which was to find a formula for classifying a given patient in the AK or BCC group on the basis of the EDS value. In discriminant analysis, a model using the classification functions under the assumption of equal a priori probabilities was applied. The classification matrix gave 100% percent compliance. The cutoff value was 1,466.
DISCUSSION AND CONCLUSIONS The goal of our study was to answer the question: Can healthy tissue, actinic keratosis lesions and basal cell carcinoma lesions be differentiated using both the backscatter properties and attenuation coefficients of skin tissue? Statistical analysis assumed the K-distribution for the envelopes of the skin backscattered echoes. That resulted from analysis of echo amplitude distributions. By comparing the Rayleigh and K-distribution models, we found that K-PDF well represented healthy dermal tissue, with the MSE eight times lower than the error calculated for Rayleigh-PDF. Our results are in agreement with the results reported by Lebertre et al. (2002), in whose study the K-distribution performed well, describing echoes from human abdominal skin samples using 35-MHz ultrasound. Other authors (Raju and Srinivasan 2002) also proved that at 30–50 MHz, skin echo amplitude departs from Rayleigh statistics. The K-distribution is particularly interesting because it allows determination of the EDS. We also found that EDS was lower for BCC lesions than for healthy skin and AK lesions (Fig. 11). The Kdistribution is sensitive to the number and uniformity of scatterers within the resolution cell. The decrease in EDS for BCC can be explained by the lower spatial density of collagen fibers (dominant scatterers) that are replaced by tumor clusters. Analogous work carried on with healthy and BCC skin samples (Petrella et al. 2012) also found that the statistical approach could be a good descriptor of tissue condition. Modeling the echo amplitude distribution with gamma and Weibull PDFs, the authors found significant differences in signals scattered in healthy and diseased skin. At the same time, histologic images revealed the dense net of collagen fibers and tumor cell clusters within the collagen fibers of healthy skin and BCC samples, respectively. In the case of AK lesions, the collagen fiber matrix in the dermis remains unaltered; thus, the scattered echo statistical properties are similar to the properties of healthy tissue (Fig. 11). The attenuation coefficient was the second parameter analyzed. Results revealed lower values for healthy skin than for BCC and AK lesions. Olerud et al. (1990)
758
Ultrasound in Medicine and Biology
reported that the amount of water contained in the skin is crucial for the attenuation of ultrasound. The more water in the skin, the lower is the attenuation coefficient. These results were also obtained by Raju and Srinivasan (2002), who determined that swelling skin, and thus tissue in which hydration is increased, had lower attenuation compared with healthy tissue. The high attenuation for AK can be explained by the low skin water content that is characteristic of this kind of lesion (King and Brucker 2011). The higher attenuation found for BCC lesions is more difficult to explain. Some (Pan et al. 1998) claim that the value of the attenuation coefficient of the skin depends on deformation of the collagen fiber matrix. They found a linear relation between the attenuation coefficient and the mechanical loading of skin samples. Changes in collagen matrix structure also occur when tumor cells and tumor cell clusters grow in the dermis, and deformation of the matrix could result in increased attenuation in BCC lesions. Although the number of cases (21 BCCs and 13 AKs) in this study is rather small to draw definite conclusions, it is worth pointing out that in all the cases considered, we obtained lower EDS values for BCC lesions and higher attenuation coefficient values for BCC and AK lesions compared with the healthy skin reference. That gives us a chance to distinguish AK cases from BCC cases because although both are characterized by a higher attenuation coefficient, only BCC has a lower EDS value. Furthermore, the discriminatory model indicated that EDS based on the statistical properties of backscatter enables differentiation of BCC and AK lesions. The cutoff value was 1,466, which means that each new patient with EDS below the cutoff value should be classified in the BCC group, whereas patients with an EDS .1,466 would be classified in the AK group. Statistical analysis also revealed the ability of a relative measure (EDSD) to differentiate BCC and AK. In conclusion, the statistical parameters, together with the attenuation coefficient, allow differentiation between basal cell carcinoma lesions, pre-cancerous lesions and healthy skin. The good performance of the described approach to tissue characterization suggests that the diagnosis of skin diseases characterized by changes in collagen matrix structure could be enhanced by ultrasonic examination. Medical consultants with whom we are working believe that the ultrasonic method can reduce the number of biopsies and increase the accuracy of estimates of the penetration depth of lesions. In cases of positive ultrasonic classification of a lesion as BCC, the lesion should be removed without confirmation of the diagnosis by biopsy. In turn, accurate evaluation of tumor tissue penetration depth is very important when planning surgical removal, resulting in maintenance of
Volume 41, Number 3, 2015
the required safety margin, on the one hand, and minimization of post-operative scarring, on the other hand. In addition, ultrasound evaluation results in a better choice of methods for removal of AK lesions. In deeper lesions, surgery is required, whereas in superficial lesions, laser therapy or cryotherapy may suffice. Acknowledgments—This work was supported in part by the Ministry of Science and Higher Education of Poland, Project N N518295140, and by National Science Centre of Poland, Project 2011/01/B/ST7/06728.
REFERENCES Bigelow T, McFarlin B, O’Brien WD, Oelze M. In vivo ultrasonic attenuation slope estimates for detecting cervical ripening in rats: Preliminary results. J Acoustic Soc Am 2008;123:1794–1800. Cham BE. Solasodine glycosides: A topical therapy for actinic keratosis. A single-blind, randomized, placebo-controlled, parallel group study with Curaderm. J Cancer Ther 2013;4:588–596. Chen X, Phillips D, Schwartz KQ, Mottley JG, Parker KJ. The measurement of backscatter coefficient from a broadband pulse-echo system: A new formulation. IEEE Trans Ultrason Ferroelect Freq Control 1997;44:515–525. Dutt V. Statistical analysis of ultrasound echo envelope. Rochester, MN: Mayo Graduate School; 1995. Fornage BD, McGavran MH, Duvic M, Waldron CA. Imaging of the skin with 20 MHz US. Radiology 1993;189:69–76. Foster FS, Pavlin CJ, Harasiewicz KA, Christopher DA, Turnbull DH. Advances in ultrasound biomicroscopy. Ultrasound Med Biol 2000;26:1–27. Fournier C, Bridal SL, Berger G, Laugier P. Reproducibility of skin characterization with backscattered spectra (12–25 MHz) in healthy subjects. Ultrasound Med Biol 2001;27:603–610. Gniadecka M, Jemec GBE. Quantitative evaluation of chronologic ageing and photoaging in vivo: Studies on skin echogenicity and thickness. Br J Dermatol 1998;139:815–821. Goodman JW. Statistical optics. New York: Wiley Classics Library; 2000. Guittet C, Ossant F, Vaillant L, Berson M. In vivo high frequency ultrasonic characterization of human dermis. IEEE Trans Biomed Eng 1999;46:740–746. Harland C, Kale S, Jackson P, Mortimer P, Bamber J. Differentiation of common benign pigmented skin lesions from melanoma by high resolution ultrasound. Br J Dermatol 2000;143:281–289. Jakeman E, Tough RJA. Generalized K distribution: A statistical model for weak scattering. J Opt Soc Am A 1987;4:1764–1772. King T, Brucker M. Pharmacology for women’s health. London: Jones & Bartlett; 2011. Lebertre M, Ossant F, Vaillant L, Diridollou S, Patat F. Spatial variation of acoustic parameters in human skin: An in vivo study between 22 and 45 MHz. Ultrasound Med Biol 2002;28:599–615. Lewandowski M, Nowicki A. High frequency imaging system with coded excitation. IEEE Trans Ultrason Ferroelectr Freq Control 2008;8:1878–1882. Litniewski J, Cieslik L, Wojcik J, Nowicki A. Statistics of the envelope of ultrasonic backscatter from human trabecular bone. J Acoust Soc Am 2011;130:2224–2232. Litniewski J, Nowicki A, Klimonda Z, Lewandowski M. Sound fields for coded excitations in water and tissue: Experimental approach. Ultrasound Med Biol 2007;33:601–607. Olerud J, O’Brien WD, Riederer-Henderson MA, Steiger DL, Debel J, Odland G. Correlation of tissue constituents with the acoustic properties of skin and wound. Ultrasound Med Biol 1990;16:55–64. Pan L, Zan L, Foster FS. Ultrasonic and viscoelastic properties of skin under transverse mechanical stress in vitro. Ultrasound Med Biol 1998;2:995–1007. Petrella LI, De Azevedo VH, Issa PR, Martins CJ, Machado CJ, Pereira WC. Statistical analysis of high frequency ultrasonic
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al. backscattered signals from basal cell carcinomas. Ultrasound Med Biol 2012;38:1811–1819. Piotrzkowska H, Litniewski J, Szymanska E, Nowicki A. Ultrasonic echosignal applied to human skin lesions characterization. Arch Acoust 2012;37:103–109. Raju BI, Srinivasan MA. High-frequency ultrasonic attenuation and backscatter coefficients if in vivo normal human dermis and subcutaneous fat. Ultrasound Med Biol 2001;27:1543–1556. Raju BI, Srinivasan MA. Statistics of envelope of high-frequency ultrasonic backscatter from human skin in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2002;49:871–882. Raju BI, Srinivasan MA, Gonzalez S, Srinivasan M. Quantitative ultrasonic methods for characterization of skin lesions in vivo. Ultrasound Med Biol 2003;29:825–838. Shankar PM, Dumane VA, Reid JM, Genis V, Forsberg F, Piccoli CW, Goldberg BB. Use of the K-distribution for classification of breast masses. Ultrasound Med Biol 2000;26:1503–1510. Shankar PM, Reid JM, Ortega H, Piccoli CW, Goldberg B. Use non-Rayleigh statistics for the identification of tumors in ultrasonic B-scan of the breast. IEEE Trans Med Imaging 1993;12:687–692.
759
Tsui PH, Yeh CK, Liao YY, Chang CC, Kuo WH, Chang KJ, Chen CN. Ultrasonic Nakagami imaging: A strategy to visualize the scatterer properties of benign and malignant breast tumors. Ultrasound Med Biol 2010;36:209–217. Turnbull DH, Starkosky BG, Harasiewicz KA, Semple JI, From L, Gupta AK, Sauder N, Foster FS. A 40-100 B-scan ultrasound backscatter microscope for skin imaging. Ultrasound Med Biol 1995;21:79–88. Wagner RF, Insana MF, Brown DG. Statistical properties of radio frequency and envelope detected signals with application to medical ultrasound. J Opt Soc Am A 1987;4:910–922. Wagner RF, Smith SW, Sandrik JM, Lopez H. Statistics of speckle in ultrasound B-scans. Trans IEEE Sonics Ultrason 1983;30:156–163. Weber L, Kirsch E, M€uller P, Krieg T. Collagen type distribution and macromolecular organization of connective tissue in different layers of human skin. J Invest Dermatol Symp Proc 1984;82:156–160. Wells PNT. Biomedical ultrasonics. New York: Academic Press; 1977. Weng L, Reid J, Shankar M, Soetanto K. Ultrasound speckle analysis based on the K distribution. J Acoust Soc Am 1991;89:2992–2995. Wortsman X, Jemec G. Dermatologic ultrasound with clinical and histologic correlations. New York: Springer; 2013.
http://dx.doi.org/10.1016/j.ultrasmedbio.2014.11.016
d
Original Contribution QUANTITATIVE SONOGRAPHY OF BASAL CELL CARCINOMA HANNA PIOTRZKOWSKA-WROBLEWSKA,* JERZY LITNIEWSKI,* ELZBIETA SZYMANSKA,y and ANDRZEJ NOWICKI* * Department of Ultrasound, Institute of Fundamental Technological Research, Warsaw, Poland; and y Dermatology Clinic, CSK MSWiA Hospital, Warsaw, Poland (Received 16 January 2014; revised 29 October 2014; in final form 23 November 2014)
Abstract—A 30-MHz ultrasonic scanner was used to collect B-scan images together with appropriate radiofrequency echoes from diseased and healthy skin regions of patients with diagnosed basal cell carcinoma and pre-cancerous lesions (actinic keratosis). Radiofrequency data were processed to obtain the attenuation coefficient and statistics of the backscattered echo signal determination (K-distribution and effective density of scatterers [EDS]). The attenuation coefficient was significantly higher for patients with basal cell carcinoma than for healthy patients. Also, the pre-cancerous skin lesions had increased attenuation. The averaged EDS values for cancer lesions were significantly lower than those for pre-cancerous lesions and healthy skin. The successful differentiation between the tissue groups examined suggests the potential value of the attenuation coefficient and EDS for carcinoma characterization. (E-mail: [email protected]) Ó 2015 World Federation for Ultrasound in Medicine & Biology. Key Words: Quantitative ultrasound, High frequency, Human skin, Skin lesions, K-distribution, Attenuation coefficient, Tissue characterization.
Foster et al. (2000) reported that although ultrasound (US) has the capability to image fine features in the skin, such as sweat gland ducts, hair follicles and veins, the diagnostic ability of skin images to reveal specific pathologies is limited. It is difficult to differentiate between benign and malignant lesions using B-scan images (Fornage et al. 1993) because both types of lesions appear hypo-echogenic compared with healthy skin. Another study indicated that both scar tissue and malignant melanoma could appear similar in US scans (Turnbull et al. 1995). Thus, biopsy is still the gold standard in final diagnosis of skin cancer; however, quantitative ultrasound can provide additional information that is potentially helpful in lesion assessment and screening tests. First, quantitative studies of skin lesions in vivo using ultrasound have mostly been limited to parameters that could be computed directly from images, recorded by using commercially available 20-MHz systems. For instance, changes in skin echogenicity (mean pixel amplitude, which is proportional to the mean backscatter amplitude) and skin thickness have been found to be related to photo-aging of the skin (Gniadecka and Jemec, 1998). The degree of acoustic shadowing, measured as the ratio of echogenicity of the retrocessional dermis to that of the perilesional dermis, was found to be
INTRODUCTION Basal cell carcinoma (BCC) is the most common cutaneous malignancy, representing 80% of all skin cancer cases. BCC arises from the basal cells of the epidermis. Although BCC is not associated with significant mortality, the associated morbidity and therapeutic costs are an increasing burden to the health care system. BCC cases are rather easily diagnosed, but a substantial number of cases require biopsy because of their similarity to other cutaneous neoplasms, both benign and malignant, making the BCC direct diagnosis ambiguous. Basal cell carcinoma can develop from precancerous growths like an actinic keratosis (AK), as well as from unchanged skin. According to Cham (2013), it is estimated that up to 50% of the population is affected by AK. The difficulty in predicting the evolution of this kind of lesion makes AK very dangerous. It may not be a threat to the health and life of the patient for a long period, but at some point it can vanish or transform into cancer. Address correspondence to: Hanna Piotrzkowska-Wroblewska, Department of Ultrasound, Institute of Fundamental Technological Research, ul. Pawinskiego 5 b, 02-106 Warsaw, Poland. E-mail: [email protected] 748
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
capable of differentiating between basal cell papilloma and malignant melanoma (Harland et al. 2000). Other parameters that require analysis of radiofrequency (RF) backscattered echoes, such as the attenuation coefficient (Guittet et al. 1999) and the apparent integrated backscatter (Fournier et al. 2001), have also been studied in healthy skin tissues in vivo. Although the statistics of ultrasonic signals have been extensively studied with respect to their potential to classify tissue types (Shankar et al. 1993, 2000; Tsui et al. 2010), such studies have not been widely carried out in skin tissue. Raju and Srinivasan (2001, 2002) studied in vivo the frequency-dependent attenuation and backscatter coefficient, as well as parameters related to echo statistics, such as the ratio of mean to standard deviation, denoted as signal-to-noise ratio (SNR), and the parameters of envelope probability density functions (PDFs) of healthy skin tissues. They found that the generalized gamma, K- and Weibull distributions modeled the envelope statistics well, whereas the Rayleigh distribution provided a rather poor fit. The use of quantitative ultrasound in the monitoring of skin tissue affected by basal cell carcinoma was proposed by Petrella et al. (2012). Their study was performed ex vivo using an ultrasound biomicroscope working at a frequency of 45 MHz. They examined SNR and shape parameters of Weibull and generalized gamma PDFs. The significant differences between lesions with various distribution patterns of tumor nests suggested that quantitative ultrasound has potential for use in carcinoma characterization. Similar results were presented in our previous study. We found that the K-distribution is appropriate for modeling statistics of the envelope of signals obtained from human skin in vivo, and the value of the shape parameter of the K-distribution can help in differentiation of healthy skin and skin affected by BCC (Piotrzkowska et al. 2012). The goal of this study was to find the quantitative measure of skin tissue backscattering properties for differentiating changes in tissue structure induced by BCC and AK. The new method allows non-invasive differentiation of the disease entities and, consequently, selection of the proper method of treatment. In the case of BCC, the most common method of treatment is surgical removal of lesions with the correspondingly large margin or, in some cases, cryotherapy or laser therapy. Actinic keratosis requires conservative treatment, locally (imiquimod or isotretinoin) or with cryotherapy or laser therapy. Because the recorded signals are dependent on both the tissue properties and the instrument used in recording the signals, methods for computing quantitative parameters, such as the attenuation coefficient and the shape parameter of the K-distribution, were investigated,
749
taking into account the compensation of the influence of the system-dependent effects. STRUCTURE OF HUMAN SKIN AND BBC LESIONS Skin is a multilayered organ, with the separate histologic layers epidermis, dermis and subcutaneous tissue. The epidermis is the thinnest layer of skin. The main cell populations of the epidermis are keratinocytes, melanocytes and Langerhans cells. The middle layer of skin is the dermis, which consists of connective tissue arranged in two layers: the more superficial papillary dermis and the deeper reticular dermis. Thin papillary dermis is made of loose connective tissue containing mostly collagen fibers. In the thicker reticular dermis, connective tissue is denser because of the coarse bundles of collagen fibers arranged in layers parallel to the skin surface. There are other structures in the dermis, such as blood vessels, lymphatic vessels, nerve fibers, portions of hair follicles and sweat glands; however; their volume percentage is much lower than that of collagen fibers. All dermal components are bound together by a gel-like ground substance made of various glyosaminoglycans and glycoproteins. The third skin layer is subcutaneous fat, further divided into lobules containing adipose cells and separated by thin fibrovascular septa. The septa consist of collagen and reticulin fibers, blood and lymphatic vessels and cutaneous nerves. Basal cell carcinoma tumors are typically characterized by a fibrous stroma surrounding clusters of tumor cells that resemble collagen fibers of the basal layer of the epidermis. Usually, these tumor cells are fairly regular with rounded nuclei and little cytoplasm. The level of skin changes (percentage of pathologic structures in the volume of tissue) reflects the progression of the lesion. In the initial phase of the disease, the number of cancerous cells is relatively low; however, in the advanced stage of cancer, the number and size of clusters increase, which reduces the proportion of collagen fibers in the dermis to about 30%. In 30-MHz ultrasonic images, all three layers of skin can be observed. BCC lesions are usually located in the dermis, and all results discussed in this article concern this layer. The echogenicity in images of the dermis is determined by a collagen component (Wortsman and Jemec 2013). Collagen fibers are 5–15 mm thick with an average density of 1250 kg/m3 and longitudinal wave sound speed of 1730 m/s (Weber et al. 1984) Thus, the resultant impedance is much higher than the impedance of BCC cells. We assume that the replacement of fibers by tumor cells leads to changes in tissue attenuation and backscatter statistics.
750
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
METHODS Acquisition system The experimental 30-MHz ultrasonic scanner and data acquisition procedure are described in detail elsewhere (Lewandowski and Nowicki 2008), and only a brief description is given here. Ultrasonic waves were generated by the 20-mm-thick spherical transducer made of the modified thick film piezoelectric lead zirconate titanate (PZT) 37 deposited on the PZT substrate (Ferroperm, Vedbæk, Denmark). Properties of the thick film transducer and the generated acoustic field are given in Table 1. The transverse pressure distribution of the probing pulse measured in focus using the needle hydrophone (Precision Acoustics, UK) is illustrated in Figure 1. The transducer was mounted in the mechanical, wobbler type-probe generating a sector scan with the 10-Hz frame rate. To increase the SNR, we used the 32bit Golay-encoded transmit sequences and compression of received echoes (Litniewski et al. 2007). Ultrasonic data were converted from analogue to digital using a 200-MHz, 12-bit resolution sampling system. The computer was used to control mechanical scanning and data acquisition. Received sequences were first compressed and then envelope detected and displayed. The compressed RF data were stored with a time gain control (TGC) curve that was used to increase the quality of skin images. One image corresponded to 200 RF echo lines, each 2048 samples long. Once the RF data were collected, further processing and analysis were done off-line on the computer using Mathcad (Mathsoft, Cambridge, MA, USA) software. Quantitative parameters Attenuation coefficient. The attenuation coefficient aðf Þ of the skin was determined using the spectral difference technique, which is based on comparison of the power spectra of backscattered signals recorded at two different depths and denoted as SAi ðf Þ and SBi ðf Þ, respectively, where i represents the number of echo lines considered, and f denotes frequency. Only the RF data from selected regions of interest (ROIs) were used. The procedure was as follows: First, from the selected part
Fig. 1. (a) Transducer pressure distribution in the focal plane. (b) Shape of the scanning pulse measured in the focus.
of the scattered echo signal in the focal zone, two segments, 0.18 ms long and separated by distance Dz 5 1 mm, were weighted using a Hanning window. Next, these segments were Fourier transformed, and the power spectra were calculated. The spectra from the same depth were averaged over all echo lines of the
Table 1. Properties of the transducer and generated acoustic field Center frequency
30 MHz
26-dB bandwidth F-number Diameter Focal length Depth of focus (26 dB) Axial resolution Lateral resolution in focus
14 MHz 2.9 3 mm 9 mm 2 mm 0.07 mm 0.14 mm
Fig. 2. Plots of the K-distribution probability density functions for various values of effective density of scatterers (EDS).
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
751
Fig. 3. B-Mode images of healthy human dermis (a), pre-cancerous skin (b) and basal cell carcinoma (c) within the indicated regions of interest.
ROI (not ,25 lines), yielding two averaged spectra SB ðf Þ and SA ðf Þ, and the attenuation aðf Þ was estimated from their log ratio. The longitudinal wave velocity of 1540 m/s was assumed (Wells 1977). The formula used to determine the attenuation was. B ðf ;zÞ 10 log SSAB ðfðf Þ$corr Þ$corrA ðf ;zÞ (1) aðf Þ 5 2 4Dz where P P SA ðf Þ5ð1=NÞ Ni5 1 SAi ðf Þ and SB ðf Þ5 ð1=NÞ Ni5 1 SBi ðf Þ. N is the number of echo lines in the ROI, and corrB ðf Þ and corrA ðf Þ denote diffraction correction coefficients dependent on frequency and distance from the focus (see Fig. 5).
The attenuation coefficient was then calculated from the slope of a least-squares linear fit of attenuation versus frequency dependence. Statistics of echo envelope. The statistics of echo signals backscattered by a medium depends on both medium structure and spatial resolution of the scanning system, which is defined by the size of the resolution cell. This size varies with pulse length and acoustic beam cross section. The soft tissue is often modeled as a collection of spatially distributed small scatterers (Wagner et al. 1983, 1987). It can be illustrated that for the large number of scatterers within the resolution cell, the signal amplitude statistics follow the Rayleigh distribution. But in biological tissue, the spatial distribution of scattering
Fig. 4. Normalized amplitudes of three selected frequency components (15, 25 and 35 MHz) of the reflected pulse power spectrum over a distance of 2 mm on either side of the focus.
752
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 5. Diffraction correction curves, corrB(f, z) and corrA(f, z) in formula (1), measured 60.5 mm from the focus.
elements is not uniform. Also, the scatterer density and size vary throughout the tissue. Thus, the effective number of scatterers in the resolution cell may not fulfill the conditions for Rayleigh statistics, particularly when the resolution cell is small. In this case non-Rayleigh statistics are considered (Goodman 2000). Jakeman and Tough (1987) pointed out that for the low effective density of scatterers, the PDF of the backscatter amplitude is given by the K-distribution. More recently, it was reported that the K-distribution describes well the statistics of the skin backscatter (Raju and Srinivasan 2002). The K-distribution is defined by pðAÞ 5 2
EDS EDS11 A b KEDS21 bA 2 GðEDSÞ
(2)
where A is amplitude, pðAÞ is probability density, b 5 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi p 4EDS=E½A2 , Kb ð$Þ is the modified Bessel function of the second kind of order b, Gð$Þ is the standard gamma function, E[ ] is the expectation operator, EDS 5 Nð11vÞ is the effective density of scatterers within the resolution cell, N is the number of scatterers and v is a constant depending on scatterer characteristics (Dutt 1995). Figure 2 illustrates the probability density function of the K-distribution plotted for various values of EDS. It is easy to see that the K-distribution covers the range of distributions, approaching the Rayleigh distribution as EDS becomes large and becoming log normal as EDS / 0. It can be seen that, in practice, the K-distribution for EDS . 10 converges to the Rayleigh distribution. The effective density of scatterers estimated from the K-distribution model depends on the actual number of scattering sites per resolution cell, as well as the uniformity of backscatter coefficient. The type and distribution of scatterers are intrinsically related to the type of tissue through which the ultrasound beam is passing. Therefore, the EDS parameter can be used to distinguish
between regions differing in spatial density of scatterer or between regions of varying scatterer cross section and, as a consequence, can be used as a parameter for tissue characterization. The EDS can be estimated from the fourth normalized moment (r4) of the K-distribution (Weng et al. 1991): E½A 2 where r4 5 ½E½A EDS 5 r4 22 2 2 4
(3)
The start and endpoints of the signal acquisition and the number of lines (ROI) were first selected from the image. The moments E½An were directly computed from the envelopes of RF signals enclosed in the ROI. Next, r4 was determined, and the EDS parameter was calculated using eqn (3). To verify whether the K-distribution is a good model to describe statistics of the amplitude of skin backscatter when using our scanner, the histograms of experimental data were calculated and fitted by the K- and Rayleigh distributions. The results obtained for the skin of healthy volunteers, discussed in the Results, reveal the superiority of the K-distribution to the Rayleigh distribution. Correct estimation of the effective density of scatterers requires a large number of data samples. To find the size of ROIs that gives the ROI size-independent EDS parameter, we used RF data corresponding to the homogenous part of the skin. A set of concentric rectangular ROIs of varying length was selected in the focal zone. Data from each ROI were analyzed for their respective EDS parameter. It was found that a 1 3 1-mm ROI box, corresponding to a 1.3-ms-long time window and 25 adjacent scan lines, is sufficient to obtain reliable results. Acquisition of ultrasonic data from the skin Skin tissues were examined in vivo. A group of patients from the Dermatology Clinic with diagnosed
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
BCC and pre-cancerous lesions (AK) participated in this study (21 and 14 cases, respectively, sex and age unknown). The approval of the ethical review board of the Medical University of Warsaw and informed consent for the study from all patients were obtained before measurements. Examinations were carried out by medical doctors. Three types of data—traditional B-mode image, set of RF echo lines and TGC curve—were stored for the skin regions where BCC was diagnosed and where precancerous lesions were localized. For all patients, reference data for the selected healthy fragments of the skin were also recorded. Healthy skin RF samples were taken from regions positioned as close to the lesion as possible. In general, the distance between lesion and healthy skin reference depends on the extension of the lesion, but in our experience, spacing from 3 to 5 cm is sufficient; however, it was always decided by the examiner. Examples of B-mode images of healthy skin and skin with BCC and pre-cancerous lesions (AK) are provided in Figure 3. Data pre-processing for attenuation estimation Attenuation in the examined tissue results in the decay of wave amplitude with depth. The latter is partially compensated by the TGC system of the receiving amplifier, which is useful for B-mode imaging, but interferes with estimation of quantitative parameters. The influence of the TGC system on RF echoes was compensated before the processing that was carried out to determine the attenuation coefficient of the skin and the statistical properties of the skin backscatter. It is important to compensate for the effects of diffraction when using the focused transducer to determine tissue attenuation from the backscatter. Focusing introduces variation of the beam amplitude with distance and shifts the mean frequency of the ultrasonic pulse in the focal region. When the acoustic pulse is approaching the focus, the amplitudes of higher-frequency components increase faster than those of lower-frequency components. For the diverging beam, the opposite effect occurs. Thus, diffraction affects the measurements of spectra and, consequently, the estimated attenuation coefficient. To compensate for the diffraction effects, several approaches have been proposed. Chen et al. (1997) developed theoretical diffraction correction curves based on the physics of acoustic wave propagation, including transducer geometry and frequency characteristics. The empirical approach was presented by Raju and Srinivasan (2002). Correction curves were determined from the acoustic data collected from the same tissue that was placed at different distances from the focus. The diffraction correction term proposed by Bigelow et al. (2008) is based on the assumption that the field
753
pattern along the focal zone can be approximated by a Gaussian function. The Gaussian depth of focus as a function of wavelength was measured by acquiring echoes from a rigid plane located at different locations along the beam axis. In our study, we compensated for focusing by processing the spectra similarly to the method proposed by Bigelow et al. (2008); however, the correction coefficients were determined directly from the amplitude spectra of echoes obtained from a rigid plane reflector located in water at various axial distances from the focus. Figure 4 illustrates the amplitudes of three selected frequency components of the reflected pulse power spectrum over a distance of 2 mm on either side of the focus. It can be seen that the 15-MHz frequency component has the larger width, which results from the weaker focusing at low frequency. For each spectral component, the correction coefficients compensating the amplitude changes induced by focusing were calculated. Correction coefficient-versus-frequency curves are provided in Figure 5 for two selected distances from the focus. They equalize the amplitude of the frequency components obtained in water, which is assumed to be a non-attenuating medium. These coefficients were used in formula (1) to compensate the power spectra collected from patients. Data selection and pre-processing for statistical analysis Focusing varies the beam cross-sectional area along the path of the beam, which changes the volume of the resolution cell. Thus, the number of scatterers contributing to the echo signal from the tissue varies, and consequently, the measured effective density of scatterers varies, too. It is important to determine the measurement conditions that reduce the influence of beam size. To this end, we used the needle hydrophone (40 mm in diameter) to measure the 26-dB width of the beam in the focal zone (62 mm). Results are illustrated in Figure 6. Pulse length remained constant in the focal zone; thus, only the variation in beam cross-sectional area was used to assess the changes in resolution cell volume. To assess the influence of resolution cell volume variation on the effective density of scatterers, we followed the technique proposed by Raju et al. (2003). We stored 20 RF images, each obtained for the scanner probe focused at different depths (step 5 0.25 mm) in the tissuemimicking phantom. For all images, the same part of the phantom (2 3 2 mm) was selected, and the appropriate RF data were used to calculate EDS values. Depending on the focus position, the same scatterers were located within the narrow beam (in focus) or the wider beam (out of focus) (Fig. 7). The results revealed that beam cross-sectional area increased maximally by 15% when the observation point
754
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 6. The 26-dB cross-sectional area of the beam emitted by the transducer used in the scanner, measured in a 62 mm focal zone (0 focus).
moved 61 mm from the focus. We also found that for this focal zone, the variation in EDS did not exceed 12% (Fig. 8). So, during analysis of skin RF data, the effective density of scatterers was calculated for ROIs that were localized in a 61-mm focal zone.
The acquired echoes, first corrected for TGC, were subsequently compensated for attenuation (Litniewski et al. 2011). For each measured RF line, the following attenuation compensating algorithm was applied. First, the spectrum of the recorded RF signal was calculated. The amplitude of each spectral component was individually compensated by increasing its value along with an increase in the pulse penetration and value of the previously determined frequency-dependent attenuation, a(f). Next, a compensated signal, RFc, was constructed on the basis of compensated spectral components. The process is described by the formula. RFc ðzi Þ 5
M X
FFTk expðaðfk Þzi Þexpð22pjfk ti Þ
(4)
k50
where k is the index of the spectral component, fk denotes the spectrum frequency component and FFT is a complex spectrum of backscattered signal. zi 5 ti $c, where c denotes velocity of the longitudinal wave in tissue (1540 m/s according to Wells 1977), andti 5 idt is the time where dt is sampling interval. The summation is carried over the whole range of frequency components of the transmitted signal (from 0 to M). The real part of RFc is desired backscattered signal compensated for attenuation. RESULTS
Fig. 7. Principle of determination of the influence of focusing on the value of the effective density of scatters (EDS). The data were collected from the same region of interest placed at different distances from the transducer.
Several authors (Raju and Srinivasan 2002; Lebertre et al. 2002) have pointed out that the K-distribution describes the statistics of the skin backscattered signal envelope much better than the Rayleigh distribution. But backscatter statistics depend greatly on the frequency and focusing conditions and, thus, should be determined individually for each scanner system used in experiments. Thus, the data for healthy skin
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
755
Fig. 8. Variation in the value of the effective density of scatters (EDS) as a function of the distance from the focus denoted as a ‘0’.
were pre-processed for statistical analysis (TGC and attenuation influence correction), and then histograms of the amplitudes were created. Also, theoretical probability density functions for the K- and Rayleigh distributions were calculated using the EDS parameters b 5 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 4EDS=E½A2 and s 5 E[A2]/2 for the K- and Rayleigh distributions, respectively. These parameters were determined from the envelopes of signals recorded from the skin. The experimental histograms were fitted to the Kand Rayleigh distributions (see Fig. 9), and the goodness of fit was assessed by calculation of the mean square error (MSE). For all cases, the MSE was several times smaller for the K-distribution fit. The mean MSEs for the K- and Rayleigh distribution fits were 0.012 and 0.083, respectively.
Attenuation coefficient and EDS were calculated including the compensation of the focusing effects, TGC and attenuation. The results obtained for BCC lesions, pre-cancerous lesions and healthy skin are illustrated in Figures 10 and 11. For all patients, attenuation coefficients calculated for healthy skin were significantly lower (see Table 2) than those for basal cell carcinoma, and ranged from 1.73 to 2.40 dB/(cm$MHz), with a mean of 2.05 dB/(cm$MHz) and standard deviation (SD) of 0.19 dB/(cm$MHz). From the data obtained for BCC lesions, attenuation coefficient values ranged from 2.60 to 3.72 dB/(cm$MHz), with a mean of 3.27 and SD of 0.42 dB/(cm$MHz). Also, the EDS values obtained for skin cancer for all patients were lower than those calculated for healthy skin.
Fig. 9. Example of histogram obtained for signals recorded from human healthy dermis together with the calculated probability density functions.
756
Ultrasound in Medicine and Biology
Volume 41, Number 3, 2015
Fig. 10. Values of the attenuation coefficient calculated for actinic keratosis (C) and healthy skin (:) (a) and basal cell carcinoma (C) and healthy skin (:) (b).
The EDS values of scatterers for healthy skin ranged from 1.61 to 2.31 (mean 51.88, SD 5 0.31), and those for skin lesions ranged from 0.82 to 1.28 (mean 5 1.03, SD 5 0.14). As in the case of BCC, attenuation coefficients for all pre-cancerous lesions were higher than coefficients measured for healthy skin. Mean (SD) attenuation coefficient values obtained for pre-cancerous lesions and healthy skin were 3.33 (0.44) and 2.12 (0.19) dB/ (cm$MHz), respectively. COMPARISON OF TISSUE GROUPS EXAMINED WITH RESPECT TO STATISTICAL DIFFERENCES Statistical analysis of the skin lesions examined was performed using Statistica Version 8 software (StatSoft, Tulsa, OK, USA). The analysis concerned the (i) attenuation coefficient (AC); (ii) EDS values determined in AK or BCC lesions and healthy skin; and (iii) differences in AC or EDS between the lesion and healthy skin for the
same patient (ACD, EDSD). Healthy skin had to be at least 3–5 cm from a lesion. When statistical analysis was conducted within a single diagnostic group (AK or BCC), the variables were considered dependent. In analyses conducted between diagnostic groups (comparison between AK and BCC), variables were treated as independent. In the AK group, one outlier was found and was not included in analysis, consequently reducing AK group size to 13 cases. The results of the statistical analysis are summarized in Table 2. Analysis of the attenuation coefficient Differences in attenuation coefficient between diseased and healthy skin were analyzed within each diagnostic group (AK and BCC). The analysis was based on Student’s t-test for dependent variables. It is quite understandable why we chose this parametric test. The statistical power of non-parametric tests is predominant in studies in which the number of data is small, as in our study.
Fig. 11. Values of EDS (effective density of scatterers) calculated for actinic keratosis (C) and healthy skin (:) (a) and basal cell carcinoma (C) and healthy skin (:) (b).
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al.
757
Table 2. Comparison of examined tissue groups with respect to statistical differences Group/parameter AK vs. healthy BCC vs. healthy AK vs. BCC
p
AC
EDS
ACD
EDSD
Statistically significant at p , 0.05 p Statistically significant at p , 0.05 p Statistically significant at p , 0.05 p
Yes ,0.0001 Yes ,0.0001 No .0.05
No .0.05 Yes ,0.0001 Yes ,0.0001
n/a
n/a
n/a
n/a
No .0.05
Yes ,0.0001
AC 5 attenuation coefficient; ACD 5 difference between lesion AC and healthy skin AC; AK 5 actinic keratosis; BCC 5 basal cell carcinoma; EDS 5 effective density of scatterers; EDSD 5 difference between lesion EDS and healthy skin EDS; n/a 5 not available.
The normality of the variables analyzed was tested with the Shapiro–Wilk test, and the Grubbs statistic was used to capture outliers. For hypothesis testing, the level of significance was a 5 0.05, and in cases where the choice was up to the researcher, the two-sided critical region was used. In both groups (AK, BCC) the mean attenuation coefficient for diseased skin was significantly higher than that for healthy skin (p , 0.0001). Differences between the AK and BCC groups caused by variation in ACs for diseased and healthy skin (ACD) were also analyzed. Next, Student’s t-test or the Cochran–Cox test (depending on the result of the F-test used to assess the equality of variances) was carried out to analyze the independent variables. There were no statistically significant differences between means. EDS analysis Analysis of EDS was based on the same method as analysis of AC. It was found that in the AK group, there was no need to reject the null hypothesis that the mean EDS values of diseased and healthy skin do not differ from each other. In the BCC group, such a hypothesis was rejected. The average EDS value for diseased skin was significantly lower than the mean EDS value for healthy skin (p , 0.0001). At the same time, the average EDS value for diseased skin in the BCC group was significantly lower than that in the AK group (p , 0.0001). The average EDS value for healthy skin in the BCC group did not differ significantly from the average EDS value for healthy skin in the AK group. Discriminant analysis Significant differences in EDS between the AK and BCC groups caused us to build a discriminatory model, the aim of which was to find a formula for classifying a given patient in the AK or BCC group on the basis of the EDS value. In discriminant analysis, a model using the classification functions under the assumption of equal a priori probabilities was applied. The classification matrix gave 100% percent compliance. The cutoff value was 1,466.
DISCUSSION AND CONCLUSIONS The goal of our study was to answer the question: Can healthy tissue, actinic keratosis lesions and basal cell carcinoma lesions be differentiated using both the backscatter properties and attenuation coefficients of skin tissue? Statistical analysis assumed the K-distribution for the envelopes of the skin backscattered echoes. That resulted from analysis of echo amplitude distributions. By comparing the Rayleigh and K-distribution models, we found that K-PDF well represented healthy dermal tissue, with the MSE eight times lower than the error calculated for Rayleigh-PDF. Our results are in agreement with the results reported by Lebertre et al. (2002), in whose study the K-distribution performed well, describing echoes from human abdominal skin samples using 35-MHz ultrasound. Other authors (Raju and Srinivasan 2002) also proved that at 30–50 MHz, skin echo amplitude departs from Rayleigh statistics. The K-distribution is particularly interesting because it allows determination of the EDS. We also found that EDS was lower for BCC lesions than for healthy skin and AK lesions (Fig. 11). The Kdistribution is sensitive to the number and uniformity of scatterers within the resolution cell. The decrease in EDS for BCC can be explained by the lower spatial density of collagen fibers (dominant scatterers) that are replaced by tumor clusters. Analogous work carried on with healthy and BCC skin samples (Petrella et al. 2012) also found that the statistical approach could be a good descriptor of tissue condition. Modeling the echo amplitude distribution with gamma and Weibull PDFs, the authors found significant differences in signals scattered in healthy and diseased skin. At the same time, histologic images revealed the dense net of collagen fibers and tumor cell clusters within the collagen fibers of healthy skin and BCC samples, respectively. In the case of AK lesions, the collagen fiber matrix in the dermis remains unaltered; thus, the scattered echo statistical properties are similar to the properties of healthy tissue (Fig. 11). The attenuation coefficient was the second parameter analyzed. Results revealed lower values for healthy skin than for BCC and AK lesions. Olerud et al. (1990)
758
Ultrasound in Medicine and Biology
reported that the amount of water contained in the skin is crucial for the attenuation of ultrasound. The more water in the skin, the lower is the attenuation coefficient. These results were also obtained by Raju and Srinivasan (2002), who determined that swelling skin, and thus tissue in which hydration is increased, had lower attenuation compared with healthy tissue. The high attenuation for AK can be explained by the low skin water content that is characteristic of this kind of lesion (King and Brucker 2011). The higher attenuation found for BCC lesions is more difficult to explain. Some (Pan et al. 1998) claim that the value of the attenuation coefficient of the skin depends on deformation of the collagen fiber matrix. They found a linear relation between the attenuation coefficient and the mechanical loading of skin samples. Changes in collagen matrix structure also occur when tumor cells and tumor cell clusters grow in the dermis, and deformation of the matrix could result in increased attenuation in BCC lesions. Although the number of cases (21 BCCs and 13 AKs) in this study is rather small to draw definite conclusions, it is worth pointing out that in all the cases considered, we obtained lower EDS values for BCC lesions and higher attenuation coefficient values for BCC and AK lesions compared with the healthy skin reference. That gives us a chance to distinguish AK cases from BCC cases because although both are characterized by a higher attenuation coefficient, only BCC has a lower EDS value. Furthermore, the discriminatory model indicated that EDS based on the statistical properties of backscatter enables differentiation of BCC and AK lesions. The cutoff value was 1,466, which means that each new patient with EDS below the cutoff value should be classified in the BCC group, whereas patients with an EDS .1,466 would be classified in the AK group. Statistical analysis also revealed the ability of a relative measure (EDSD) to differentiate BCC and AK. In conclusion, the statistical parameters, together with the attenuation coefficient, allow differentiation between basal cell carcinoma lesions, pre-cancerous lesions and healthy skin. The good performance of the described approach to tissue characterization suggests that the diagnosis of skin diseases characterized by changes in collagen matrix structure could be enhanced by ultrasonic examination. Medical consultants with whom we are working believe that the ultrasonic method can reduce the number of biopsies and increase the accuracy of estimates of the penetration depth of lesions. In cases of positive ultrasonic classification of a lesion as BCC, the lesion should be removed without confirmation of the diagnosis by biopsy. In turn, accurate evaluation of tumor tissue penetration depth is very important when planning surgical removal, resulting in maintenance of
Volume 41, Number 3, 2015
the required safety margin, on the one hand, and minimization of post-operative scarring, on the other hand. In addition, ultrasound evaluation results in a better choice of methods for removal of AK lesions. In deeper lesions, surgery is required, whereas in superficial lesions, laser therapy or cryotherapy may suffice. Acknowledgments—This work was supported in part by the Ministry of Science and Higher Education of Poland, Project N N518295140, and by National Science Centre of Poland, Project 2011/01/B/ST7/06728.
REFERENCES Bigelow T, McFarlin B, O’Brien WD, Oelze M. In vivo ultrasonic attenuation slope estimates for detecting cervical ripening in rats: Preliminary results. J Acoustic Soc Am 2008;123:1794–1800. Cham BE. Solasodine glycosides: A topical therapy for actinic keratosis. A single-blind, randomized, placebo-controlled, parallel group study with Curaderm. J Cancer Ther 2013;4:588–596. Chen X, Phillips D, Schwartz KQ, Mottley JG, Parker KJ. The measurement of backscatter coefficient from a broadband pulse-echo system: A new formulation. IEEE Trans Ultrason Ferroelect Freq Control 1997;44:515–525. Dutt V. Statistical analysis of ultrasound echo envelope. Rochester, MN: Mayo Graduate School; 1995. Fornage BD, McGavran MH, Duvic M, Waldron CA. Imaging of the skin with 20 MHz US. Radiology 1993;189:69–76. Foster FS, Pavlin CJ, Harasiewicz KA, Christopher DA, Turnbull DH. Advances in ultrasound biomicroscopy. Ultrasound Med Biol 2000;26:1–27. Fournier C, Bridal SL, Berger G, Laugier P. Reproducibility of skin characterization with backscattered spectra (12–25 MHz) in healthy subjects. Ultrasound Med Biol 2001;27:603–610. Gniadecka M, Jemec GBE. Quantitative evaluation of chronologic ageing and photoaging in vivo: Studies on skin echogenicity and thickness. Br J Dermatol 1998;139:815–821. Goodman JW. Statistical optics. New York: Wiley Classics Library; 2000. Guittet C, Ossant F, Vaillant L, Berson M. In vivo high frequency ultrasonic characterization of human dermis. IEEE Trans Biomed Eng 1999;46:740–746. Harland C, Kale S, Jackson P, Mortimer P, Bamber J. Differentiation of common benign pigmented skin lesions from melanoma by high resolution ultrasound. Br J Dermatol 2000;143:281–289. Jakeman E, Tough RJA. Generalized K distribution: A statistical model for weak scattering. J Opt Soc Am A 1987;4:1764–1772. King T, Brucker M. Pharmacology for women’s health. London: Jones & Bartlett; 2011. Lebertre M, Ossant F, Vaillant L, Diridollou S, Patat F. Spatial variation of acoustic parameters in human skin: An in vivo study between 22 and 45 MHz. Ultrasound Med Biol 2002;28:599–615. Lewandowski M, Nowicki A. High frequency imaging system with coded excitation. IEEE Trans Ultrason Ferroelectr Freq Control 2008;8:1878–1882. Litniewski J, Cieslik L, Wojcik J, Nowicki A. Statistics of the envelope of ultrasonic backscatter from human trabecular bone. J Acoust Soc Am 2011;130:2224–2232. Litniewski J, Nowicki A, Klimonda Z, Lewandowski M. Sound fields for coded excitations in water and tissue: Experimental approach. Ultrasound Med Biol 2007;33:601–607. Olerud J, O’Brien WD, Riederer-Henderson MA, Steiger DL, Debel J, Odland G. Correlation of tissue constituents with the acoustic properties of skin and wound. Ultrasound Med Biol 1990;16:55–64. Pan L, Zan L, Foster FS. Ultrasonic and viscoelastic properties of skin under transverse mechanical stress in vitro. Ultrasound Med Biol 1998;2:995–1007. Petrella LI, De Azevedo VH, Issa PR, Martins CJ, Machado CJ, Pereira WC. Statistical analysis of high frequency ultrasonic
Quantitative sonography of basal cell carcinoma d H. PIOTRZKOWSKA-WROBLEWSKA et al. backscattered signals from basal cell carcinomas. Ultrasound Med Biol 2012;38:1811–1819. Piotrzkowska H, Litniewski J, Szymanska E, Nowicki A. Ultrasonic echosignal applied to human skin lesions characterization. Arch Acoust 2012;37:103–109. Raju BI, Srinivasan MA. High-frequency ultrasonic attenuation and backscatter coefficients if in vivo normal human dermis and subcutaneous fat. Ultrasound Med Biol 2001;27:1543–1556. Raju BI, Srinivasan MA. Statistics of envelope of high-frequency ultrasonic backscatter from human skin in vivo. IEEE Trans Ultrason Ferroelectr Freq Control 2002;49:871–882. Raju BI, Srinivasan MA, Gonzalez S, Srinivasan M. Quantitative ultrasonic methods for characterization of skin lesions in vivo. Ultrasound Med Biol 2003;29:825–838. Shankar PM, Dumane VA, Reid JM, Genis V, Forsberg F, Piccoli CW, Goldberg BB. Use of the K-distribution for classification of breast masses. Ultrasound Med Biol 2000;26:1503–1510. Shankar PM, Reid JM, Ortega H, Piccoli CW, Goldberg B. Use non-Rayleigh statistics for the identification of tumors in ultrasonic B-scan of the breast. IEEE Trans Med Imaging 1993;12:687–692.
759
Tsui PH, Yeh CK, Liao YY, Chang CC, Kuo WH, Chang KJ, Chen CN. Ultrasonic Nakagami imaging: A strategy to visualize the scatterer properties of benign and malignant breast tumors. Ultrasound Med Biol 2010;36:209–217. Turnbull DH, Starkosky BG, Harasiewicz KA, Semple JI, From L, Gupta AK, Sauder N, Foster FS. A 40-100 B-scan ultrasound backscatter microscope for skin imaging. Ultrasound Med Biol 1995;21:79–88. Wagner RF, Insana MF, Brown DG. Statistical properties of radio frequency and envelope detected signals with application to medical ultrasound. J Opt Soc Am A 1987;4:910–922. Wagner RF, Smith SW, Sandrik JM, Lopez H. Statistics of speckle in ultrasound B-scans. Trans IEEE Sonics Ultrason 1983;30:156–163. Weber L, Kirsch E, M€uller P, Krieg T. Collagen type distribution and macromolecular organization of connective tissue in different layers of human skin. J Invest Dermatol Symp Proc 1984;82:156–160. Wells PNT. Biomedical ultrasonics. New York: Academic Press; 1977. Weng L, Reid J, Shankar M, Soetanto K. Ultrasound speckle analysis based on the K distribution. J Acoust Soc Am 1991;89:2992–2995. Wortsman X, Jemec G. Dermatologic ultrasound with clinical and histologic correlations. New York: Springer; 2013.
![[Superficial basal cell carcinoma versus nodular basal cell carcinoma].](/assets/img/hold.png)
