| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Developments |
1 From the Depts of Radiology (S.P.P.), Obstetrics and Gynecology (S.K.S.), and Pathology (W.A.W.), Dartmouth Hitchcock Med Ctr, One Medical Center Dr, Lebanon, NH 03756; Thayer School of Engineering, Dartmouth College, Hanover, NH (K.D.P., A.H., P.M.M., B.W.P.); and Dept of Community and Family Medicine, Norris Cotton Cancer Ctr, Lebanon, NH (T.D.T., M.R.G.). Received Apr 10, 2003; revision requested Jun 27; revision received Sep 8; accepted Oct 14. Supported by National Institutes of Health grant P01-CA80139. Address correspondence to S.P.P. (e-mail: steven.p.poplack@hitchcock.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
© RSNA, 2004
Index terms: Breast, 00.1299 • Infrared and near-infrared spectroscopy, 00.1299
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
With the hope of improving contemporary breast imaging and deepening our understanding of the biology of breast disease, we have developed three alternative imaging techniques. None of these techniques involve the use of ionizing radiation or sound waves; instead, they operate in different regions of the electromagnetic spectrum and exploit observed electromagnetic differences between normal tissue and breast malignancies (5–7). These imaging techniques are near-infrared (NIR) spectroscopy, electrical impedance (EI) spectroscopy, and microwave imaging (MI) spectroscopy.
The tissue properties involved in these imaging techniques include light scattering and absorption (at NIR spectroscopy) and permittivity and conductivity (at EI and MI spectroscopy). Permittivity reflects the capacity of tissue to store electrical charge, while conductivity is a measure of its ability to conduct current. These tissue properties can be used to estimate specific characteristics of the breast, which differ when breast cancer is present as compared with when a benign condition or normal breast tissue is present (5–7). NIR spectroscopy enables quantification of hemoglobin and deoxyhemoglobin concentrations and provides information on tissue vascularity and oxygenation. At EI spectroscopy, measures of permittivity and conductivity reflect tissue cellularity and cell membrane integrity (8). Permittivity and conductivity in the MI spectroscopic frequency range reflect tissue water content (7).
Although there have been reports on the electromagnetic property values in the breast, these data are predominantly based on results of analysis of ex vivo tissue specimens and a few in vivo studies (9,10). To our knowledge, there has been no previous evaluation of electromagnetic properties over the spectral range from the near–direct current wavelengths measured at EI spectroscopy to NIR wavelengths in asymptomatic women with normal mammographic findings. To interpret the importance of each electromagnetic property in disease states, it is critical to establish a range of normal values for each property and to determine how these properties may vary with different clinical characteristics. Thus, the purpose of our study was to provide representative data for these electromagnetic properties in the normal female breast as potential benchmarks for the future development of electromagnetic breast imaging.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
Mammography was performed with Lorad MIV dedicated screen-film units (Hologic, Danbury, Conn)—equipped with HTC grids (Hologic) and Kodak Min-R2000 film (Eastman Kodak, Rochester, NY)—and standard processing techniques. Screening mammography consisted of acquisition of standard craniocaudal and mediolateral oblique views of each breast. All subjects reported no current clinical breast concerns and no prior history of breast intervention.
The initial negative mammographic interpretation was rendered by one of several clinical radiologists who served as the primary reader. Each mammogram was evaluated independently by a second reader (S.P.P.) with 7 years of dedicated mammography experience who confirmed normality and assessed breast composition. The second reader excluded subjects if any mammographic abnormality, including definitely benign findings such as typically benign calcifications, intramammary lymph nodes, or calcified fibroadenomas, was detected. In cases in which there was a discrepancy in categorization of breast composition between the first and second readers, a third reader with 14 years of dedicated mammography experience independently reviewed the examination results and categorized breast composition.
All subjects underwent imaging with the electromagnetic techniques within 5 months of screening mammography. Our final study group included 23 women. From our clinical breast imaging database, we first generated a list of women with negative mammographic findings who were then stratified by age and breast density. Subjects from this list from each of four age groups (ie, 40–49 years, 50–59 years, 60–69 years, and 70–79 years) with varying breast compositions consisting of "almost entirely fat," "scattered fibroglandular density," "heterogeneously dense," and "extremely dense" (11) were consecutively contacted (S.K.S.). In subsequent analyses, owing to the small number of subjects in each subset, we (M.K.G., T.D.T.) collapsed these four density subsets into two larger groups: The "less dense" group included breasts that were almost entirely fat plus breasts with scattered fibroglandular density, and the "more dense" group included breasts that were heterogeneously or extremely dense.
We initially recruited 15 of 16 consecutive subjects but were unable to identify (and therefore omitted) a subject in her seventies who had extremely dense breasts. We recruited four additional consecutive subjects from the computer-generated list to fill a void resulting from reclassification of breast composition by the consensus (third) reader. As is further detailed below, the MI spectroscopic data were initially compromised in five subjects owing to faulty system hardware that went undetected during the first few examinations, leading to the recruitment of four additional consecutive women of the same age and density strata and a repeat MI spectroscopic examination in one subject. We were unable to obtain MI spectroscopic data in three subjects or recover a conductivity image in one subject due to out-of-plane effects in small breasts. In two of the four additionally recruited women, the EI spectroscopic examination had to be aborted because of a faulty electrode (in one woman) or poor grounding (in the other). This yielded a total subject population of 46 breasts in 23 women for NIR spectroscopy, 42 breasts in 21 women for EI spectroscopy, and 42 breasts in 21 women for MI spectroscopy.
An author (S.K.S.) established subject disease status through linking with a statewide mammography-pathology registry (13) and hospital information systems and performing supplementary phone interviews. (The statewide registry captures clinical history and interpretation data from mammographic encounters and data on all benign and malignant disease for participants.) One subject was interviewed by phone by our study coordinator (S.K.S.) because there was no follow-up information available for her from the other data sources. Participants were considered healthy if there was no record or self-report of benign or malignant disease. Follow-up time was recorded. Mean and median follow-up periods for disease status were 23 and 25 months, respectively.
Electromagnetic Imaging Examinations
The technical underpinnings of the three imaging techniques used in this study have been described in considerable detail elsewhere (14–16). In this section, we summarize salient features of each technique’s clinical operation and its expected performance on the basis of results of previous phantom studies (17–20). The total time involved in performing the bilateral multimodality examinations was typically 90 minutes. This included subject preparation (eg, gowning), breast positioning, movement among the equipment used for the three imaging techniques, data collection, and, often, a short break between the parts of the examinations. Examination sessions were not repeated and test-retest reproducibility was not evaluated in this study. Reports of measurement reproducibility in phantoms and clinical breast examinations have appeared elsewhere (20–22). Variation in the imaging system measurements was also not evaluated; however, at our institution, the instruments are regularly calibrated in phantom experiments to ensure consistent performance.
The three methods (NIR, EI, and MI spectroscopy) have a number of common characteristics:
Electromagnetic imaging.—Each technique involves exposing a cross-section of breast tissue to nonionizing radiation and recovering data on the electromagnetic properties of different tissue components within the path of transmission. The specific nature of these data depends on the transmission wavelength or frequency. NIR spectroscopy involves transmission of wavelengths in the nanometer span, EI spectroscopy involves transmission of frequencies in the kilohertz scale, and MI spectroscopy involves transmission of frequencies in the megahertz range.
Subject positioning.—The subject lies prone on a specially designed (in-house) examination table that has an opening for the pendant breast and is similar in configuration to a dedicated prone stereotactic biopsy table.
Interface positioning.—Source-sensor arrays (designed in-house) are positioned circumferentially around the pendant breast at a location of interest referenced (in centimeters) to the chest wall. Data are acquired for an approximately 1-cm section of breast tissue in the coronal plane. In this study, a single section at the mid-breast level was examined with NIR and EI spectroscopy, while the entire breast was evaluated with MI spectroscopy in three to seven sections, depending on breast size. In a clinical examination, the entire breast would probably be imaged with each technique.
Image reconstruction.—Electromagnetic properties are reconstructed at each frequency or wavelength on the basis of computational models of the relevant electromagnetic field interactions occurring in the breast (eg, light diffusion for NIR spectroscopy, electrical current flow for EI spectroscopy, and microwave propagation for MI spectroscopy). In this study, electromagnetic property value averages were obtained for the imaged volume. However, pixel values for each property could also be calculated and displayed as a property map image. Representative property map images obtained with each technique, along with a corresponding coronal MR image (which was not obtained as part of this study), are shown in Figure 1.
|
Data calibration and quality control.—The data obtained with each imaging system were calibrated against a known reference, and steps were taken to ensure quality control at the time of examination. In the cases of NIR and MI spectroscopy, measurements of a homogeneous reference medium (Fresenius Kabi Clayton, Clayton, NC) with known properties at all wavelengths and frequencies were recorded prior to an examination. These data were used to remove errors in source-sensor measurements caused by systematic variations in channel-to-channel characteristics (eg, cable losses and phase shift). For EI spectroscopy, the reference was an electronic circuit with an impedance in the range of that of breast tissue whose components had been independently measured. Details of the calibration process for each system have been reported (14–16). During an examination, data were monitored for sensor saturation, and gain settings were adjusted accordingly. In the cases of NIR and EI spectroscopy, where skin contact is made, the quality of the contact was evaluated through an initial (partial) data acquisition; array positioning was adjusted if contact was deemed inadequate.
NIR Spectroscopy
With this technique, a circumferential array of 16 fiber-optic bundles (Cuda Fiberoptics, Jacksonville, Fla) moves radially to directly contact the surface of the pendant breast. The fiber array accommodates breast diameters ranging between 4 and 20 cm with a 0.5-µm precision and is mounted on a stage, which moves vertically over a distance of 20 cm from the chest wall to the nipple. Sixteen separate fibers are used for light delivery and detection. At any given time during image acquisition, one fiber serves as the light source and the remaining 15 fibers act as light detectors. The light source then switches to the next delivery fiber, and the process is repeated until light signals have been recorded for all source locations around the breast. Complete data acquisition (16 sources times 15 detectors) occurs within approximately 30 seconds. Five wavelengths (660, 761, 785, 808, and 826 nm) are used during an examination. Including breast positioning time, a multispectral single-plane breast examination can be completed in 5–10 minutes.
The risks associated with an NIR spectroscopic breast examination are minimal. Thermal effects caused by tissue exposure to NIR energy have been documented, and safety limits on power density have been established (23) (standard power density: 200 mW/cm2 in this spectral band). These safety limits were not exceeded in this study; approximately 10 mW/cm2 of NIR energy was applied to the breast. Direct eye exposure to NIR light has the potential to cause vision impairment, but this risk was minimal in our study because the signals were fiber-coupled to the breast and the imaging array was shielded by the examination table and the subject’s body once she was positioned. Participants were offered protective eyewear for the examination. (All subjects in this study were informed of the risk of visual impairment but declined to use protective eyewear.)
NIR-wavelength light spreads through tissue as a diffusive wavefront, which can be mathematically modeled in terms of the wavelength-dependent optical properties of absorption and scattering. Through the use of modeling techniques, the differences in the transmitted and the detected light are related to the spatial distribution of light absorbers and scatterers in breast tissue. There are four distinct tissue elements that absorb wavelength-specific light in the breast: oxygenated and deoxygenated hemoglobin, water, and fat. By exploiting the known spectra of these primary NIR absorbers in the breast, we can generate maps of total hemoglobin concentration and oxygen saturation in each section of breast tissue.
We obtained up to 14 images (six absorption images, six scattering images, one total hemoglobin image, and one oxygen saturation image) per section of breast tissue. Image reconstruction occurred in less than 5 minutes per wavelength (with a 1-GHz processor [Dell Computer, Round Rock, Tex]). We examined each section of breast tissue with five separate wavelengths and reconstructed property map images of each section. However, we report image averages (rather than pixel values) of total hemoglobin concentration, oxygen saturation, and absorption and scattering coefficients at a single wavelength, 785 nm, as representative of the breast response.
Results of phantom studies (17,20) revealed a depth dependency in the contrast resolution and in the accuracy of imaging quantitative values of absorption, scattering, and total hemoglobin concentration. For near-surface (up to approximately 3 cm in depth) regions, high-contrast objects (ie, those with a relative contrast to their backgrounds of more than 5:1) as small as 2–3 mm can be detected but not characterized, whereas low-contrast (2:1) objects 5 mm and larger can be characterized with an error of 10%–15% in a property value. Imaging centrally (at a 5–6 cm depth) is more difficult, and performance degrades to a high-contrast detection resolution greater than 5 mm and a low-contrast characterization resolution of 1 cm (or more), with a property error of 20%.
EI Spectroscopy
The EI spectroscopic examination involves use of a 16-element array of silver and silver chloride electrodes (In Vivo Metric, Ukiah, Calif) that are brought into direct contact with, and electrically coupled to, the pendant breast with a thin layer of conductive gel (Signa gel; Parker Laboratories, Orange, NJ). The electrode array can accommodate breast diameters of 5–18 cm and can move vertically over a distance of more than 25 cm. For this study, 10 distinct voltage frequencies (10, 20, 40, 50, 70, 125, 225, 525, 750, and 950 kHz) were applied to a midcoronal section of breast tissue, and the resultant current was detected. The source-detector sequences were varied for each driving voltage in up to 15 different patterns of application. This yielded a data set of 240 voltage and current pairs. Acquisition time for a full multispectral examination (240 measurements at each of 10 frequencies) for a single imaging plane was approximately 2 minutes once the electrodes were in contact with the breast.
The risks associated with an EI spectroscopic breast examination are minimal. At the frequency band and power level used for this EI spectroscopic technique, there are no known risks to healthy volunteers. However, subjects with implanted electrical devices (eg, pacemakers or neurostimulators) were excluded from the study, since the operation of these devices could be compromised by the signals applied during an EI spectroscopic examination. The safety limit for applied current in the frequency band of operation used here (10–950 kHz) is 10 mA (23,24). The EI spectroscopy system actively monitors the measured current for each applied voltage during a frequency sweep and adjusts the levels accordingly to ensure strict adherence to the safety standard.
As at the NIR spectroscopic examination, changes in the measured currents at the EI spectroscopic examination relate to the intrinsic electrical properties of the breast. For the EI spectroscopic examination, the tissue electrical conductivity and permittivity can be estimated by mathematically modeling the current flow. Image reconstruction occurs off-line and takes approximately 30 minutes per frequency with a 1-GHz processor (Dell Computer) and nonoptimized analysis software (written in-house). Using this modeling technique, we generated absolute images of electrical conductivity and permittivity at each of the 10 driving frequencies from the applied voltages and measured current amplitudes and phases acquired during an examination. A complete EI spectroscopic examination of a single section of breast tissue yields a total of 20 images (10 conductivity maps and 10 permittivity maps). In this study, we report the mean conductivity and permittivity at two frequencies, 125 kHz and 950 kHz, as representative of the complete breast response.
Results of phantom experiments have shown (18,20) that EI spectroscopic imaging performance is also depth dependent. Study results confirm that high-contrast objects are visible 5 cm below the surface if they are 2 cm or larger. Smaller high-contrast objects (<1 cm in size) are also visible if they are located superficially (within 3 cm of the "skin" surface). Moderate-contrast (4:1) abnormalities 1 cm in size can be seen with an accuracy of 20%–25% when also located superficially under the skin. Difference imaging (ie, subtracting a phantom image with an abnormality from an image of the same phantom without the abnormality) improves visibility and localization considerably. In the case of difference imaging, 1-cm high-contrast objects are evident at depth (to approximately 8 cm), and detection of objects smaller than 5 mm is possible superficially (to approximately 2 cm of depth).
MI Spectroscopy
The MI spectroscopy system illuminates the breast over the frequency range of 300–900 MHz. It involves use of an array of 16 monopole antennas (fabricated in-house) positioned circumferentially around the pendant breast. Unlike the arrays in the NIR and EI spectroscopy systems, the array in the MI spectroscopy system is fixed at a diameter of 15.2 cm and does not directly contact the breast. For MI spectroscopy studies, the pendant breast and array are immersed in a saline bath, which serves to broaden the range of transmission frequencies and limit stray microwave radiation. The array moves vertically over a span of 6 cm and is adjusted in 1-cm increments through a range of three to seven positions (depending on breast size) from the top of the immersion tank (nearest the chest wall) to the level of the nipple. During image acquisition, a single antenna emits one of seven microwave frequencies, which is detected by the nine antennas forming an arc around the opposite side of the breast. A complete data set, sufficient for image reconstruction at seven frequencies ranging from 300–900 MHz in 100-MHz increments, consists of 144 measurements of field amplitude and phase (16 sources times nine detectors) and can be acquired in approximately 90 seconds.
The risks associated with the microwave breast examination are nominal. Safety standards exist for power density exposures over the spectral band involved, and the upper bound has been set at 1 mW/cm2 (25). Our MI spectroscopy system transmits less than 5 mW of total power, which has been determined to produce exposure levels well below this threshold (approximately 0.2 mW/cm2). There are no scientifically known biohazards from these electromagnetic transmissions at low power, although some controversy exists regarding the health effects of prolonged exposure at power line (26) and mobile communication (27) frequencies.
Microwave data are collected in a scheme analogous to the one used in computed tomography, but the reconstruction of property images relies on the same mathematic modeling techniques used with NIR and EI spectroscopic imaging. In the case of MI spectroscopy, a model describing coherent electromagnetic wave propagation in tissue is used to approximate the tissue properties of electrical conductivity and permittivity. Image reconstruction follows data acquisition, requiring approximately 10 minutes per section per frequency at a workstation (HP ES40 Alpha server; Hewlett Packard, Palo Alto, Calif).
A complete set of images from a single breast consists of up to 98 images: seven permittivity and seven conductivity images for each breast cross-section, corresponding to the seven acquisition frequencies for as many as seven sections of data. In this study, we present mean MI spectroscopic property values at a single wavelength, 600 MHz. In contrast to the NIR and EI spectroscopic examinations (which in this study were limited to measurement of a single section), the property values at MI spectroscopy were averaged over two to three mid-breast sections.
Investigators who performed phantom studies (19,20) with the MI spectroscopy system estimate its high-contrast detection resolution to be 2–4 mm, depending on depth and illumination frequency. Although depth dependency does exist, it is not as strong as that with the NIR and EI spectroscopy systems in the 300–900 MHz spectral band. However, objects in the geometric center of the imaging array are less well resolved owing to reconstruction artifacts. In general, low-contrast (2:1 and 4:1) abnormalities in the 5–10-mm range can be characterized with an error of no more than 10%–20% in the recovered electromagnetic properties. For imaging experiments in which saline is used as the coupling fluid (which creates a high contrast with the breastlike phantom regardless of the properties of the embedded abnormality), the permittivity image is more accurate (10%–15% error) than its conductivity counterpart (20%–25% error).
Statistical Methods
Subject data were summarized (T.D.T., M.K.G.) by using percentages, means, and SDs. The spatial mean of each electromagnetic property was computed by averaging the pixel values in the reconstructed breast image. For EI spectroscopy, the outer 1-cm ring of tissue was excluded at pixel averaging because of artifacts known to be associated with electrode contact with the breast surface. For MI spectroscopy, since the exact location of the breast surface was not known, the location of the breast surface was inferred on the basis of the steep gradient in electrical properties associated with the transition from breast tissue to the saline background. In contrast to NIR and EI spectroscopy, at which a single 1-cm section of tissue at the mid-breast level was examined, for MI spectroscopy, pixel values were averaged over two to three mid-level breast sections. Means and SDs for the subject pool were calculated (T.D.T., M.K.G.) according to electromagnetic property from the spatial averages over the imaged regions of the individual breasts. Correlations between right and left breast mean values for each property, along with P values that were based on the results of t testing for Pearson correlation coefficients (S-PLUS 6.0; Insightful, Seattle, Wash) (28), were calculated (T.D.T., M.K.G.). P values less than .05 were considered to indicate statistically significant differences.
Multiple linear regression models were fit (T.D.T., M.K.G.) to relate log-transformed property values to individual subject characteristics. The regression coefficients were then exponentiated to obtain the adjusted percentage change in property values corresponding to differences in subject characteristics. Because the models included both left and right breast data, a random effect for each individual was included to adjust for within-person correlation. Standard errors for effect estimates from the models were used to compute P values on the basis of a z score. The models were fit with the program "lme" from the statistical package S-PLUS 6.0 (Insightful).
Correlations between properties from different techniques, along with P values based on t test statistics for Pearson correlation coefficients (S-PLUS 6.0; Insightful) (28), were also calculated (T.D.T., M.K.G.). It should be noted that, on the basis of results of a statistical hypothesis test, a study of this size—that is, one involving 23 subjects—can enable detection of a correlation of 0.52 with a two-sided significance level of .05 and a power of .80.
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
Salient mammographic and clinical characteristics of the 23 subjects are listed in Table 1. All subjects were white, which reflects the ethnicity of the population served by our institution (ie, 98% white).
|
|
|
|
|
|
|
|
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
We stratified subjects by age and radiographic breast density, two factors that affect the accuracy of mammography (29). We sought to understand whether electromagnetic techniques are affected by these and other factors, including BMI, menopause status, and exogenous hormone use.
The two breasts of the same subject are usually symmetric at mammography. We might subsequently expect the breasts of the same subject to exhibit similar electromagnetic properties. Although repeatability was not evaluated in this analysis, the strong correlation of electromagnetic property measurements between the right and left breasts implies a high degree of consistency (ie, intrasubject repeatability) for each of these imaging techniques.
NIR Spectroscopy
NIR spectroscopy is of particular clinical interest because it can enable the detection of malignancy on the bases of tissue vascularity and oxygenation. Our NIR spectroscopic technique allowed us to map total hemoglobin concentration and oxygen saturation in each section of breast tissue. Hemoglobin concentration is an indicator of tissue vascularity, which is increased in breast malignancy (30) owing to angiogenesis. Malignant tissue is also less oxygenated, probably because of increased metabolic activity and/or an ineffective oxygen distribution system (31). The total hemoglobin concentration average of 24 µm ± 12.20 and the oxygen saturation average of 69.3% ± 9.07 observed in this study are in good agreement with values reported by other groups (9,10).
Statistically significant associations were evident between subject characteristics (ie, BMI and breast density) and total hemoglobin concentration and absorption at 785 nm. Total hemoglobin concentration and absorption at 785 nm decreased by 30% and 25%, respectively, when radiographic breast density decreased. This makes sense physiologically, assuming that blood content (and hence hemoglobin concentration and absorption) in an area of the breast correlates with blood vessel density. Results of a parallel study in which the microvasculature in histologic sections of benign and malignant breast tissue was analyzed (Wells W, Daghlian CP, Tosteson TD, et al, unpublished data) support this premise. There is an even more pronounced relationship (ie, decreases in hemoglobin concentration and absorption of 51% and 47%, respectively, at 785 nm) with increasing BMI. There was also a significant (P = .02) decrease in the scattering coefficient at 785 nm with increasing BMI. Assuming that higher BMI correlates with greater amounts of adipose tissue (12), the 29% decrease in average scattering coefficient suggests that fat is less scattering than glandular tissue. This trend is consistent with results of previous studies (6,9).
No statistically significant associations were observed between oxygen saturation and subject characteristics. This lack of correspondence suggests that tissue composition may not influence the degree to which blood is deoxygenated in the normal breast. However, oxygen saturation may have relevance in the disease states. Tromberg and colleagues (31) reported an average decrease in oxygen saturation in cancers relative to oxygen saturation in normal breast tissue.
EI Spectroscopy
EI spectroscopy has clinical appeal because it enables the estimation of electrical properties that appear to be associated with tissue cellularity and cell membrane structure. Cancer is more cellular than benign breast disease or normal tissue (Wells W, Daghlian CP, Tosteson TD, et al, unpublished data). Cell membrane structure and function also change with malignancy (32). The EI spectroscopy system that we used is different from available systems that involve use of electrical current or voltage to characterize breast disease (33).
The average property trends at EI spectroscopy that we observed are generally consistent with results of a small number of ex vivo breast specimen studies (5,32,34). In the present study, conductivity was not observed to be frequency dependent over the range of 125–950 kHz but rather was centered on values near 0.46 
–1 · m–1, with SDs on the order of 25% of the mean. Permittivity means varied more dramatically both between subjects at the same frequency (with SDs of approximately 35%–40% of the means) and as averages at different frequencies (with an approximately 40% decrease in mean value). The significant permittivity decrease with frequency and the corresponding modest increase in conductivity observed in our study follow the overall dispersion characteristics noted in previous studies (34,35), although our values for relative permittivity are considerably higher than those noted in the previous studies. However, our values for relative permittivity represent in vivo averages estimated from tomographic images, whereas the comparable studies in the literature involved ex vivo specimen measurements. The higher values for relative permittivity are consistent with some data for fat (36).
There was a statistically significant decrease of 35% in relative permittivity at 125 kHz with decreasing breast density. Assuming that radiographic categorization is a measure of the predominant tissue constituents (fat vs fibroglandular tissue), one would expect the less-dense breast to have less conductivity and lower permittivity relative to the denser breast. The one exception in the data reported here is relative permittivity at 950 kHz, which showed little change with density. This observation remains unexplained but does not constitute a statistically significant finding. Like previous researchers (5,34), we observed small age-related variations (3%–23%) in properties at EI spectroscopy that were not statistically significant. Unlike at NIR spectroscopy, at EI spectroscopy, changes in property values related to breast density were not paralleled by similar variations with BMI. The large and significant increase (207%, P < .001) in relative permittivity at 950 kHz with increasing BMI remains unexplained. The conductivity changes with increasing BMI were small (ie, less than 20%) and not statistically significant.
MI Spectroscopy
The MI spectroscopic technique evaluated in this study is quite different from its progenitor, breast thermography. Like NIR and EI spectroscopy, it involves the use of mathematic modeling to measure the dielectric properties of conductivity and permittivity in a tomographic section of breast tissue. Conductivity and permittivity in the microwave spectrum (ie, at MI spectroscopy) mirror tissue water content. This is relevant because breast cancer appears to have higher water content than benign or normal tissue (37).
In this study, for MI spectroscopy, conductivity averages at 600 MHz were 0.28 –1 · m–1, with a 25% SD relative to the mean, while the permittivity averages were 16.2 ± 7.16 relative to air. These values are consistent with those in a previous report (7).
We observed statistically significant decreases of 34%–40% in both conductivity and permittivity with decreasing breast density. Like those at NIR spectroscopy, the property values at MI spectroscopy generally decreased (2%–11%) with increasing age and BMI (37%–46%) but not to statistically significant levels. To the best of our knowledge, no previous studies have involved the investigation of electromagnetic property changes with age over the frequency range of 300–3,000 MHz. These trends suggest a frequency-dependent (ie, at 600 MHz) reduction in conductivity and permittivity with increasing fat content.
Intermodality Correlations
Scatterplots of alternative property measures exhibited a wide range of correlations with varying levels of statistical significance. Figure 4 presents a subset of examples illustrating this range. The strongest correlation (r = 0.80, P < .01) occurred between total hemoglobin concentration at NIR spectroscopy and conductivity at MI spectroscopy. This implies that MI spectroscopic properties are also sensitive to local increases in blood volume related to a higher blood vessel density that is probably due to increased ionic concentration and/or water content. However, neither EI spectroscopic (r = –0.02) nor MI spectroscopic (r = –0.19) properties correlated with oxygen saturation at NIR spectroscopy, suggesting that oxygenated blood and deoxygenated blood are electrically similar over these respective frequency bands.
Although EI spectroscopy and MI spectroscopy involve measurement of the same electromagnetic properties (ie, conductivity and permittivity), the modest correlations (r = 0.52 for relative permittivity) between these modalities imply that somewhat different anatomic or electrophysiologic mechanisms are responsible for the underlying tissue interactions. For example, permittivity and conductivity values at EI spectroscopic frequencies are dramatically different in neoplastic tissue from their values in normal (electrically polarized) epithelial cells (8). These tissue-specific differences probably result from differences in membrane potentials, which appear to be altered or completely lost in cancer (38). For MI spectroscopy, water content, rather than cellular membrane structure, seems to account for differences between normal and malignant tissue. The proportions of free and membrane-bound water molecules respond differently when exposed to an applied high-frequency electromagnetic field at microwave wavelengths. Alterations in vascularity, cellularity, and stromal characteristics observed in breast malignancy may induce changes in water molecules that would translate to localized changes in conductivity and permittivity at MI spectroscopy.
Study Limitations
The imaging systems used in our study are technically complex, not yet standardized, and only available in a few laboratories. Their clinical utility is unproved, and their promise for improving the detection and/or diagnosis of breast disease is based on results of case studies and ex vivo specimen measurements. Their clinical implementation has practical constraints; because examinations are performed with the patient in the prone position, some axillary and posterior tissue adjacent to the chest wall cannot be evaluated. Each modality appears to have a resolution limit (see Materials and Methods section). At NIR spectroscopy, the NIR emissions cannot penetrate a very thick breast (ie, a breast with a diameter > 12 cm). Finally, our data are based on a relatively small sample size of all white women and may not be representative of data in women of other ethnic groups. Nonetheless, the results appear consistent with past experience. They potentially provide an important baseline against which results of future studies can be compared. Hopefully, future research would include larger, multicenter evaluations with standardized instrumentation in an ethnically diverse population.
Radiologists are aware of the shortcomings of mammography and US. In one recent report, more than 25% of asymptomatic breast cancers were not detected with mammography, while more than 75% of women undergoing breast biopsy did not have cancer (4). Mammography exposes patients to ionizing radiation and discomfort from compression, both of which probably contribute to noncompliance with screening recommendations. Nontraditional breast imaging may improve these deficiencies and allow us to better understand breast disease at the cellular or molecular level.
We have presented average data on 10 electromagnetic property measures in a common cohort of healthy subjects. For NIR spectroscopy (at 785 nm), data include the means, medians, SDs, and right-to-left breast correlations of total hemoglobin concentration, oxygen saturation, and absorption and scattering coefficients; for EI spectroscopy (at 125 and 950 kHz) and MI spectroscopy (at 600 MHz), data include the means, medians, SDs, and right-to-left breast correlations of conductivity and permittivity.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: BMI = body mass index, EI = electrical impedance, MI = microwave imaging, NIR = near-infrared
Author contributions: Guarantors of integrity of entire study, S.P.P., K.D.P.; study concepts and design, S.P.P., K.D.P.; literature research, S.P.P., K.D.P., A.H., P.M.M., B.W.P., T.D.T., W.A.W.; clinical studies, S.P.P., K.D.P., A.H., P.M.M., B.W.P., S.K.S., W.A.W.; experimental studies, all authors; data acquisition, A.H., B.W.P., P.M.M., T.D.T., M.R.G., S.K.S.; data analysis/interpretation, T.D.T., M.R.G., K.D.P.; statistical analysis, T.D.T., M.R.G., K.D.P.; manuscript preparation and definition of intellectual content, all authors; manuscript editing, S.P.P., K.D.P., T.D.T., M.R.G., W.A.W.; manuscript revision/review and final version approval, S.P.P., K.D.P.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. P. Poplack, T. D. Tosteson, W. A. Wells, B. W. Pogue, P. M. Meaney, A. Hartov, C. A. Kogel, S. K. Soho, J. J. Gibson, and K. D. Paulsen Electromagnetic Breast Imaging: Results of a Pilot Study in Women with Abnormal Mammograms Radiology, May 1, 2007; 243(2): 350 - 359. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. B. Wolbarst and W. R. Hendee Evolving and Experimental Technologies in Medical Imaging Radiology, January 1, 2006; 238(1): 16 - 39. [Abstract] [Full Text] [PDF]
|
|
| <||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
