Using Near-Infrared Light To Detect Breast Cancer Sergio Fantini, Erica L. Heffer, Horst Siebold and Oliver Schütz 24 Optics & Photonics News November 2003 1047-6938/03/11/0024/6-$0010 Optical Society of America
Optical imaging is a safe, painless technique that may become a valuable clinical tool for the detection and diagnosis of breast cancer. We present an approach to optical mammography based on intensity-modulated illumination and phase-sensitive detection (frequency-domain optical mammography). We show how the diagnostic value of single-wavelength images (which display the blood distribution and blood-vessel architecture within the breast) can be enhanced by multi-wavelength information associated with the local oxygenation level of the blood. The idea of using light to noninvasively detect breast cancer has been revisited in the past few years as a result of theoretical and experimental advances in optical spectroscopy and imaging of the human breast. Optical imaging benefits from the high optical contrast associated with spatial inhomogeneities in the concentration and oxygenation of blood within the breast, although spatial resolution is limited by the diffusive nature of light propagation in breast tissue. Optical imaging is also sensitive to water and lipid concentration in breast tissue. The sensitivity to blood concentration, blood oxygenation, water content and lipids accounts for the unique diagnostic features of optical imaging compared to established breast imaging modalities. In the future, the introduction of optical contrast agents capable of selectively labeling breast tumors may open new opportunities in the optical detection of breast cancer. In 2003, it is estimated that there will be approximately 180,000 new cases of breast cancer and just over 40,000 deaths from the disease in the United States. Breast cancer is the most common form of cancer in women and the second leading cause of cancer death in women (after lung cancer). 1 Since early detection is key to successful treatment, improving the effectiveness of diagnostic tools can result in a better clinical outcome for patients. Optical mammography, either alone or in conjunction with other imaging modalities, is a promising technique whose effectiveness and potential role in clinical practice is currently being explored in a number of research laboratories worldwide. X-ray mammography X-ray mammography is currently the primary method for breast cancer screening. Recent randomized clinical trials have shown that screening mammography has reduced mortality from breast cancer by 25 percent to 30 percent in women between the ages of 50 and 70, and by 18 percent among women between the ages of 40 and 50. 1 Despite the key role played by X-ray mammography in the detection of breast cancer, some limitations have been acknowledged by the medical community. The interpretation of mammograms is difficult in that it involves the identification of tissue abnormalities such as microcalcifications, architectural distortions, asymmetrical densities, masses and densities that have developed since the previous mammogram. 1 The accuracy of the identification depends heavily on the interpreter s experience and skill. Furthermore, mammograms from women who have radiodense breast tissue (a condition that has been linked to an increased risk for breast cancer) are even more difficult to interpret and in some cases cannot be evaluated because of an insufficient level of X-ray transmission. The high incidence of false positive results with X-ray mammography accounts for the fact that approximately three-fourths of all biopsies end up being benign. In addition to the limitations of X-ray mammography, other issues pertaining to the increased health risks associated with the patient s exposure to X-ray radiation are relevant. In certain populations, particularly those with mutations of BRCA1 and BRCA2 (the genes responsible for breast cancer susceptibility), repeated exposure to X-ray radiation might actually increase the risk of developing breast cancer. Efforts are being made to improve the performance of X-ray mammography. For example, full-field digital mammography enables physicians to manipulate contrast and magnification with one exposure 1 and allows for computer-aided detection and diagnosis. This method has been shown to improve upon radiologists readings by increasing the number of cancers found in radiographically dense breasts, but still must be accompanied by an experienced professional s interpretation to reduce the incidence of false positive findings. Other techniques for breast imaging Ultrasound imaging is commonly used as an adjunct to X-ray mammography in determining whether lesions that appear to be suspicious in X-ray mammograms are solid masses or fluid-filled cysts. Magnetic resonance imaging, performed after the intravenous administration of a gadolinium-based contrast agent, is being considered as a tool to discriminate benign from malignant breast disease and to screen women who are at high risk of developing breast cancer (because they carry mutations in BRCA1 or BRCA2) or who have radiographically dense breasts. Scintimammography is based on detecting the radiation generated by radionuclides that concentrate in breast malignancies or provide indications on the level of metabolic activity. Aside from the fact that this method requires the injection of a radioactive material with associated radiation health risks similar to those associated with X-ray mammography, it is unable to detect cancers smaller than 1 cm in size [Ref. 1] and its clinical utility is still being evaluated. Newer technologies are also being investigated. Among these are: infrared (Facing page) The prototype for frequency-domain optical mammography, developed by Siemens AG, Erlangen, Germany. Here, the subject s breast is examined in the craniocaudal projection. November 2003 Optics & Photonics News 25
a (mm -1 ) 0.1 1 01 Hb HbO 2 Fat Water 001 650 700 750 800 850 900 950 1000 thermography, which measures an increase in surface temperature caused by the increased metabolic rate of the underlying tumor; microwave imaging, which is based on the dielectric contrast associated with different levels of water content; electrical impedance imaging; and opto-acoustic tomography. Near-infrared imaging of the breast Optical imaging, the research focus of our group, is another alternative technology for breast imaging and cancer detection. Optical breast imaging uses nearinfrared light, typically in the wavelength range of 650 nm to 1,000 nm. In this spectral region, the optical absorption of breast tissue is sufficiently low to allow adequate light transmission through the whole breast. The optical properties of the breast in this spectral region are mainly determined by the concentration and oxygenation of hemoglobin and by the water and fat content of breast tissue. Figure 1 shows the near-infrared absorption spectra of deoxy-hemoglobin (Hb), oxyhemoglobin (HbO 2 ), fat and water for typical concentrations in breast tissue. 2 Wavelength (nm) Figure 1. Absorption spectra of the four dominant near-infrared absorbing species in the human breast: deoxy-hemoglobin (Hb), oxy-hemoglobin (HbO 2 ), fat and water. These absorption spectra refer to typical concentrations in the breast of 10 M for Hb, 15 M for HbO 2, 80% (by volume) for fat and 15% (by volume) for water. [Reproduced with permission from Ref. 2.] The idea of using light to image breasts is not new. It was first proposed in the late 1920s, when a broadband light source was used in transmission geometry and the breast was visually inspected on the side opposite the source. In the 1970s and 1980s, technical improvements such as the use of sensitive video cameras as detectors led to a renewed interest in breast transillumination techniques. After considerable evaluation of the results by the medical community, those transillumination techniques were abandoned because of their inadequate clinical performance. The major difficulty with breast transillumination results from the fact that near-infrared light is strongly scattered within tissue. Specifically, the diffusive nature of light propagation in breast tissue limits the spatial resolution of optical mammography and the sensitivity of the technique to deep tumors. The diffusive nature of light propagation in breast tissue also complicates the spectroscopic determination of the concentrations of the various absorbing species (deoxy-hemoglobin, oxy-hemoglobin, fat, water) within the breast. Diffuse optical imaging and spectroscopy of the breast have recently improved as a result 2000 Neoplasia Press Inc. of the application of theoretical models (such as transport theory and diffusion theory) to describe light propagation in tissue, and thanks to the introduction of time-resolved experimental techniques in the time domain (with pulsed light sources) or frequency domain (with intensity-modulated light sources). Another recent development involves the optical study of dynamic features that can identify and monitor a number of physiological phenomena, including those associated with cancer. 3 In this article we describe a frequency-domain research prototype for optical mammography developed by Siemens AG, Medical Solutions (Erlangen, Germany), and some of the image processing techniques that exploit the spatial and spectral information content of the raw optical data to maximize the diagnostic value of the breast images. Frequency-domain optical mammography The opening image (page 24) shows the configuration for a breast measurement using the prototype of a frequencydomain imager. 4 The patient s breast, placed between two glass plates, is lightly compressed to maintain stability. The glass plates are oriented horizontally to image the breast in a craniocaudal view; they can be rotated by 45 degrees or 90 degrees to collect images in oblique or mediolateral views. Figure 2 shows a block diagram of the instrument. The illumination and collection optical fibers, placed respectively on the top and bottom plate, are scanned in tandem in a raster pattern across the breast. A complete scan of the breast takes 2 minutes to 4 minutes depending on the size of the breast. Four laser diodes emitting at 690, 750, 788 and 856 nm are used to image the breast, while a fifth laser at 1310 nm is used to detect the edge of the breast and drive the breast scanning. All five lasers are coupled to the illumination fiber. The intensities of the four measurement lasers are modulated at a frequency of about 70 MHz, each slightly offset from the others (69.50 MHz, 69.80 MHz, 70.20 MHz, 70.45 MHz), while the probe laser is modulated at 500 khz. A small fraction (~ 5%) of the transmitted signal 26 Optics & Photonics News November 2003
collected by the collection fiber is sent to a silicon photodiode detector. The 500 khz signal (corresponding to the 1310 nm pilot laser) is used to drive the optical scanning by stopping the acquisition along a line when the signal exceeds a given threshold. The majority of the detected signal is sent to a photomultiplier tube detector, the output of which is analyzed by use of a homodyne-inquadrature scheme to yield the amplitude and phase of the intensity-modulated data at the four wavelengths. The parallel plate geometry used here is not ideal for image reconstruction. Indeed, the data collected in the configuration shown in Fig. 2, where the illumination and collection fibers are collinear, provide two-dimensional projection images of the breast but do not lend themselves to image reconstruction. Offaxis illumination and collection fibers improve the feasibility of image reconstruction and depth discrimination in this parallel plate geometry; a circular geometry (with optical fibers placed around the breast) has been shown to be better suited to image reconstruction. 5 There are, however, three major advantages associated with the parallel plate geometry used here that render it an attractive approach. High reproducibility The stable and reproducible breast positioning during the exam accounts for highly reproducible results on any given patient. No need for calibration Because scanning involves a single illumination fiber and a single collection fiber (coupled to a single optical detector), there is no need to calibrate to correct for the individual throughput of multiple optical fibers or the individual sensitivity of multiple detectors. The lack of a mechanical contact between the optical fibers and the breast also guarantees a relatively uniform optical coupling with the breast throughout the scanning of the fibers. Fine spatial sampling rate Spatial scanning allows for a relatively fine spatial sampling rate of 0.5-1.0 mm -1 which, for practical Laser diodes 690 nm 750 nm 788 nm 856 nm 1310 nm Pilot laser 500 khz 70.45 MHz 70.20 MHz 69.80 MHz 69.50 MHz Frequency synthesizers Homodyne in quadrature detection reasons, is harder to achieve in configurations based on the use of multiple illumination/collection optical fibers. Correction of edge effects: N-images The fact that the breast is compressed lightly between the glass plates in optical mammography (as opposed to the firm compression used in X-ray mammography) accounts for a strong variability in the thickness of the breast between the plates. Anterior and lateral tissues, closer to the edge of the projected breast image, are thinner than the base. The optical data collected closer to the breast edge are significantly affected by these geometrical factors, which are referred to as edge effects. Taking advantage of frequencydomain data, we have combined the amplitude and phase images into a socalled N-image, which corrects for edge effects, enhances the contrast of breast lesions and extends the useful imaging area to the whole breast. 6 The correction is based on the observation that diffusion theory predicts an approximately linear dependence of the phase on the tissue thickness and a much stronger phase sensitivity to the thickness than to spatial changes in the optical properties of the tissue. As a result, the phase can be used to determine the tissue thickness at any Illumination fiber Scanner driver X-Y scanner X-Y scanner 500 khz ~5% Collection Photodiode fiber ~95% Photomultiplier tube Data processing Figure 2. Block diagram of the prototype for frequency-domain optical mammography. given point. Because we have empirically found that, on average, the amplitude is inversely proportional to tissue thickness (since the stronger dependence of amplitude on tissue thickness predicted by diffusion theory is partially compensated by increased photon losses through the side of the breast at areas close to the breast edge), we have introduced an edge-corrected parameter N,defined as: N(x,y) = r 0 A 0 /[r(x,y)a(x,y)], where r(x,y) and A(x,y) are the tissue thickness and the amplitude, respectively, at pixel (x,y), r 0 is the maximal breast thickness and A 0 is the amplitude at a reference pixel. Figure 3(a) shows the images based on raw amplitude and phase data (at 690 nm) on a representative patient (n. 184, 74 years old) with a cm cancer in the right breast. The edge effects cause a large dynamic range in the rawdata images, with high amplitude and low phase values at the breast edge as a result of the higher transmission and lower optical path length through thinner tissue regions. The tumor, which is indicated by the arrow, is visible with low contrast in the amplitude image but not visible at all in the phase image. Figure 3(b) shows the edge-corrected N-image (at 690 nm), which does not present the drastic variations at the breast edges as the amplitude image does. In November 2003 Optics & Photonics News 27
addition, the cancer location (arrow) corresponds to a dark region indicating an increase in optical attenuation, and some other regions within the breast show increased attenuation as well. Enhanced display of spatial features: N -images To improve the visibility of more detailed structures, we further processed the (a) Raw data Amplitude (690 nm) 1 1 (c) Enhancement of spatial features 856 nm 788 nm (d) Spectral information Phase (690 nm) 1 1 1 1 750 nm 690 nm Low Oxygenation image Oxygenation index High N-image using an algorithm based on a second derivative operator similar to those commonly employed for edgedetection. The enhanced display of spatial structures within the breast is visible in the second-derivative N -images displayed in Fig. 3(c). The N -images were created by means of the following steps. 7 First, the original N-images were smoothed with a low-pass spatial filter. (b) Correction of edge effects 1 N -images N-image (690 nm) 1 1 1 Figure 3. Illustration of image processing steps in our approach to optical mammography. (a) Raw amplitude and phase data. (b) Correction of edge effects by combination of amplitude and phase data into a single parameter that we call N. (c) Enhancement of the display of spatial features by a spatial second-derivative operator. (d) Spectral analysis of the optical data used to generate images of the relative oxygenation over the breast. Figure depicts the craniocaudal projection of the right breast of a 74-year-old woman with a cm cancer. Next, the second derivative was calculated in four directions: horizontal, vertical and along the two diagonals using a forward-difference discrete approximation of the second derivative. We took the minimum value of the second derivative calculated along the four directions to enhance the visibility of directional structures such as blood vessels. The secondderivative images display pixels with a negative second derivative (i.e., attenuation peaks) in grayscale, while pixels with a positive second derivative are set to white, and the outside of the breast is set to black. The N -images displayed in Fig. 3(c) do show much more structural detail with respect to the N-image of Fig. 3(b). However, the additional detail, by itself, does not necessarily aid in identifying the tumor. Additional steps are required to differentiate the various structures visible in the second-derivative images. This can be done by use of spectral (or multiwavelength) information associated with the level of blood oxygenation in the breast. Spectral information: oxygenation-index images We calculated an oxygenation index by performing a least-squares fit of the second derivative values of N(N ) at four wavelengths with a linear combination of the extinction spectra of oxy-hemoglobin and deoxy-hemoglobin. Since this fit provides a relative measurement of hemoglobin oxygenation on an arbitrary scale, we refer to the recovered parameter as an oxygenation index rather than as the oxygen saturation of hemoglobin. 8 The oxygenation index is only calculated at pixels where the N value is negative at all four wavelengths, indicating an area associated with a peak in attenuation. The oxygenation-index image shown in Fig. 3(d) provides an indication of the effectiveness of this method in differentiating between cancerous and healthy tissue. The region in which the tumor is located shows a clustered region of low oxygenation index values [blue/green in Fig. 3(d)]. A more challenging case: a tumor < 0.5 cm in size Figure 3 illustrates the main steps performed to analyze the frequency-domain 28 Optics & Photonics News November 2003
optical data to display spatial features and relative levels of oxygenation within the breast. The specific clinical case reported in Fig. 3 refers to a relatively large tumor, cm in size. Figure 4 illustrates the potential of this approach in detecting smaller tumors: it shows an oblique view of the right breast of a 47-year-old patient with an invasive ductal carcinoma, < 0.5 cm in size, at a location indicated by the arrow. The edge-corrected N-image [Fig. 4(a)] shows several blurred structures, which are better discriminated in the enhanced secondderivative image [Fig. 4(b)]. The cancer is not readily identifiable in the N and N images, while it appears as a highlighted hypoxic region in the oxygenation-index image based on spectral information. This method of detection relies on the sensitivity of near-infrared light to blood concentration in breast tissue and on the potential of spectrally determined oxygenation data in discriminating tissue regions having different metabolic rates of oxygen. These characteristics may make it useful in differentiating between cancerous lesions, benign lesions and normal tissue structures in optical mammograms. Exogenous contrast agents for optical mammography The images presented in this article are exclusively based on the intrinsic optical contrast in the human breast, which in the near-infrared is mostly associated with oxy-hemoglobin, deoxy-hemoglobin, water and lipids. Despite the encouraging results presented in this article, it is has not yet been established that intrinsic optical contrast agents are sufficient to effectively perform optical imaging of the breast for cancer detection. Use of extrinsic contrast agents although it requires an intravenous injection can offer unprecedented opportunities and possibly lead to a more powerful approach (both in terms of sensitivity and specificity) to the optical detection of breast cancer. For example, it is possible to detect specific enzyme activity by use of autoquenched near-infrared fluorescence probes, the fluorescence emission of which is restored in the presence of enzymes that are overexpressed by tumors. In animal studies, it has been 18 16 14 12 10 8 6 4 2 0 (a) N-image (b) N -image (c) Oxygenation image 0 2 4 6 8 Figure 4. (a) Edge-corrected image (N-image), (b) second-derivative image (N ) and (c) oxygenation-index image of the right breast, oblique projection, of a 47-year-old woman with a cancer less than 0.5 cm in size. Cancer location is indicated by the arrow. shown that such fluorescence signals can be used to detect submillimeter sized breast tumors in vivo up to a depth of 7 millimeters to 10 millimeters (which it should be possible to extend to 5 centimeters to 6 centimeters as needed for imaging of the human breast). 9 Other approaches aim at detecting near-infrared fluorescent dyes (typically, the clinically approved indocyanine green or structurally related dyes) that exhibit preferential accumulation in cancerous tissue. 10 Conclusion Breast cancer affects a large number of women worldwide and is a leading cause of cancer death among women. The gold standard for screening, X-ray mammography, has been proven to decrease mortality rates in certain age groups, although it is characterized by certain recognized limitations. Optical imaging is one of the emerging technologies that might play an important role in screening and clinical applications, either as a stand-alone technique or in conjunction with other imaging modalities. Intrinsic optical contrast alone may be enough to detect breast tumors and to differentiate malignant and benign lesions. However, this requires taking full advantage of the 18 16 14 12 10 8 6 4 2 0 0 2 4 6 8 18 16 14 12 10 8 6 4 2 0 0 2 4 6 8 Low High Oxygenation index sensitivity to blood/water/fat content in breast tissue, as well as the oximetry capabilities of optical mammography. The introduction of extrinsic contrast agents, while incompatible with broad screening applications, may open new opportunities for optical mammography and further improve the performance of a potentially powerful clinical tool. Acknowledgments We acknowledge support from the National Science Foundation, award BES-93840. References 1. Mammography and Beyond: Developing Technologies for the Early Detection of Breast Cancer, S. J. Nass et al., Eds, (National Academy Press, Washington, D.C., 2000). 2. B. J. Tromberg et al., Neoplasia 2, 26-40 (2000). 3. H. L. Graber et al., IEEE Trans Med Imaging 21, 852-66 (2002). 4. L. Götz et al., Akt. Radiol. 8, 31-3 (1998). 5. B. W. Pogue et al., Optics Express 4, 270-86 (1999). 6. S. Fantini et al., Med. Phys. 23, 149-57 (1996). 7. V. E. Pera et al., J. Biomed. Opt. 8, 517-24 (2003). 8. S. Fantini et al., Proc. SPIE 4955, 183-90 (2003). 9. R. Weissleder et al., Nat. Biotechnol. 17, 375-8 (1999). 10. V. Ntziachristos et al., Proc. Natl. Acad. Sci. USA 97, 2767-72, (2000). Sergio Fantini (sergio.fantini@tufts.edu) and Erica L. Heffer are with the Tufts University Department of Biomedical Engineering and Bioengineering Center, Medford, Mass. Horst Siebold and Oliver Schütz are with Siemens AG, Medical Solutions, Erlangen, Germany. Tell us what you think: http://www.osa-opn.org/survey.cfm November 2003 Optics & Photonics News 29