Special Section on Clinical Near-Infrared Spectroscopy and Imaging

Review of optical breast imaging and spectroscopy

[+] Author Affiliations
Dirk Grosenick, Herbert Rinneberg

Physikalisch-Technische Bundesanstalt, Abbestrasse 2-12, 10587 Berlin, Germany

Rinaldo Cubeddu, Paola Taroni

Politecnico di Milano, Dipartimento di Fisica, Piazza Leonardo da Vinci 32, 20133 Milano, Italy

J. Biomed. Opt. 21(9), 091311 (Jul 11, 2016). doi:10.1117/1.JBO.21.9.091311
History: Received January 15, 2016; Accepted June 13, 2016
Text Size: A A A

Open Access Open Access

Abstract.  Diffuse optical imaging and spectroscopy of the female breast is an area of active research. We review the present status of this field and discuss the broad range of methodologies and applications. Starting with a brief overview on breast physiology, the remodeling of vasculature and extracellular matrix caused by solid tumors is highlighted that is relevant for contrast in optical imaging. Then, the various instrumental techniques and the related methods of data analysis and image generation are described and compared including multimodality instrumentation, fluorescence mammography, broadband spectroscopy, and diffuse correlation spectroscopy. We review the clinical results on functional properties of malignant and benign breast lesions compared to host tissue and discuss the various methods to improve contrast between healthy and diseased tissue, such as enhanced spectroscopic information, dynamic variations of functional properties, pharmacokinetics of extrinsic contrast agents, including the enhanced permeability and retention effect. We discuss research on monitoring neoadjuvant chemotherapy and on breast cancer risk assessment as potential clinical applications of optical breast imaging and spectroscopy. Moreover, we consider new experimental approaches, such as photoacoustic imaging and long-wavelength tissue spectroscopy.

Figures in this Article

Breast cancer is the most common cancer of women and the second leading cause of death after cardiovascular disease in many countries. About one in eight women will be diagnosed with this disease during their lifetimes in the United States and in European countries.1,2 Early diagnosis of breast cancer is essential to ensure a high chance of survival for the affected women. Therefore, an important task is to provide diagnostic tools with high sensitivity for early detection of breast cancer and high specificity to avoid false positive results.

Today’s first line imaging modality is x-ray mammography. Several countries have implemented mammography screening programs with the aim of early detection of the disease. However, the sensitivity of conventional x-ray mammography is only about 75%.3 For radiographic dense breast tissue, which is more common in younger women, the sensitivity can even drop below 50%.3 Considering that frequent x-ray exposure can promote the development of cancer on a long time scale mammographic screening is often discussed controversially. Magnetic resonance imaging (MRI) of the breast offers higher sensitivity, yet its specificity is rather poor.4,5 Moreover, it is characterized by high costs and long examination times prohibiting screening. Currently, this method is in use as an additional modality for selected women with suspicious lesions. Similarly, breast ultrasound is often used as a supplementary tool. Its results are strongly dependent on the examiner’s interpretation. In general, biopsies, which are at the end of today’s breast cancer detection clinical work flow, show a large number of false-positive cases for the established imaging modalities.6,7

Besides the problem of detection and differentiation, tools are needed to support or monitor the therapy of breast cancer. A main part of treatment is the surgical removal of the cancerous tissue. The today’s preferred way of breast-conserving surgery requires methods to safely detect tumor margins. Furthermore, the treatment of large cancers often starts with neoadjuvant chemotherapy (NAC) to shrink the size of the tumor before surgery. This shrinking process needs to be monitored by a suited method. Currently, MRI is most often used to this end.8 However, a less costly method not requiring application of contrast agents is highly desirable, in particular for frequently repeated use.

During the last 15 years, optical imaging of the breast has been investigated by many research groups as well as by companies in order to develop tools that could yield considerable contributions to the mentioned steps in breast cancer management, i.e., to the detection of breast cancer, to its differentiation, and to monitor its treatment. Starting from the idea to detect breast cancer by near-infrared spectroscopy from contrast in hemoglobin concentration and blood oxygen saturation, the potential of optical imaging of the breast can now be realistically assessed due to the large amount of data obtained from clinical studies. However, definite conclusions are still difficult to draw, as clinical studies were typically conducted with different instruments, different methods of data analysis, and various clinical protocols, making results often difficult to compare. Although the original aim of developing a new tool for screening that could compete with x-ray mammography could not be reached so far, optical breast imaging has found potentially new areas of application, such as monitoring of neoadjuvant therapy progress and determination of risk populations for breast cancer development. Furthermore, the application of contrast agents seems to be a promising way for breast cancer imaging.

In this paper, we review the development of optical breast imaging by near-infrared spectroscopy, including contrast agent enhanced methodology. We start with a discussion of breast physiology and with the main features of breast imaging instrumentation and data analysis approaches. Then, we consider the available results on optical properties of the healthy breast and of malignant and benign lesions and the various methods to improve the contrast between healthy and diseased tissue. We review research on monitoring neodjuvant chemotherapy and on risk assessment as potential applications. And finally, we consider new experimental approaches, such as photoacoustic imaging (PI) and long wavelength broadband spectroscopy.

Biomedical Background
Overall structure and composition of normal breast tissue

Optical mammography and diffuse optical spectroscopy of the human breast probe the absorption, scattering, and fluorescence properties of various components of healthy and diseased breast tissue. The human female breast consists mainly of glandular, adipose, and connective tissue, together with blood and lymphatic vessels, and it contains several simple mammary glands (lobes), each draining through a separate lactiferous duct. The lobes branch into several lobules consisting of intralobular ducts separated by rather loose (intralobular) connective tissue, containing microvasculature and small lymphatic channels. Each intralobular ductal tree terminates in a cluster of alveoli that will differentiate to produce milk on exposure to lactogenic hormones. The intralobular connective tissue and ductal network are surrounded by the interstitial connective tissue, being dense, less cellular, containing variable proportions of adipose tissue and extracellular matrix (ECM), and representing over 80% of the human breast volume.9

Besides blood and lymphatic vessels, the connective tissue (stroma) consists of various stromal cells and the ECM (see Fig. 1), providing a scaffold for stromal cells (e.g., fibroblasts, adipocytes, cells of the immune system, e.g., lymphocytes, macrophages).10 The ECM is composed of water, proteins, and polysaccharides.11 Proteoglycans (PGs) and fibrous proteins (e.g., collagens, fibronectin, elastins) are the two main classes of macromolecules of the ECM, forming an intricate interlocking mesh.11 The polysaccharides glycosaminoglycans (GAGs) are usually attached to ECM proteins to form PGs. PGs fill the majority of the extracellular interstitial space within the tissue in the form of a hydrated gel.11 PGs have a net negative charge that attracts sodium ions and in consequence water, keeping the ECM and resident cells hydrated. Unlike other GAGs, the polysaccharide hyaluronic acid (or “hyaluronan”) contained in the ECM is not bound to matrix proteins. Hyaluronic acid in the extracellular space confers upon tissues the ability to resist compression providing a counteracting swelling force by absorbing significant amounts of water. Collagens constitute the main structural element of the ECM and, together with other fibrous proteins, provide mechanical strength and elasticity to the tissue.11 Nonactivated tissue fibroblasts secrete various ECM proteins (collagens and elastins) and PGs including hyaluronic acid.11 The ECM, containing various peptides, e.g., growth factors and enzymes, is being constantly remodeled for tissue homeostasis. Such remodeling is regulated by a careful balance between intracellular matrix synthesis, secretion, modification, e.g., crosslinking of collagens by lysyl oxidases (LOX) and enzymatic degradation, e.g., by matrix metalloproteases (MMPs).12 The ECM not only represents scaffolding for the stromal cells but also conveys biochemical and biophysical signals to cells.

Graphic Jump Location
Fig. 1
F1 :

Structure and function of ECM in (a) normal and (b) tumorous glandular epithelial tissue (e.g., breast). (Reprinted with permission from Frantz et al.11)

Abnormal vasculature and lymphatics of solid tumors

Solid tumors are not simply a collection of neoplastic cells but have been considered to be abnormal organs, with abnormal vasculature and lymphatics, abnormal ECM, and populations of stromal cells differing from normal host tissue.10,13 The vasculature of solid tumors differs distinctly from that of surrounding host tissue. Vasculature in normal tissue is arranged in a hierarchy of arteries–arterioles–capillaries–venules and veins, and grows under strict control of intervessel distances to ensure sufficient supply of oxygen and nutrients to cells by diffusion after extravasation, whereas tumor vasculature develops in a chaotic manner without such control, leading to chaotic interconnectivity of vessel segments and spatial vascular heterogeneity. In solid tumors, necrotic regions and regions of low microvessel density may occur, whereas tumor blood vessels are more abundant at the tumor–host interface.14

Tumors may co-opt existing host vasculature for supply of oxygen and nutrients,15,16 yet new vasculature must generally develop for tumors to grow beyond 1 to 2 mm3 in size.17 New blood vessels may be grown from existing vasculature by angiogenic sprouting, followed by growth of sprouts and their fusion with existing vessels to form new perfused blood vessels (neoangiogenesis).16,18,19 Vascular sprouting is initiated by vascular epithelial growth factor (VEGF), which is secreted by, e.g., hypoxic tumor cells. VEGF works in concert with other growth factors, e.g., angiopoietin-1 (ANG-1) and angiopoietin-2 (ANG-2). ANG-1 is involved in vessel maturation.19 On the other hand, ANG-2 disrupts the connections between the endothelium and perivascular cells and thus promotes cell death and vascular regression. Yet, in conjunction with VEGF, ANG-2 promotes neovascularization. By destabilizing existing vessels, ANG-2 allows for the formation of sprouts, thus contributing to neoangiogenesis, provided VEGF is present at sufficient concentration.16,19,20 In solid tumors, guidance proteins (EphB4) have been shown to act as negative regulators of blood vessel branching and vascular network formation, switching the vascularization program from sprouting angiogenesis to circumferential vessel growth.21 Neovascularization in solid tumors involves many complex processes of different origin and molecular pathways,22 whereas processes of angiogenesis are strictly controlled under physiological conditions; this is no longer true in solid tumors, and tumor vessels turn out to be immature, fragile, tortuous, dilated, with uneven diameters, and known to form (large diameter) arteriovenous shunts. Often, it is even difficult to distinguish arterioles from venules in solid tumors, i.e., the classification of tumor vessels as arterioles, capillaries, and venules is no longer adequate.13,14,18,23

Tumor cells that are considerably farther apart from nearby capillaries than the diffusion limit of oxygen (typically 100 to 200  μm) suffer from chronic (diffusion limited) hypoxia. Furthermore, the structurally abnormal tumor vasculature results in spatially and temporally heterogeneous blood flow, affecting tissue oxygenation (acute or perfusion-limited hypoxia).24 From intravital dorsal window microscopy on tumor models, it is known that blood flow through tumor capillaries is frequently sluggish and at times may even be stationary and reverse direction.23 It follows that blood flow through tumors may not follow a constant unidirectional path. In addition, red blood cell (RBC) flux varies greatly among tumor vessels; many tumor vessels do not carry RBCs but contain plasma only.24 Solid stress, caused by tumor cell proliferation and increased ECM deposition in tumors, may compress or block existing vasculature, impeding blood flow and, hence, the supply of oxygen and nutrients.9,2527

Tumor hypoxia is associated with poor prognosis, because it causes resistance to standard therapies and promotes more aggressive tumor phenotypes.28 Hypoxic tumor cells are known to be resistant to ionizing radiation, since oxygen is needed to stabilize radiation-induced DNA defects and, in addition, are considered to be resistant to some anticancer drugs.28,29 Furthermore, T-cells are dependent on normal oxygen levels for migration in tumor tissue, suggesting that hypoxia indirectly regulates antitumor immunity by restricting T-cell access.10

Tumor vessels may exhibit high permeability to macromolecules, e.g., to plasma proteins such as albumin.18,30 Generally, lymphatic drainage is impaired in tumors, since lymphatic vessels are sparse or even absent. Because of the leakiness of tumor vasculature, interstitial fluid pressure rises from, e.g., 0 mmHg in normal breast tissue up to 30 mmHg in breast cancers or even higher,13 impeding extravascular fluid flow and transport of extravasated macromolecules including chemotherapeutics, causing the so-called enhanced permeability and retention (EPR) effect.30,31 Macromolecules administered intravenously or small molecules, e.g., drugs that associate with plasma proteins, may not extravasate from normal vasculature, yet may leak into the interstitial matrix from tumor vasculature due to its enhanced permeability and stay there because of impaired lymphatic drainage.

Abnormal extracellular matrix of solid tumors

It has long been known that tumor-derived ECM is biochemically distinct in its composition compared with normal ECM (Fig. 1).12 Breast cancer progression is associated with changes in ECM composition, with inflammatory cell infiltration, and differentiation of fibroblasts.9 Tumor-derived ECM differs from that of the host tissue, owing to the disruption of the balance between ECM synthesis and secretion and owing to alterations in the normal levels of matrix-remodeling enzymes, such as MMP and LOX.12 The ECM remodeling observed in tumors includes increased deposition of collagen, fibronectin, PGs, substantial MMP-dependent cleavage, and increased levels of LOX-dependent matrix crosslinking. There is evidence for an increased deposition of ECM in hypoxic tumor regions.32 Differentiated fibroblasts (e.g., myofibroblasts) deposit large amounts of ECM proteins.11 The expression of MMPs is often highly upregulated in solid tumors with MMPs produced by myofibroblasts and tumor cells.11 ECM deposition and leukocyte infiltration are often very pronounced at the tumor-stroma border.10 The majority of increased tumor and adjacent tissue stiffness occur as a result of increased ECM deposition.12 Furthermore, increased LOX activity results in increased ECM stiffness.12 As illustrated in Fig. 2, the expanding tumor mass exerts compressive stress (tissue solid stress) on the surrounding tissue, on the ECM, on intratumoral vasculature and lymphatics.9,25 NIR imaging is directly affected by the modifications of the ECM occurring in tumors, resulting in changes of the tissue absorption spectrum. For example, additional deposition of collagen increases long-wavelength (1060 nm) absorption, while tissue solid stress on tumor vasculature may affect blood flow, resulting in changes of oxyhemoglobin concentration and corresponding modifications of the absorption spectrum.

Graphic Jump Location
Fig. 2
F2 :

Extrinsic and intrinsic forces on (a) normal and (b) tumorous glandular epithelial tissue (e.g., breast). (Reprinted with permission from Frantz et al.11)

Expected solid tumor-host optical contrast

Within the optical window, extending from about 635 up to 1060 nm breast tissue optical spectroscopy yields the absorption coefficient μa(λ) and reduced scattering coefficient μs(λ) of the host and tumor tissue. From average absorption coefficients μa(λ), concentrations of the main tissue constituents deoxyhemoglobin (HbR), oxyhemoglobin (HbO2), and thus, total hemoglobin HbT=HbR+HbO2, water, lipids, and collagen can be inferred using Beer’s law. Total tissue hemoglobin concentration, HbT, can be converted to vascular volume fraction rBV (vascular volume density) according to HbT=MCHC×H×rBV where MCHC is the mean corpuscular hemoglobin concentration and hematocrit H is assumed to be homogeneous within the vasculature. Because of angiogenesis and circumferential vessel growth occurring in tumors, one expects higher vascular volume fraction and thus higher total tissue hemoglobin concentrations in tumorous compared to host tissue HbT(T)>HbT(N) (see Sec. 4.1). Tissue blood oxygen saturation, defined as StO2=HbO2/HbT, depends not only on vascular structure but also on transvascular oxygen loss, i.e., on tissue metabolic rate of oxygen consumption and on vascular oxygen supply, i.e., on perfusion. From tumor biology, the concentration of collagen is expected to be higher in breast tumors compared to surrounding host tissue. Likewise, because of the additional deposition of the ECM in tumors and the hydrophilic nature of PGs contained, higher water content in breast tumors is predicted.

Apart from absorption, light scattering in tissue provides information on tissue structure and composition. Light scattering in tissue is dominated by Mie scattering and, therefore, probes density and size of biological cells; however, scattering by other tissue structures, such as collagen fibrils, cannot be excluded. Because of tumor cell proliferation and infiltration of various cells (e.g., inflammatory cells, differentiated fibroblasts) into the tumor stroma, cell density and, hence, the reduced scattering coefficient μs are expected to be increased in tumors compared to their surrounding host tissue. Furthermore, formation of large modified collagen bundles likely affects photon scattering (see Fig. 2). As discussed in Sec. 4.1, most of these predictions are borne out by experimental data.

Historical Development of Optical Breast Imaging

First attempts to see suspicious lesions in the female breast using visible light were reported by Cutler in 1929.33,34 The breast was transilluminated by holding a small size powerful lamp against the lower surface of the breast and by observing shadows of the light on the upper breast side by eye that arose from high-absorbing tissue structures. In the 1970s and 1980s, transillumination imaging of the breast was further developed by using improved light sources in combination with light detection by sensitive films or video cameras.3539 However, several clinical studies showed that sensitivity and specificity of the so-called lightscanning or “diaphanography” method were low in comparison to x-ray mammography.4043 The lightscanning approach was very simple to apply, but it had major intrinsic disadvantages, including the absence of discrimination between scattering and absorption of the tissue and the limited exploitation of spectral information.

New efforts in optical breast imaging started in the 1990s. At that time, mathematical models of light propagation in tissue became available that permitted separation of the scattering and the absorption properties of the tissue. In this way, absorption properties could be exploited by the methods of near-infrared spectroscopy to determine the composition of the tissue, i.e., to measure the concentrations of the main tissue absorbers including blood oxygen saturation. Basic concepts of optical breast imaging developed at this time are valid until today.

Classification of Optical Breast Imagers

Normally, optical images of the breast are obtained by switching through a sufficient number of point-like light sources realized by illuminated optical fibers and, for each source, detecting the light at one or several suited detector positions. The instruments for optical breast imaging and spectroscopy can be classified with respect to the temporal profile of the laser radiation employed and according to their measurement geometry. Furthermore, a few instruments have been designed to detect fluorescence arising from an exogenous contrast agent. The three well-known groups of instruments are time-domain, frequency-domain, and continuous-wave (CW) systems. Briefly, time-domain systems measure the broadening of short (picosecond) laser pulses after propagation through the tissue. Absorption coefficients μa and reduced scattering coefficients μs of the tissue can be derived from the analysis of the detected pulse shape using an appropriate model of photon propagation. These coefficients characterize the tissue volume sampled by photons depending on the particular source–detector combination. For source–detector separations of a few centimeters, the width of the broadened pulses amounts typically to a few nanoseconds. A well-suited technique for the detection of these pulses is time-correlated single-photon counting. In the past, hardware for time-domain systems was expensive. However, it is now becoming possible to perform time-resolved data acquisition with more cost-competitive instrumentation.

Frequency-domain systems use intensity-modulated laser radiation and measure its demodulation and phase shift after passing through tissue. In principle, it is sufficient to perform such measurements at one modulation frequency to separate the absorption and the scattering coefficient of the sampled tissue volume. Typically, a modulation frequency of about 100 MHz is chosen. Frequency-domain measurements correspond to investigating the Fourier spectrum of time-domain measurements at one frequency. By employing several modulation frequencies up to at least 1 GHz, the information content of frequency domain measurements increases and becomes comparable to that of typical time-domain investigations.

CW systems use continuously emitting lasers or light-emitting diodes at several near-infrared wavelengths, or broadband light sources. The instruments detect the attenuation of the transmitted light. More specifically, the CW technique yields the attenuation coefficient κ=μa/D, which is a combination of the absorption coefficient μa and the diffusion coefficient D=1/3μs. To obtain absolute values of the absorption coefficients, prior knowledge on the reduced scattering coefficient is required, e.g., from additional investigations with the frequency-domain or time-domain technique or by using representative values from literature. However, because light scattering dominates over absorption within the optical window, the latter approach may lead to systematic errors. Alternatively, it is possible to assess both absorption and scattering properties by performing CW measurements at more than one source–detector separation, which has been used, e.g., to obtain properties of small (almost homogeneous) tissue regions.44 Generally, it is also possible to exploit spectral features for the separation of absorption and scattering by using specifically selected optical wavelengths.45 The main advantage of the CW approach is that light sources and detectors are comparably cheap. Therefore, such systems can be equipped with a large number of source and detector positions at low costs.

The geometry for optical breast imaging can be divided into three main groups (Fig. 3). In one approach, the breast is compressed between two parallel plates. This geometry is very close to the concept of x-ray mammography. It offers a simple way to compare an optical transillumination image with a corresponding x-ray mammogram. A second approach consists of investigating the freely pending breast with the woman being in prone position. This geometry has high similarity with MRI of the breast. The third basic approach is the use of a handheld probe that can be positioned at selected locations of the breast or moved over its surface similar to a breast ultrasound detector. It is obvious that these three geometries offer the possibility for combining optical imaging with x-ray, MR, or US breast imaging, which was pursued by several groups, mainly to gain prior information on spatial tissue composition, lesion size, and position when analyzing their optical and functional properties.

Graphic Jump Location
Fig. 3
F3 :

Examples of measurement geometries: (a) parallel-plate geometry with moving source fiber and few detector fibers in transmission and reflection,46 (b) parallel-plate geometry with fixed source and detector fibers (left) and CCD camera detection (right) (reprinted with permission from Choe et al.47), (c) freely pending breast geometry with 255 source and 255 detector fibers (© 1999 IEEE. Reprinted with permission from Colak et al.48), (d) freely pending breast with 32 sources and 32 detectors (reprinted with permission of Optical Society of America from Enfield et al.49), (e) circular arrangement of 48 fibers in three rings with 16 translation stages to bring the fibers in contact with the tissue (reprinted with permission from Pogue et al.50), and (f) handheld probe (reprinted with permission from O’Sullivan et al.51).

Overview on Optical Breast Imagers

Figure 3 shows several examples of the measurement geometry and the source–detector arrangements that were realized in instrumentation for clinical studies on optical breast imaging.

A simple way to realize the compression geometry is the application of two transparent plates with variable distance [Fig. 3(a)]. Optical transillumination images of the breast can be obtained by moving a source fiber on one side and a detector fiber on the other side in tandem. Such systems have first been developed by the German companies Siemens and Carl Zeiss.52,53 Both systems were equipped with frequency-domain instrumentation. Later, the Physikalisch-Technische Bundesanstalt54 and the Politecnico di Milano55 used this geometry to build-up the first time-domain scanning optical mammographs. At Tufts University, the compression geometry has been employed more recently in a hybrid instrument combining frequency-domain measurements at a few near-infrared wavelengths with broadband CW spectroscopy56 and in an instrument using solely CW radiation.57 With a scanning step size of a few millimeters or even less, typically, more than 1000 source–detector combinations (scan positions) are sampled.

As an example for the parallel plate geometry, the schematic in Fig. 3(a) shows the latest source–detector fiber arrangement of the PTB instrument with several detection fibers in transmission and also detection fibers in reflection.46 The various fibers in transmission permit the detection of transillumination images at the implemented optical wavelengths under different projection angles, which can be exploited to reconstruct three-dimensional (3-D) images of the tissue.46,58 The additional detection fibers in reflection can be used to improve the 3-D resolution close to the surface, as shown in phantom experiments.59 However, with a transparent plate, only distances from the source fiber below, typically, 1 cm can be exploited due to multiple reflections occurring within the plate. Figure 3(a) shows a second source fiber (dashed line), that was added to the PTB optical mammograph for fluorescence measurements. The tandem scanner concept with several offset fibers in transmission is also employed in the latest version of the CW optical mammograph developed at Tufts University.57

Figure 3(b) illustrates the parallel-plate instrument developed at University of Pennsylvania. This hybrid CW-frequency domain device uses the compression geometry with a source–detector arrangement optimized for tomographic reconstruction. The patient is in the prone position with both breasts hanging in a tank filled with a scattering fluid. The opaque compression plate on the left-hand side is equipped with 45 fixed source fibers and 9 fixed detector fibers. The frequency-domain approach is employed to measure the diffusely reflected light at four wavelengths. The CCD camera on the right-hand side in Fig. 3(b) is used to measure the diffuse transmittance of the breast by the CW approach at up to six wavelengths.47 This device was extended with an option for fluorescence measurements using indocyanine green (ICG) as contrast agent.60 There were also two time-domain devices developed that use the parallel-plate geometry with matching fluid chamber, the commercial system Softscan of Advanced Research Technologies (ART) Inc., Montreal,61 and a laboratory prototype at the Physikalisch-Technische Bundesanstalt Berlin.62 Both these devices employed the concept of scanning source and detector fibers along transparent plates. Depth resolution was achieved by either offset detection channels in transmission similar to Fig. 3(a) or by two CCD cameras similar to Fig. 3(b). The PTB device was also capable of performing fluorescence investigations. The company DOBI Medical International Inc. developed the parallel-plate device ComfortScan for imaging pressure-induced changes in the blood oxygen saturation of tumors. This instrument uses flat-field illumination at 640 nm by LEDs and CCD camera detection of the transmitted light.63

Figures 3(c) and 3(d) show examples of devices with a cup-like chamber for investigations of the freely pending breast. The cup in Fig. 3(c) belongs to the CW instrument built by Philips, Eindhoven. In this device, a total of 255 source fibers and 255 interleaved detection fibers were used to reconstruct the tissue attenuation coefficient at three optical wavelengths.48 The same geometry was used in the fluorescence instrument developed about 10 years later.64 The time-domain instrument in Fig. 3(d) was built at the University College London. It is equipped with 32 source fibers and 32 detection fiber bundles and provides 3-D images of the absorption and scattering properties at two wavelengths.49 Both these devices with cup geometry use a scattering fluid to get high quality optical coupling and to work with a well-defined geometry for reconstruction of the optical properties. In order to account for breasts of different sizes, cups with different diameters can be used.

At Dartmouth College, a frequency-domain instrument with six optical wavelengths was developed for investigations on the freely pending breast that does not need a scattering fluid [see Fig. 3(e)]. Here, 48 fibers, which are arranged on three rings, are brought in contact with the tissue under slight pressure.50 In a newer version, the wavelength range of the device was extended up to 948 nm employing CW lasers.65 The breast imager developed by NIRx Technologies uses the same principle of pressure-induced optode contact. From these four wavelengths, the CW instrument permits simultaneous investigations on both breasts of the patient. It has been designed for dynamic investigations to record the physiological response of the breast tissue to specific interventions, such as the Valsalva maneuver or to dynamically observe the effect of a contrast agent bolus.66,67 The company, Imaging Diagnostics Systems, Inc., Fort Lauderdale, followed the principle of CT scanners and developed devices in which the freely pending breast of the patient in prone position is scanned by moving a laser beam and a detector array circularly around the tissue. By changing vertical positions, contiguous slices of the breast are acquired at one optical wavelength.68,69 Hereby, the detectors are not in contact with tissue. An extended version of these devices was prepared for fluorescence investigations.

Detection limits of diffuse optical tomography systems were investigated theoretically and numerically.70,71 The method consists of analyzing raw numerical phantom data by means of a chi-square test, obtained from forward simulations together with a realistic noise model, derived from the system hardware. Both parallel-plate and cup geometries were compared with respect to detection limits of heterogeneities at various positions within the tomographic volume investigated. In cup geometry, low detection sensitivity was obtained at the upper center of the cup [cf., Figs. 3(c) and 3(d)], i.e., close to the chest wall, where the tissue is exclusively sampled by source–detector combinations with large separations. In slab geometry, detection sensitivity shows only small variations between the outer and the inner tissue regions, since the breast is sampled with constant source–detector separation. Generally, for smaller breast sizes, lesions of 5-mm diameter could be detected in almost all parts of the compressed breast (parallel-plate geometry) and in the outer parts of the uncompressed breast (cup geometry), whereas for larger breasts, the detection limit moved toward 7.5-mm lesion size when a lesion-to-background absorption contrast of 21 was assumed.

The third group of optical breast imagers comprises devices with handheld probes. As an example, Fig. 3(f) shows the probe of the diffuse optical spectroscopic imaging (DOSI) device developed at the University of California, Irvine.51,72 This device combines the frequency-domain and the CW approach. It contains two source–detector pairs. The first one is used to perform frequency-domain measurements at six optical wavelengths ranging from 650 to 860 nm. In contrast to the frequency-domain instruments discussed above, the modulation frequency is varied here from 50 to 1000 MHz. Therefore, the amount of information obtained for the sampled tissue volume is comparable to that of time-domain approaches discussed above. The second source–detector pair is connected to a tungsten halogen white-light source and a spectrometer to record broadband reflectance spectra from 650 to 1000 nm. The scatter power law: Display Formula

μs(λ)=μs(λ0)(λ/λ0)b,(1)
which is fitted to frequency-domain data at the six discrete wavelengths, provides a scatter correction for the CW reflectance spectra. In a second step, the absorption spectra are extracted by best fitting the corrected CW reflectance spectra to the photon diffusion model and used to fit the chromophore concentrations. To obtain spatially resolved information, the probe is positioned, e.g., at steps of 10 mm, along a line or a two-dimensional (2-D) grid on the breast at the (known) lesion position with the patient in supine position.73

At the University of Pennsylvania, a handheld CW imager was used for breast imaging with a three wavelengths LED source and eight surrounding silicon diode detectors 4-cm apart from the source.74 Another CW device was developed by ViOptix, Inc., Fremont.75 Handheld probes have also been used for blood flow characterization in breast tissue and tumors employing diffuse correlation spectroscopy (DCS; cf. Sec. 3.5). A detailed overview about the various handheld NIR devices for breast imaging and other applications can be found in the review by Erickson and Godavarty.76

Multimodality Imaging

Several groups have combined optical breast imaging with other clinical breast imaging modalities. Hereby, the conventional clinical modality provides structural information about the breast tissue that is exploited in the reconstruction of the optical and physiological properties. In this way, problems of low spatial resolution and diffuse blurring of reconstructed optical data can be overcome, and optics can provide, e.g., metabolic information about lesions not accessible by the conventional modalities. At the University of Connecticut, a handheld probe device has been developed for combined investigations of breast tumors by NIR light and ultrasound. This probe contains a commercial US detector array together with 12 source and 8 detection fibers for diffuse reflectance frequency-domain measurements at several source–detector distances.77 In the present version, three optical wavelengths between 660 and 830 nm are used.78

The combination of optical and ultrasound measurements in a handheld probe is technically simple, whereas the combination of optical and MR measurements is more challenging. To avoid interference of the optoelectronic components with the high magnetic field, long fibers or fiber bundles have to be used to deliver and collect the light inside the MR bore. Furthermore, the restricted space inside the bore limits the number of source and detector fibers that can be installed. The first demonstration of concurrent optical and MR imaging of breast tissue was reported by Ntziachristos and Ma,79 who placed a parallel-plate patient interface for time-resolved transmittance measurements inside the MR tomograph. This instrument allowed comparison of contrast-enhanced optical absorption imaging with ICG versus dynamic contrast enhanced MR imaging80 and provided intrinsic hemoglobin and oxygen saturation contrast StO2 for malignant and benign breast lesions.81 At Dartmouth College, a 16 fiber ring holder interface was applied inside the MR scanner for simultaneous frequency-domain optical and MR interventions.82 During reconstruction of the optical images, the structural information from the MR investigations was used as prior information. This group also reported successfully incorporating water and fat information from MR imaging to improve the accuracy of the reconstructed hemoglobin concentration.83

At the Massachusetts General Hospital, a combined optical and x-ray breast imager was developed that uses the parallel-plate geometry. Optical measurements are performed in transmittance by a hybrid frequency-domain and CW approach using a source–detector grid designed for 3-D reconstruction of the optical properties. The device can also be applied for functional monitoring of the breast tissue.84

Data Analysis and Reconstruction

One aim of data analysis is the generation of optical mammograms, i.e., the generation of 2-D or 3-D images that display lesions and structures inside the breast with high contrast. Another aim is the determination of the optical and physiological parameters of lesions and of healthy breast tissue. Hereby, the generation of images is not really required. Generally, optical mammograms can provide such values. However, the accuracy of values characterizing lesions is often limited by the diffuse blurring and by partial volume effects. These limitations can be overcome by using prior information about the size and location of the lesions together with adequate heterogeneous models.

The generation of optical mammograms for the circular tomographic geometry requires a reconstruction of the optical or physiological properties of the breast. For the parallel-plate devices, reconstruction can also be applied, but it is not mandatory. Since the distance between source fiber and the detection fiber in scanning devices with parallel-plate geometry is the same at all scan positions, data at each scan position can be analyzed independently.

Figure 4 illustrates some general features of optical mammograms obtained by the various data analysis methods. Transillumination images from parallel-plate instruments [Figs. 4(a)4(c)] often exhibit not only the carcinoma but also superficial blood vessels and other localized regions of high vascularization. Reconstructed slices from parallel-plate instruments show more blurred structures [Figs. 4(d) and 4(e)]. Similarly, reconstructed images from circular tomographic measurements [Figs. 4(f) and 4(g)] display the carcinoma often within a nonuniform background with additional structures showing high correlation to the source and detector fiber positions.

Graphic Jump Location
Fig. 4
F4 :

Examples of absorption and HbT optical mammograms of patients with carcinomas (labeled by arrows or lines): (a) late-gate transillumination image at 670 nm from a parallel-plate instrument displaying a 3.5-cm carcinoma (from Rinneberg et al.85); (b) transillumination image at 785 nm from a parallel-plate instrument displaying difference in absorption coefficient Δμa from average background analyzed within an inhomogeneous model (from Quarto et al.86); (c) HbT transillumination image with 3.9-cm carcinoma from a CW parallel-plate instrument (reprinted with permission from Anderson et al.87), (d) reconstructed slice of relative total hemoglobin concentration (designated rTHC) with a 2.2-cm carcinoma from a parallel-plate instrument with matching fluid (reprinted with permission from Choe et al.47), (e) reconstructed slice of HbT with a 2.5-cm carcinoma obtained with prior knowledge from x-ray mammography (reprinted with permission of the Radiological Society of North America from Fang et al.88), (f) reconstructed slice (HbT map) with a carcinoma about 4 cm in size (reprinted with permission from Wang et al.65), and (g) reconstructed absorption image with carcinoma (indicated by yellow circle) from a CW tomographic instrument with matching fluid (image reproduced from van de Ven et al.89).

Homogeneous models

Together with time-domain parallel-plate instruments, so-called late-gate and early-gate intensity images are used to generate optical mammograms showing absorbing lesions like carcinomas and lesions with reduced light scattering like cysts with high contrast. Hereby, the different effects of localized scattering and absorbing objects on time-resolved transmittance measurements are exploited.90 The late-gate intensity is mostly sensitive to absorption properties, and consequently, a late-gate image at a certain wavelength displays the spatial distribution of the major absorber at that wavelength. Thus, late-gate intensity mammograms at wavelengths around 800 nm are directly correlated to the distribution of (total) hemoglobin in the tissue, whereas late-gate mammograms at wavelength between 650 and 690 nm are more sensitive to the distribution of deoxyhemoglobin in tissue [Fig. 4(a)]. Correspondingly, mammograms taken at about 925 and 975 nm are correlated to the local distribution of lipids and water in the tissue, respectively. In frequency-domain measurements, the absorption properties of tissue are typically displayed by optical mammograms showing demodulation, whereas phase information yields mammograms related to scattering properties. The visual contrast of absorbing objects in late-gate or demodulation images can be enlarged by plotting reciprocal values of these quantities or using a second-derivative method.53,54,91

However, simple intensity mammograms at different wavelengths give only qualitative information about optical and physiological properties of lesions. In order to determine these properties quantitatively, time-domain or frequency-domain data have been analyzed by different models of light propagation. A simple approach to determine the tissue optical properties at each scan position is the application of the diffusion model for the homogeneous infinite slab. In this way, optical mammograms showing absorption coefficients and reduced scattering coefficients can be generated. From corresponding data at different wavelengths, maps of hemoglobin concentration HbT and tissue blood oxygen saturation StO2, water, lipid, and collagen content can be derived relying on Beer’s law. Information on number density and equivalent size of the scattering centers (typically cell organelles, membranes, and so on) can be obtained from the dependence of the reduced scattering coefficients on wavelength together with appropriate Mie calculations.92,93 To improve the robustness and stability of the fitting procedure that aims at estimating a rather high number of unknowns from data collected at few wavelengths, a spectrally constrained global fitting procedure has also been effectively applied.94 Specifically, the concentrations of oxy- and deoxyhemoglobin, water, lipids, and collagen, together with the reduced scattering coefficient at 600 nm and scattering power b, were fitted directly to time-resolved transmittance curves measured at 7 wavelengths, using the Beer law to relate the absorption properties to the concentrations of the main tissue constituents and the approximation to Mie theory to model the scattering properties.

When a homogeneous model is applied, the contrast between lesions and the surrounding tissue is underestimated since the lesion typically fills only part of the banana-like volume between source and detector (partial volume effect). In order to enhance the contrast in these maps, the application of a second-derivative approach was proposed.95

The discussed models for the parallel-plate geometry need to be improved at the edges of the breast since the breast of decreasing thickness no longer fills the space between both glass plates. To avoid artifacts, distributions of times of flight can be scaled using the mean times of flight of the detected photons as a rough measure of the tissue thickness at the various scan positions.96 This correction works also for fluorescence measurements.46 In frequency domain, demodulation data can be corrected for edge effects by exploiting the measured phase information.97

Measurements in reflection geometry with the handheld probe, as shown in Fig. 3(f), have been analyzed by using the diffusion model for the homogeneous semi-infinite medium.98 Also, in this analysis, the contrast of the lesion to the surrounding tissue is underestimated due to partial volume effects. By plotting the results obtained at different positions of the probe, a mammogram with a small number of pixels is obtained.73

Heterogeneous models

More realistic values of lesion optical properties and functional parameters from parallel-plate devices have been obtained by the application of inhomogeneous models considering the lesion as an object in an otherwise homogeneous background medium. One approach is the model of diffraction of photon density waves by a spherical object.99 Furthermore, a random walk model has been used to derive tumor optical properties.100 Other approaches are perturbation models like the first-order Born approximation or an empirical Padé approach.101,102 Since the true shape of the lesion is typically not known, the spherical shape has been assumed in these models. The latter two models have also been employed to generate optical mammograms by assuming a virtual sphere of predefined size located in the midplane of the breast. In this way, the partial volume effect of the homogeneous model discussed above is reduced and the lesion contrast in the optical mammogram is improved. Generally, low order perturbation models are at their limit of validity in a large number of cases, since lesion size and absorption contrast can become large.103 Recently, a higher order perturbation model has been employed for the analysis of the optical and functional parameters of malignant and benign lesions [see Fig. 4(b)].86

The results of the inhomogeneous models strongly depend on the assumptions about the size and the location of the lesion (distance to the parallel plates). The latter information can be easily derived from measurements with offset fibers [cf., Fig. 3(a)].99 However, the size cannot be reliably determined from the optical measurements due to the diffusive nature of the light propagation. Therefore, often the size information from conventional clinical imaging modalities or from pathological findings has been used as prior information. Hereby, one should have in mind that the extension of the vascular bed of a tumor, which is responsible for the optical contrast, could deviate from the clinical size estimation.

Reconstruction of optical properties and functional parameters

Optical mammograms based on the circular tomographic geometry are obtained by 2-D or 3-D reconstruction of optical properties and functional parameters. Typically, the diffusion model is used as forward model. Arbitrary geometries of the breast can be handled by the finite-element method. Time-domain data from the freely pending breast covered by a matching fluid have been analyzed by applying the TOAST software package.49 At Dartmouth College, a frequency-domain implementation of the finite element method was employed.104 Later, this model was developed toward a spectrally constrained approach that directly fits the functional parameters, hereby exploiting the scatter power law [Fig. 4(f)].105 Philips used a back projection algorithm to reconstruct the attenuation coefficients for their CW instrument.48 In their fluorescence, mammograph reconstruction of attenuation data was performed by a linear Rytov approximation assuming constant scattering [Fig. 4(g)]. For reconstruction of the fluorescence data, a Born approximation was used.64 Data from the dynamic breast imager of NIRx are analyzed by a linear perturbation method, too.66