1.

Introduction

Imaging plays a central role in cancer diagnosis, therapy, and prognosis primarily through the detection of anatomically defined abnormalities, including calcifications that are often present in later-stage breast cancer. With the wealth of information provided by the now-maturing areas of genomics and proteomics, the identification of molecular markers and targets promises contrast-enhanced, diagnostic cancer imaging with a sensitivity and specificity to early cancer that is not possible with conventional, anatomical imaging. Molecular approaches promise to improve diagnostic cancer imaging and ultimately to affect the quality of cancer patient care.

While molecular imaging with exogenous agents may be universally attractive for most conventional imaging modalities such as magnetic resonance imaging (MRI), nuclear imaging [such as gamma imaging, positron emission tomography (PET), and single-photon emission computed tomography (SPECT)], and x-ray computed tomography (CT), molecular imaging requires sufficient signal arising from minute quantities of targeting agent. Magnetic resonance (MR) and x-ray imaging techniques typically require millimolar amounts of an exogeneous contrast agent, but recent advances in MRI have demonstrated that potential contrast can exist at concentrations as low as micromolar amounts. However, the amounts of exogeneous contrast agent required in MR and x-ray imaging are still substantial for sufficient signals for human imaging. Hence, the challenge of molecular diagnostic imaging of metastatic cancer typically lies with nuclear and optical techniques, which enable imaging using exogenous contrast agents at levels that are orders of magnitude smaller (<nanomolar).

Over the past 15 years, near-infrared (NIR) optical imaging approaches have been developed for breast cancer screening that are based upon an endogenous absorption contrast resulting from the nonspecific process of angiogenesis in order to discriminate normal from diseased tissues.^{1 2 3 4 5 6 7} However, for diagnostic detection of early metastatic lesions in which the angiogenesis-induced contrast is nonexistent or insufficient for tomographic NIR imaging of the breast, exogenous contrast based upon targeting and reporting fluorophores will be necessary.^{8 9 10 11} In the Photon Migration Laboratory at Texas A&M University, we have sought to develop fluorescence-enhanced optical tomography for breast cancer diagnosis via molecular imaging techniques. The approach depends upon delivering molecularly targeting NIR-excitable fluorescent contrast agents, which specifically target metastatic cancer cells within the breast tissue or, for prognostic indication of the extent of disease, within the lymph system draining the breast. Detection of cancer cells within the subsurface axillary lymph nodes, the internal mammary lymph nodes located beneath the medial sternum, or within the sentinel lymph node (or the lymph node closest to the primary breast tumor), which can be as great as 4 cm deep, is typically performed via biopsy following surgical resection. In the case of sentinel lymph nodes, it is performed following anatomical mapping using nuclear imaging and detection techniques, and prior to surgical resection and pathology.

However, before clinical feasibility trials involving the administration of a fluorescence contrast agent for diagnostic and prognostic imaging can begin, the efficacy of tomographic fluorescence-enhanced optical imaging in clinically relevant volumes must be demonstrated in phantom studies. To date, three-dimensional fluorescence-enhanced optical tomography has been limited to in vivo measurements in small animal models^{12 13} or phantoms with unrealistic volumes or geometries^{14 15 16} in comparison to what would be expected clinically.

In this paper, we present experimental tomographic imaging results involving two distinct measurement geometries pertinent to diagnostic and prognostic imaging of the breast within clinically relevant volumes and measured using an intensified charge-coupled device camera (ICCD) modified for time-dependent measurements of light propagation and fluorescence generation. In Sec. 2 we outline the necessity for the differing measurement geometries with point illumination and collection and area illumination and collection, and describe the light propagation model for fluorescence-enhanced optical imaging. The clinically relevant phantoms employed, the unique instrumentation involved in both measurement geometries, and a brief description of tomographic algorithms are given in Sec. 3. The experimentally obtained images that demonstrate the feasibility for state-of-the-art 3-D fluorescence-enhanced optical tomographic reconstructions on clinically relevant tissue phantoms are presented for the point and area illumination and collection geometries. Finally, these 3-D image reconstructions are discussed in light of the current ongoing clinical feasibility studies of fluorescence-enhanced optical imaging for mapping sentinel lymph nodes in breast cancer patients.

2.

Background

The time-dependent technique for measuring light propagation and fluorescence generation used here involves the frequency-domain approach. In this approach, intensity-modulated, incident excitation light is delivered at a single point, at a series of points, or across an area on the tissue boundary, typically at modulation frequencies of 30 to 200 MHz. The excitation light or “photon density wave” propagates as a spherical wave (in the case of point illumination) or a planar wave (in the case of area illumination). Owing to the absorption and scattering properties of the tissue, the excitation wave is amplitude attenuated and phase shifted relative to the incident wave. Upon encountering fluorescently tagged volumes within the tissue, a fluorescent photon density wave is generated with additional amplitude attenuation and phase lag, owing to the fluorescence decay kinetics of the fluorophore. This fluorescent photon density wave then propagates throughout the tissue, and during transit it experiences additional amplitude attenuation and phase lag owing to the tissue’s optical properties until it reaches the tissue boundaries where it can be collected separately from the propagated excitation wave by optical filters.

Typically, all tomographic NIR imaging approaches (whether based upon endogenous or exogenous contrast) involve inverting measurements of light propagation between source–detector pairs of single, sequential points of excitation illumination and excitation (or emission) collection. Yet as shown in Fig. 1, illumination by excitation light from a single point interrogates a relatively small portion of tissue volume. Consequently, measurements among a series of single pairs of illumination and collection points must be accomplished using either spatial scanning or replication of the transmitters and receivers across the surface of the tissue. While these measurements are sufficient for the inverse spectroscopy problem, a limited set of point illumination and collection sites may produce sparse datasets for 3-D image reconstructions, especially when the fluorophore accumulates in an area not interrogated by the excitation light (Fig. 1). As a result, a high density of measurements in a relatively confined volume is typically required for imaging purposes, and simultaneous point illumination and area illumination and collection schemes in reflectance and\or transillumination geometries may be more appropriate.

Figure 1

Different measurement geometries. (a) Sequential point illumination of single excitation source using optical fibers. (b) Simultaneous point illumination of two excitation sources using optical fibers. (c) Simultaneous point illumination of multiple excitation sources using optical fibers. (d) Area illumination using an expanded beam of the excitation source. (c) and (d) are equivalent below the tissue surface, if the density of multiple-point illuminating sources is high.

Yet to date, while tomographic imaging algorithms have been developed for single illumination and collection measurements, to our knowledge there have been no attempts to tomographically reconstruct images from NIR measurements involving simultaneous point illumination and collection geometries or involving area illumination and collection geometries. In this work, we have extended tomographic imaging to include dual points of simultaneous illumination, as well as area illumination in order to enable interrogation of a larger tissue volume, as is appropriate for clinically relevant tissue volumes. Both simultaneous multiple-point sources and area illumination are being explored as means of (1) increasing the spatial coverage and the amount of NIR source light that can be safely delivered, and (2) decreasing the data acquisition time. Conceptually, as the number of point sources increases, the multiple-point illumination geometry [Fig. 1(c)] will mimic the area illumination geometry [Fig. 1(d)]. However in practice, the instruments for these two illumination geometries offer different opportunities and challenges since in one case optical fibers must contact the surface of the domain, and in the other case no optical fibers are involved. The general mathematical formulation of the light propagation model for both geometries is briefly presented in Sec. 2.1.

3.

Materials and Methods

3.1.

Phantoms

Parallel studies were performed using two different phantoms with differing measurement geometries. These include a breast-shaped phantom employing point illumination and collection geometry, and a semi-infinite phantom employing area illumination and collection geometry.

3.1.1.

Breast-shaped phantom using point illumination and collection geometry

For tomographic evaluation of metastatic spread of cancer in the breast, we employed a breast-shaped phantom as shown in Fig. 2(a). The hemispherical portion of the phantom (10 cm diameter) was used to mimic the breast tissue and the extended chest wall regions around the breast (cylindrical portion, 20 cm diameter and 2.5 cm height). The measurement geometry for point illumination and collection was adapted to the breast phantom using optical fibers of 1-mm diameter that were located in concentric rings along the hemispherical surface of the phantom.²⁴ Most of the work performed in optical tomography employs a sequential point illumination and collection geometry, with recent developments toward imaging simultaneous multiple collection fibers using CCD cameras in continuous-wave (cw)^{12 13} and frequency-domain photon migration (FDPM) approaches.^{24 25 26} In our work, the collection of a light signal from the phantom surface was expedited by employing a CCD camera for simultaneous acquisition of data from multiple collection fibers that were interfaced as 2-D arrays onto an interfacing plate. Apart from a single-point illumination technique, a simultaneous dual in-phase point illumination technique [see Fig. 1(b)] was employed as a first step in evaluating the feasibility of tomographic reconstructions from measurements acquired using simultaneous multiple-point illumination geometry [see Fig. 1(c)].²⁵ Fluorescence measurements were acquired from the entire hemispherical surface of the phantom; that is, 3-D boundary surface measurements were acquired for 3-D tomographic reconstructions.

Figure 2

Breast-shaped phantom with the cup-shaped portion (10-cm diameter) representing the breast tissue and the cylindrical portion (20-cm diameter and 2.5-cm height) representing the extended chest wall regions around the breast. Point illumination and collection measurement was employed on the hemispherical surface using optical fibers. (b) Semi-infinite phantom of 8×8×8 cm ³ volume. Area illumination and detection was employed by illuminating and collecting a signal from a given area on the top surface of the phantom.

3.1.2.

Semi-infinite phantom using area illumination and collection geometry

For tomographic evaluation of the depth of a sentinel lymph node and for mapping lymph flow, current clinical feasibility trials involve area illumination of the upper quadrant of the breast, which is closest to the axilla, and area collection of the reemitted fluorescence signal. The phantom analogy to these clinical trials involves a semi-infinite model (8×8×8 cm ³ volume) [see Fig. 2(b)] employing area illumination and collection geometry. Here, the measurement geometry does not require any optical fibers to illuminate or collect the light signal. An expanded beam of modulated NIR light was incident on the top surface of the semi-infinite phantom and the emitted fluorescent signal was imaged directly using an intensified CCD camera detection system. Unlike point illumination of the excitation source, where the point source is not characterized, area illumination of the excitation source requires spatial characterization of the incident modulated NIR light. Hence, cross-polarizers are used to directly measure the excitation light specularly reflected (in terms of amplitude and phase) from the phantom’s surface.^{27 28}

As shown later, 3-D reconstructions were feasible with the semi-infinite phantom, even with 2-D boundary surface imaging from a single planar wave source. This measurement geometry mimics the prognostic imaging of a sentinel lymph node (from a 2-D boundary surface) in an actual clinical application.

3.3.

Instrumentation

Time-dependent fluorescent measurements were acquired in the frequency domain using a gain-modulated intensified CCD imaging system. A schematic of the instrumentation, along with the different measurement schemes employed in two different phantom experiments, is provided in Fig. 3. Modulated NIR light (783- or 785-nm wavelength) at 100 MHz was launched onto the phantom surface either through optical fibers or an expanded beam. In the breast phantom, dual excitation sources for a simultaneous dual-point illumination geometry were also obtained by splitting the laser source using a 50:50 beamsplitter (20Q20BS.2, Newport Corp., California). The modulating fluorescence signal emitted from both phantoms was acquired via a Nikon lens by the photocathode of the image intensifier (FS9910C Gen III model, IIT Night Vision, Roanoke, Virginia), which was also modulated at 100 MHz.

Figure 3

Instrumentation for fluorescence-enhanced optical imaging in the frequency domain, using the gain-modulated intensified CCD camera for detection. The instrumentation setup is similar for both phantom systems, except for the measurement geometry employed in either phantom.

For measurement on the breast phantom using single- and dual-point illumination and multiple-point collection, the fluorescent signal collected by the fibers, whose ends were coupled to an interfacing plate, was acquired by the photocathode of the image intensifier. In contrast, fluorescence emitted from the boundary surface of the semi-infinite phantom was directly imaged onto the photocathode of the image intensifier. The fluorescent photons are converted to electrons at the photocathode, amplified at the multichannel plate, and converted back to photons at the phosphor screen of the image intensifier. These photons are in turn imaged by the CCD camera, which was either lens coupled (in the semi-infinite phantom system) or optically coupled (in the breast phantom system) to the phosphor screen of the image intensifier. A 16-bit CCD camera (Series AT200, model S1512B, Photometrics Ltd., Tucson, Arizona) was used for the semi-infinite phantom studies and a 12-bit CCD camera (TE/CCD-512-EFT Photometric CH12, Roper Scientific, New Jersey) was used for the breast phantom studies. The use of an ICCD detection system enhanced the data acquisition rates for both the phantoms, relative to the use of single-photon multiplier tubes (PMTs) for sequential detection, as employed in our earlier imaging studies.^{14 15 16}

Measurements on both the phantoms were performed using the homodyne technique, since the laser diode and the photocathode of the image intensifier were modulated at the same frequency (100 MHz) using two oscillators that were phase locked by a 10-MHz reference signal. A phase delay varying from 0 to 2π was introduced between the two oscillators in order to obtain phase-sensitive constant-intensity images.³² The ac intensity and phase shift were determined from the fast Fourier transform (FFT) that was performed on the 32 steady-state images acquired by varying the phase delay between the two oscillators. The data acquisition time varied with the integration time of the CCD camera (0.2 to 1 s), the number of phase delays acquired (32 in our case), and the total number of repetitions of each steady-state image (5 to 10). Further details regarding instrumentation and data acquisition procedures for both phantoms are described elsewhere.^{24 27 32 33}

The weak fluorescent signal was collected by blocking the excitation light at the phantom surface using a stack of optical filters, including an 830-nm interference filter (F10-830.0-4-2.0 model, CVI Laser Corp., Albuquerque, New Mexico) and a holographic filter (HNPF-785.0-2.0 model, Kaiser Optical Systems Inc., Ann Arbor, Michigan). However, the filters were not very efficient in completely rejecting the 785-nm excitation light, limiting the detection of deeply embedded targets in large phantom volumes regardless of the measurement geometries.^{11 25} Work is currently underway to customize optical filters that can efficiently reject the excitation light without further attenuating the weak reemitted fluorescent signal.

For all experimental conditions, the acquired fluorescence measurements in terms of ac intensity (I_ac) and phase shift (θ) were referenced with respect to the measurement location having the maximum fluorescence ac intensity, in order to account for instrument response functions that affect the transmission of fluorescent signal from the phantom surface to the CCD detector array.^{24 28}

4.

Results

The fluorescence measurements acquired from all experimental studies matched the simulated measurements predicted by the forward model reasonably well. Three-dimensional image reconstructions were made using the respective inversion approaches for the two phantoms; that is, a Bayesian approach for breast phantom studies and a modified truncated Newton’s approach for the semi-infinite phantom studies.

Reconstructions obtained from the breast phantom experiments containing a single 1-cm ³ volume target located 1.4 cm deep with single- and dual-point illumination geometries are presented as a series of contour slices in the y -plane in Fig. 4.²⁵ Similar contour-sliced plots of the reconstructed semi-infinite phantom are presented in which a single 1-cm ³ volume target was located 1.4 cm deep from the boundary surface (see Fig. 5).⁴¹ In both the phantoms and for all three differentexperimental cases, the single target was successfully reconstructed close to its true location without any spatial artifacts. Single, 1-cm ³ volume targets located 2 cm deep were also reconstructed in the breast phantom.²⁵ For brevity and for comparison with the reconstructions from the area illumination and collection measurements, only the results from the 1-cm deep target are presented here.

Figure 4

(a) A series of contour slices along the y -plane of an actual breast phantom, representing the true location of the 1.4-cm ³ target located 1 cm from the hemispherical phantom surface. (b) A series of contour slices along the y -plane of the reconstructed breast phantom using single-point illumination. (c) Contour slices along the y -plane of the reconstructed breast phantom using simultaneous dual-point illumination.

Figure 5

(a) A series of contour slices along the y -plane of an actual cubic phantom, representing the true location of the 1-cm ³ target 1 cm from the top surface. (b) A series of contour slices along the y -plane of the reconstructed phantom, representing the reconstructed location of the target.

Image reconstructions were also performed for the breast phantom experiment containing three small-volume targets (0.5 to 0.5 cm ³) located up to 1.4 cm deep from the phantom surface (experiment 4 of Table 1). The initial reconstructions used a discretized mesh phantom (3857 unknowns) similar to that used for the single 1-cm ³ volume target studies. Using this coarsely discretized mesh phantom, one of the three targets was not completely located and appeared as an artifact [see Figs. 6(b) and 6(e)], even though the reconstructions converged. Upon employing a more finely discretized breast phantom (11,906 unknowns) under similar reconstruction conditions, all three targets were distinctly located without any artifacts [see Figs. 6(c) and 6(f)].²⁶ This result could probably be attributed to the level of FE mesh discretization, in which the distance between two consecutive nodes in the mesh, if equivalent to the radius of the target, could reduce the probability of reconstructing the target. It should be noted that these breast phantom studies, which attempt to mimic the multifocality of lesions within the breast, are preliminary and more work is under way to assess the significance of discretization for detecting the smallest possible target volumes.

Figure 6

(a) y-x and (d) y-z planar views of the actual breast phantom. (b) and (e) Reconstructed breast phantom using coarsely discretized mesh. (c) and (f) Reconstructed breast phantom using a finely discretized mesh. Three targets of 0.5 to 0.6 cm ³ volume were used in the actual breast phantom of ∼1087 cm ³ volume and measured using single-point illumination geometry.

5.

Discussion

Future diagnostic and prognostic breast imaging with NIR must be directed toward molecular targeting of single and multiple lesions. To realize these goals, simultaneous development of (1) fluorescence-enhanced optical imaging systems using clinically relevant phantom systems and different possible measurement geometries, and (2) reconstruction algorithms to solve for the highly ill-posed inversion problem, where sparse surface measurements are used to determine the 3-D interior optical property maps, must be undertaken. Only a few examples of such reconstructions are presented here, in order to demonstrate for the first time the feasibility of tomographically reconstructing large 3-D phantoms of clinical relevance, and using different measurement geometries. Work is progressing toward assessment of their robustness using different measurement geometries and under varying experimental conditions, in order to aid in the clinical translation of the emerging fluorescence-enhanced optical imaging technology.

Breast phantom studies using point illumination and collection geometry have the potential for a direct translation into clinical diagnostic imaging of breast tissue by molecularly targeting the metastatic spread of lesions or early-stage lesions using a systemic, intravenously injected fluorescent contrast agent. The feasibility of employing dual-point illumination measurement geometry in breast phantoms can be easily translated into simultaneous multiple-point illumination geometry for clinical feasibility trials, prior to which the sensitivity and specificity of this imaging modality must be assessed using different phantom models.

The preliminary semi-infinite phantom studies using area illumination and collection geometry have demonstrated the possibility of prognostic imaging of sentinel lymph nodes and of locating nodes for surgical resection. Although initial clinical feasibility trials are using a nonspecific contrast agent (ICG), parallel feasibility studies are currently being performed to develop and assess the performance of tumor-specific contrast agents through small-animal studies.⁴²

Future development critically depends upon improving rejection of excitation light, enabling tomographic reconstructions of smaller targets at greater depths and with lower concentrations (up to nanomolar amounts) of the fluorophores. The ability to collect a signal from 100 fmol of ICG at tissue depths of 4 cm and greater¹¹ needs to be effectively transferred to clinical breast imaging.