1.

Introduction

In the last decades numerous methods for screening breast cancer have been applied, including x-ray mammography (for an overview see Ref. 1), magnetic resonance imaging,^{2 3} positron emission tomography,⁴ and ultrasound.⁵ During the last few years a great deal of effort was expended to develop instrumental systems for transilluminating the breast and the brain with near-infrared (NIR) light,^{6 7 8 9 10 11} taking advantage of the fact that NIR light is nonionizing and harmless at doses sufficient for medical diagnosis. A disadvantage of NIR transillumination imaging is the low spatial resolution that results from multiple scattering of photons when thick tissues are investigated.^{12 13 14} The contrast in optical mammograms is based on different scattering and absorption properties of tumor and normal tissue. The contrast in optical transillumination images^{8 15} based on scattering alone is not sufficient to discriminate the different types of tissues. On the other hand, increased vascularization and blood volume of tumors allow us to detect the lesions at adequate contrast based on absorption differences.^{8 16 17 18} However, the tumors detected in these in vivo studies are rather large and exhibit a high degree of angiogenesis, therefore hemoglobin contrasts. In images relying on differences in absorption small tumors might be overlooked because of a low degree of angiogenesis and of superimposed areas due to large blood vessels, if these areas interfere geometrically with the target tissue. In this case a different mechanism to create contrast is needed.

Troy et al.;¹⁹ reported on reduced scattering and absorption coefficients, determined ex vivo by about 115 histologically classified specimens from 88 patients. The absorption coefficient was, however, underestimated due to blood drainage during excision. These measured absorption coefficients, therefore, do not reflect malignant in vivo conditions. Only for five patients was the reduced scattering coefficient determined for tumor and corresponding normal tissue of the same patient. However, direct comparison of the two types of tissue for the same individual is crucial because of the large biological variability of normal breast tissue.

As a basis for the development of a contrast agent-based transillumination, for about 20 patients it was determined if there was a difference in the reduced scattering coefficients between normal and tumor tissue of the same patient. Since the alteration of tissue scattering properties by exogenous agents would require pharmacologically adjusted compounds, e.g., small air-filled particles with tumor-localizing properties (which are expected to be difficult to achieve) we focused on NIR dyes modifying tissue absorption and fluorescence. The use of a fluorescent contrast agent with high affinity to tumor tissue might be the method of choice^{15 20 21} to overcome present limitations of optical transillumination imaging. A detailed theoretical description of the propagation of fluorescence photons in homogeneous media with a local inhomogeneity is given in Refs. 22 and 23. The use of fluorescent contrast agents with excellent enrichment was described in phantoms,²⁴ and subsequently applied in animal studies (dogs) for the detection of naturally occurring mammary disease^{25 26} in reflection mode. In these studies photomultiplier-based and intensified charge coupled device (CCD)-based frequency-domain measurements were taken.

The aim of the present study was to investigate methods to improve the contrast of optical mammograms. One particular objective was to elucidate possible differences in the reduced scattering coefficient between tumor tissue and normal tissue of the human breast performing measurements in vitro. Second, we explored how contrast can be improved in transillumination images, using 5 cm thick liquid scattering phantoms with tissuelike optical properties and with a local inhomogeneity modeling a geometrically defined tumor area of enhanced absorption and fluorescence. To our knowledge, this is the first time such investigations have been performed using a time-domain mammograph. The particular NIR fluorescent dye used might be suitable as an optical contrast agent and was developed for the demarcation of mammary tumors by fluorescence.

2.

Experiment

3.

Results and Discussion

In Figure 1(a) the reduced scattering coefficient of tumor tissue and corresponding normal tissue is depicted in the wavelength range between 600 and 800 nm for a particular patient. As expected the reduced scattering coefficient decreases with increasing wavelength for both tumor and normal tissue in accordance with results obtained by Tromberg et al.;¹⁸ in vivo at six wavelengths between 670 and 960 nm. This wavelength dependence was similar for all specimens.

Figure 1

Reduced scattering coefficient of tumor tissue and normal breast tissue of one selected patient versus wavelength (a). Reduced scattering coefficients of normal breast tissue and corresponding tumor tissue (connected by straight lines) of the same patient taken at λ=600 nm (b) and λ=700 nm (c). A total of 20 patients were investigated.

No general conclusions can be drawn from the ratios of the reduced scattering coefficients of tumor tissue and normal tissue ex vivo. This is evident from Figures 1(b) and 1(c) depicting the reduced scattering coefficients of tumor tissue and normal tissue obtained from 20 patients at λ=600 nm (b) and λ=700 nm (c). Obviously, the reduced scattering coefficients cannot be used to discriminate tumor and normal tissue. In Ref. 19 the same conclusion was reached. Absorption coefficients of tumor and normal tissue in vivo cannot be deduced from measurements ex vivo. In addition, it is known that detection of tumors in optical mammograms reflecting differences in absorption might be hampered by absorption of large blood vessels or of nonmalignant lesions. Therefore, we studied exogenous agents capable of adding biological specificity to optical imaging procedures, in particular optical mammography. For this purpose, we used the SIDAG tricarbocyanine dye,^{29 30} which exhibits a high extinction coefficient and a moderate fluorescence quantum yield.

In Figure 2 the normalized absorption and fluorescence emission spectra of the SIDAG fluorophor are shown. Due to the broad absorption spectrum peaked at 753 nm, the dye can be excited effectively between 670 and 750 nm. The wavelengths of the lasers used for frequency-domain and time-domain measurements are indicated by black solid and dashed lines, respectively. The fluorescence spectrum of SIDAG exhibits a maximum around 792 nm, which corresponds to a Stokes shift of about 40 nm. It follows that the major parts of the absorption and emission spectra cover a spectral range where scattering and absorption of tissue is low. This allows recording of fluorescence photons originating from thick tissue samples like the human breast.

Figure 2

Normalized absorption (solid line) and emission (dotted line) spectra of the dye SIDAG in bovine plasma (concentration 2 μmol/L). Vertical solid and dashed lines indicate the wavelengths used in the frequency-domain and the time-domain experiments, respectively.

Using this dye contrast enhancement between tumor and surrounding tissue was studied in a rat model with a tumor grown from the tumor cell line MTLn-3. As demonstrated by Riefke et al.;,²⁰ this dye permitted demarcation of the tumor by fluorescence at the highest contrast about 24 h after intravenous injection. For illustration, fluorescence images of a tumor-bearing rat are shown in Figure 3. Figure 3(a) illustrates the fluorescence image prior to the application of the dye, Figure 3(b) the image taken 24 h after intravenous (i.v.) application of 2 μmol of SIDAG/kg of body weight. Figure 3(c) shows the normalized fluorescence intensity along the dotted line through the tumor [see Figure 3(b)] illustrating the high contrast between tumor and surrounding tissue. The fluorescence intensity of the remaining parts of the body was slightly higher than the autofluorescence intensity recorded initially. It follows that contributions from the SIDAG fluorescence were observed outside the tumor as well. The unspecific autofluorescence observed in rats can be caused by pheophorbides contained in the food. We have selected a specific diet to reduce this background fluorescence in animals.

Figure 3

Fluorescence images of a tumor bearing rat (MTLn-3) at a dose of the dye SIDAG 2 μmol/kg body weight: (a) anterior view before i.v. administration and (b) anterior view 24 h after administration. Intensities are represented by gray scales corresponding to counts. It should be noted that the gray scales differ by a factor of 10. (c) Shows the fluorescence spectrum recorded along the dotted line of (b).

We investigated whether at concentrations corresponding to in vivo data²⁹ this dye is suited to detect the inhomogeneity located 2.5 cm below the surface of phantom No. 1, simulating a compressed human breast of 5 cm thickness. Measurements were performed in transmission geometry using the frequency-domain mammograph and different concentrations of the dye inside the inhomogeneity of phantom No. 1. Source and detector optical fibers were moved in tandem in steps of 2 mm along a line crossing the tube. At each scan position amplitude modulation and phase shift were measured at the excitation wavelength of 690 nm. First, data were taken without any dye added to study the influence of the Teflon tube itself. Under these conditions the tube caused a phase shift of about 7° due to the higher reduced scattering coefficient of Teflon compared to the scattering solution. The tube was not visible in the modulation amplitude data, i.e., the absorption coefficient of the tube material was comparable to that of the scattering solution.

In Figure 4(a) normalized modulation amplitudes are shown recorded along the scan (x) direction. There was no dye added to the scattering medium surrounding the Teflon tube. The concentration of SIDAG inside the tube was varied between 0.2 and 2.0 μmol/L. Depending on the dye concentration, the modulation amplitude of the transmitted light was reduced from about 95 to 65 at the location of the tube. The reduction in modulation amplitude is caused by the increased absorption of the laser light inside the tube.

Figure 4

Normalized modulation amplitudes along a line scan perpendicular to the Teflon tube at x=10 cm of phantom No. 1. The concentration of the dye within the tube was varied between 0.2 μmol/L (open diamonds), 0.5 μmol/L (closed circles), 1.0 μmol/L (open squares), and 2.0 μmol/L (closed triangles), no dye was added to the surrounding scattering solution (a). The dye concentration outside the tube was 0.1 μmol/L, the concentration within the tube was varied between 0.1 μmol/L (solid line), 0.2 μmol/L (open diamonds), 0.4 μmol/L (closed circles), and 0.8 μmol/L (open squares). (b) Taking the solid line of (b) as the baseline, (c) illustrates modulation amplitudes corrected for the influence of the wall of the tube.

The normalized modulation amplitudes shown in Figure 4(b) correspond to a dye concentration of 0.1 μmol/L in the scattering medium surrounding the tube. Inside the tube, the concentration was varied between 0.1 and 0.8 μmol/L. Under these conditions, the absorption coefficient of the tube is no longer matched to that of the surrounding medium. When the dye concentrations of the scattering medium inside and outside the Teflon tube are the same (0.1 μmol/L) the smaller absorption by the Teflon should result in an increased modulation amplitude. This is visible in Figure 4(b) where the increase in amplitude amounts to about 10 compared to the surrounding medium. Due to the mismatch in absorption coefficients between the tube and the scattering solutions this curve represents the baseline for investigations with other dye concentrations inside the tube. As expected, modulation amplitude decreases with respect to the baseline with increasing dye concentration inside the tube. Figure 4(c) shows the corresponding relative changes of the modulation amplitude with respect to the (smoothed) baseline. At the highest dye concentration used (0.8 μmol/L) the observed decrease in amplitude amounts to about 6.

To minimize the influence of the tube material, we replaced the Teflon tube by a dialysis tube with a much thinner wall in phantom No. 2. This phantom was investigated using the time-domain mammograph. From distributions of time of flight of photons measured at the laser wavelength of 750 nm, photon counts in different time windows were calculated. Besides the total photon count N _tot , an early time window was analyzed extending from t=0 up to 1.73 ns. The width of the early time window was chosen such that at a reference position far from the inhomogeneity photon counts N₁ in this time window amounted to 10 of all photons N _tot detected at this position. When the scattering solution was the same inside and outside the tube it could not be detected by N _tot or N₁, independent of the dye concentration. Besides absorption of laser photons by the dye, we also investigated the fluorescence signal. To this end, the phantom was scanned a second time with a band pass filter in front of the detector to spectrally select the fluorescence signal. From measured distributions of times of flight of fluorescence photons, we determined total fluorescence photon counts N _tot,fl and photon counts N _1,fl in an early time window extending from t=0 up to 2.25 ns. Again, the early time window was chosen such that at the reference position photon counts N _1,fl amounted to 10 of all fluorescence photons (N _tot,fl ) detected at this position.

The results shown in Figure 5 were obtained at a dye concentration of 0.02 μmol/L in the surrounding medium and a tenfold higher dye concentration in the scattering solution contained in the tube. Figure 5(a) displays time-integrated intensity, i.e., N _tot , of transmitted laser photons normalized to the corresponding value at the reference position. At the position of the tube the number of transmitted photons detected is reduced by about 10 compared to the surrounding medium. A comparable reduction was obtained for early arriving photons. In Figure 5(b) the number of detected fluorescence photons N _1,fl of the early time window is depicted, normalized to the corresponding value at the reference position. At the position of the tube N _1,fl is higher by 10, whereas the total intensity N _tot,fl is larger by about 5 only. To improve contrast we have calculated the ratio of the fluorescence and transmittance signals depicted in Figures 5(b) and 5(a). Figure 5(c) shows the resulting image where contrast enhancement amounts to about 22. In Figure 5(d) line scans are shown corresponding to the transmittance, fluorescence, and ratio images. The line scans were derived by averaging all eight scan lines of the images.

Figure 5

Image of a cylindrical inhomogeneity 7 mm in diameter with a tenfold higher concentration of the fluorescence dye SIDAG compared to the surrounding medium (phantom No. 2). Normalized total photon counts N _tot at the laser wavelength (a). Normalized fluorescence photon counts N _1,fl corresponding to the early time window (b). Ratio of normalized fluorescence phantom counts N _1,fl and normalized total photon counts N _tot at the laser wavelength (c). Mean normalized photon counts and corresponding ratios derived from eight line scans along the x axis see [(a)–(c)], (d).

Our results may be compared to recent reports on NIR imaging of phantoms and tumors using protease activated fluorescent probes. Weissleder and co-workers^{21 33} injected 250 pmol (5 μmol/L×50 μL) of the dye Cy5.5 subcutaneously into the flank of nude mice. This amount could be detected through about 1 mm of skin at a signal/noise ratio=9.4 using a 10 s acquisition time. In our case (see Figure 5) the concentration of 1 μmol/L used corresponds to 385 pmol/cm tube length. The fluorescent object was detected against a fluorescing background at a contrast/noise ratio of about 5 through 5 cm of scattering solution at an acquisition time of 1 s per scan position. Although both experiments are not strictly comparable, the same amount of fluorescent dye was detected in each case.

We conclude this section by discussing the time required to record laser pulse fluorescence mammograms. It was shown in Ref. 8 that time-domain optical mammograms could be recorded within 3–5 min by sampling transmitted laser photons at each of 1000–2000 scan positions using 5 mW of incident laser power and for breast thickness between 4 and 8 cm. It follows from our phantom experiments that fluorescence intensity for the dye concentrations investigated is weaker by at least a factor of 10–20 compared to excitation light. This reduced signal level could be partly compensated, e.g., by increasing data acquisition time or incident laser power, or by parallel data acquisition. Provided that similar dye concentrations and degrees of enrichment can be achieved in vivo as in our phantom experiments, recording of time-domain fluorescence mammograms may be feasible for compressed breasts of moderate thickness.

4.

Conclusion

The ex vivo results presented in this paper illustrate that reduced scattering coefficients alone do not allow us to generally discriminate small (1 cm³) breast tumors from normal breast tissue and from nonmalignant lesions of human breast with sufficient specificity. A promising way to improve the specificity of optical mammography is the application of contrast agents to modify the absorption and fluorescence properties of small tumors. It was shown that the SIDAG dye was enriched in a mammary tumor of a rat model up to a ratio of 6:1 and that no significant fluorescence originated from surrounding tissue. If this dye were approved for human studies it might favorably be used as a potential fluorescence tumor marker. Furthermore, we demonstrated that absorbing and fluorescent inhomogeneities containing this dye at concentrations as low as 0.2 μmol/L within a 5 cm thick, scattering, absorbing, and fluorescent phantom could be detected by recording time-resolved transmittance. Improved contrast was obtained by calculating the ratio of fluorescence and transmittance images, the latter taken at the wavelength of laser photons.