The results of SOCT imaging of GNRs in human breast tissue samples are summarized in and . To confirm that the ρ values derived from SOCT correlate with changes in GNR density, we first compare the mean ρ values taken from each stack of SOCT images, plotted along the y direction (). For Sample 1, the control image stack is used to construct the a priori spectral database. As expected, the image stacks obtained from Sample 1 after GNR injection show a clear increase in mean ρ, whereas the image stack before treatment is nearly identical to that of the control. We note that there appears to be a positive bias in ρ for the untreated samples, which needs to be further investigated. As the injected GNR solution was at a relative density ρ of 1 (equivalent to that used for the spectral calibration), it is consistent that the resulting concentrations measured within the tissue samples were some fraction of this number.
Fig. 5 GNR contrast-enhanced SOCT images of a human breast carcinoma tissue sample. (a)–(f) Series of B-mode images before injection, sampled every 0.5 mm along y; (g)–(q) images after the second injection of GNR solution in Sample 2 (cf. (more ...)
For Sample 2, the two image stacks after GNR injections show a sizable increase in mean ρ values near y=0 relative to the control sample, as expected from the central location of the injection site. The apparent extent of GNR diffusion is approximately 1 mm from the origin. However, we note that these tissues are fixed in formalin and are expected to be less permeable than live or unfixed tissues, so the extent of GNR diffusion may be greater in a living subject.
The GNR distribution can also be superimposed onto standard OCT images, and resolved with sub-millimeter resolution in the x and z dimensions (). The structure of the tissue samples are shown in red (OCT signal), while the relative density of GNRs are plotted in green or blue (ρ>0 or ρ<0, respectively). While there is no physical meaning to negative density values, their intensities may reflect the noise levels of this technique, or a disturbance in tissue structure. For example, a few purplish regions indicating negative ρ can be found near the surfaces in images (j)–(l). A possible reason for this apparent signal depletion may be due to the forced hydration of nearby tissue caused by GNR injections, producing a local reduction in endogenous tissue scattering.
Inspection of the B-mode SOCT images acquired for Sample 2 after the second GNR injection reveals an increased number of pixels with high, positive ρ values, mostly well beneath the surface of the tissue (). This result is not unexpected, as the injection needle was inserted 1–2 mm below the tissue surface. The regions of green in images (j)–(m) are particularly bright and continuous, correlating strongly with the injection site near the center of the image stack (). In comparison, the control SOCT images () exhibit only a slight positive bias. Overall, the relative GNR densities observed in the SOCT images are consistent with the mean ρ values plotted in .
The results above suggest that the relative GNR density needed for detection in human breast tissues is about 0.3. The molar extinction coefficients of our GNRs are estimated to be in the range of 2–4 × 109
at plasmon resonance, based on characterization data for GNRs of comparable size.29–31
Based on these values, the undiluted concentration of GNRs used in this study is 10–20 nM, suggesting the limit of detection to be 3–6 nM (μt
). Lower limits are possible if the samples are spectroscopically homogeneous, as reported by others where absorption sensitivity to 5 cm−1
was obtained in homogeneous tissue phantoms.32
Considering that this study has addressed the added complexity of imaging within heterogeneously scattering tissues with irregular boundaries, the experimental SOCT sensitivity to GNRs is within an expected range.
It is worth mentioning that SOCT images of GNRs have been obtained previously in liquid tissue phantoms (Intralipid), using a different SOCT algorithm to compute the cumulative spectral shift.17
One problem with the cumulative algorithm is that while the noise is somewhat mitigated, the locations of the GNRs cannot be depth-resolved. Furthermore, a comparison of SOCT images obtained from liquid and solid specimens indicates a higher level of spectroscopic noise in the liquid phantoms. This is attributed to Brownian motion of the oil emulsion, causing Doppler shifts that scramble the spectrum. In this work, the SOCT algorithms are depth-resolved for solid tissue imaging, and produce sufficiently low noise to enable the detection of GNRs in a heterogeneous medium. While the experiments reported here demonstrate imaging after interstitial injection of nanorods into tumor specimens, we expect that in realistic in vivo
imaging scenarios the enhanced permeation and retention effect may provide tumor-specific imaging after intravenous administration, as demonstrated previously using nanoshells to image mouse tumors.33