After our work with simulations, we tested the Gen3 imager and reconstruction schemes with tissue phantoms that roughly mimic optical the properties and contrast of breast tissue. Hollow targets,16 mmin diameter (Figure 4.34) with1 mmthick walls, were filled with optical contrast
and submerged in a background solution.
Figure 4.44 shows the reconstruction using data from our Gen3 imager with the tissue phan- toms. The background solution was mixed with India ink and Intralipid to have an approximate absorption and scattering ofµa = 0.05 cm−1 andµ0s = 5 cm−1 at785 nm. The targets were
filled to have3×chromophore concentration (India ink, upper left) and2×scattering (Intralipid,
lower right) relative to the background. The top slices are closer to the source plane and the lower slices are closer to the detector plane as in the 3D simulations.
Again, as expected (but this time with experiment rather than simulation), we find that FD imaging reduces the crosstalk between absorption and scattering quite dramatically. There is a “frame” of higher contrast levels on the edges of the absorption image in the FD reconstruction in Figure 4.44, but that region lies outside the sensitivity range of our measurement and suggests
Figure 4.44: 3D reconstruction of tissue phantom. The upper left target is India ink with3× µa
contrast and the lower right target has a higher Intralipid concentration with2× µ0s contrast compared to the background. The CW reconstructions (left) have relatively low contrast and high crosstalk while the targets are well resolved spatially. In the FD reconstructions (right) noticeable artifacts near the source are apparent, but significantly less crosstalk betweenµaand
µ0sis observed. The top slices are closer to the source plane and the lower slices are closer to the detector plane (as in the 3D simulations).
that updated/better initial guesses of the background properties could improve our reconstruc- tion. In addition, the image appears to suffer from artifacts near the source plane; these artifacts near the source and detector surfaces are known to arise in DOT, and methods exist that can be employed to ameliorate them, e.g., spatially variant regularization. However, we did not im- plement the spatially variant regularization technique here; it will be explored in future work. Nevertheless, the spatial structure of the reconstructed targets are well resolved in the transverse direction with some target broadening/elongation towards the source plate for both absorption and scattering objects. Again, a lower number of sources (60 of209) and detectors (∼ 320of
5000pixels) was utilized in the tomography so that we could carry out the reconstructions faster
(∼6−8hours/reconstruction) with this data set. (Note, the small number of detectors is perhaps
less of an issue than the sources, because adjacent pixels are averaged in software to increase the SNR.)
Although the spatial structure of the images is reasonably good, one nagging question con- cerns contrast (or lack thereof) of the tissue phantoms and to a lesser extent, contrast in our 3D simulations. The actual reconstructed contrast is less than expected. This effect has been ob- served previously in DOT devices [10,44,181], and it has largely been attributed to the spreading of the reconstruction contrast that arises when the target (in the reconstructed image) is spread out in size compared to the true target dimensions. The effective contrast in such images depends on a product of the target volume and target contrast (e.g., when a reconstructed absorption tar- get is broadened in size (volume), then its absorption contrast is decreased in proportion to the volume). This problem is exacerbated in 3D, since there is an additional stretching in the lon- gitudinal direction. In addition, some of the underestimation of optical contrast is also due to
systemic offsets inherent in the instrument or reconstruction method which can be compensated with calibration phantoms [182]. In fact, an area of research that these results suggest should be pursued, concerns how to rescale the image contrast; one approach to this problem is to utilize standard targets as part of every experimental run. In the case of our absorption and scattering
Figure 4.45: 3D reconstruction of the multi-chromophore data with Nigrosin (top left) and IR806 (bottom right) dyes with2×background absorption. The absorption reconstructions looks simi- lar to earlier iterations in the simulation (Figure 4.42(a)). Increasing iterations past∼40for the data does not improve image quality as it is noise limited. Contrast level is low, but the main spatial features separate nicely.
targets, the integrated signal isΓ =δµa·V, whereδµais the fractional difference from the back-
ground, andV is the target volume. Here we will use the approximate shape of the target object to estimate volume; using the FWHM to determine the bounding volume and integrating the
δµaabsorption within that volume, we determine∆Γµa = Γreconµa /Γµexpecteda = 0.35. This result
(i.e., obtained by integration over the volume of the target object) is reasonable, especially given that our contrast is very large and will thus contribute to the signal in the nonlinear regime (i.e., outside of the small perturbing limit), and given that the object shapes are not perfect spheres, and given that our target objects have optical index of refraction differences with respect to the background that are not accounted for in the reconstructions. Similarly for the scattering con- trast we found∆Γµ0s = 0.36. For comparison, in the CW case, we found∆ΓCW,µ0s = 0.16and
∆ΓCW,µ0s = 0.05. The resolution of the scattering target is quite good (except in the longitudi- nal direction as we have discovered with our simulations, but remember we do not employ any spatially variant regularization here). As expected the absorption targets are broader and larger than the scattering targets; in the future, we will explore more TV parameters to increase the absorption resolution.
In a different type of study, we employed two tissue phantoms to test the multi-spectral capa- bilities of the Gen3 imager. The two targets shown in Figure 4.34 were each filled with Nigrosin and IR806, respectively, to give approximately2×the absorption and the same scattering as the
background Intralipid solution at785 nm(µa= 0.04 cm−1,µ0s= 8cm−1). For this data set we
increased the number of sources (209of209), while keeping the number of detectors the same
as the previous experiment (∼320of5000).
There is a clear separation of the chromophores in the image with a small amount of crosstalk, especially in the IR806 reconstruction (∼10%). Less crosstalk is apparent in the Nigrosin chro-
mophore reconstruction, and this improvement could be due to the strong spectral sensitivity of the Nigrosin across the different wavelengths; by comparison, IR806 has very low contrast at the lower wavelengths. At higher iteration number, the simulations suggest that the two chro- mophores will eventually be differentiated. Unfortunately, at this time the data reconstruction is noise limited to 80 iterations after which no further improvements or difference in image quality can be found. For Nigrosin,∆ΓN igrosin = 0.90which accounts for most of our contrast. The volume contrast for IR806 was also quite good with∆ΓIR806= 0.78. The crosstalk in the scat- tering images from the absorbing targets were found to be∼10%in this situation. Two reasons
might explain the improve contrast: the increased number of sources could improve resolution and the lower perturbation (2×background instead oftimes) of the absorbers puts the problem more in the perturbative limit. Improving resolution with increased numbers of sources and de- tectors, improving our SNR which gives greater numbers of off-axis measurements, improving absorption resolution with different regularization, are all obvious issues we hope to explore in the near future with the Gen3 imager as a result of these experiments.