2.1.

Limited-Projection-Angle FMT-XCT Acquisition and Coregistration

Acquisition and coregistration of FMT and XCT datasets were performed as previously described in Ref. 25. In brief, anesthetized mice were placed in an imaging cartridge consisting of two transparent parallel plates. These served the purpose of immobilizing the mice during the different measurement steps and for translation between imaging devices. Also, it provided a common coordinate system that could be used for coregistration of all datasets due to its rigid shape.

FMT images were acquired in transillumination mode using a commercially available limited-projection-angle FMT device (FMT 2500, VisEn Medical, Woburn, Massachusetts—now PerkinElmer) with two channels at $680/700$ and $750/770\mathrm{nm}$ excitation/emission wavelength, respectively. The acquisition time for each channel was 5 to 10 min.

XCT data were acquired using a micro-PET-XCT device (Inveon, Siemens Preclinical Solutions, Knoxville, Tennessee) in 10 to 15 min and automatically reconstructed by the manufacturer’s software. The reconstructed datasets were exported in digital imaging and communication in medicine (DICOM) format to be coregistered with the FMT data. Segmentation of bones, lungs, and heart was performed as previously described.^25,32 For the subcutaneous tumor mice, an additional segmentation of the tumor area was performed. This area was selected manually based on the XCT images.

2.2.

PET Acquisition, Coregistration, and Reconstruction

PET data were acquired using the same micro-PET-XCT device as for XCT data acquisition. After XCT imaging, the mouse was automatically passed on into the PET device. [${}^{18}\mathrm{F}$]-FDG-PET acquisition took around 8 min.

Reconstruction of PET data was realized by the manufacturer’s software and DICOM images were exported for coregistration to FMT-XCT data. Since PET and XCT data were acquired with a hybrid device, coordinates for coregistration of both datasets were given by the device manufacturer, and coregistration to FMT-XCT reconstructions was therefore straightforward.

2.3.

FMT-XCT Reconstruction

The reconstruction process was performed as described in Ref. 25, except that automatic schemes for spatially varying regularization were researched and employed here instead of manually chosen priors.

Briefly, photon propagation was modeled based on a solution of the diffusion equation, implemented using a finite element solver.²² For a set of measurements between multiple source-detector pairs contained in vector $y$ and obtained from an unknown fluorescence distribution $x$, the solver calculated a weight matrix leading to a linear matrix equation

Each row in $W$ contains the relative contribution of all different voxels in the image $x$ for a single source-detector measurement. Correspondingly, different rows contain a similar calculation for each source-detector pair measurement.

The inversion of Eq. (1) leads to an ill-posed problem that needs regularization of the residual norm, which is often realized by a Tikhonov type regularization matrix.^9,11 It is known that the reconstruction of $x$ can be improved by integrating anatomical information in the regularization term of the inverse problem.^20,25 This results in the following expression:

2.4.

Automatic Estimation of Regularization Factors for Reconstructions Using Priors

A data-driven two-step inversion method was previously proposed in the context of 360-deg systems^21,31 for automatically estimating spatially varying regularization factors in the $L$-matrix, employed to establish priors in Eq. (2). The two-step inversion method consists of a first inversion step, performed using anatomical information as prior information in the forward model and employing Eq. (2) with $L=I$ ($I$ being the identity matrix) to derive the relative fluorescence strengths per segment. Regularization factors for each segment are then estimated from reconstructed intensity values in the different tissue segments; typically, proportionally to the calculated mean signal intensity per segment. The second inversion step incorporates these segment-specific regularization factors into the diagonal of the $L$-matrix in order to achieve spatially varying regularization in combination with an adequate regularization parameter $\lambda$. The advantage of the two-step data-driven approach is that it is free of user-based assumptions or heuristic selection of regularization factors.

Stand-alone FMT reconstructions based on limited-projection-angle (planar) geometries tend to have surface artifacts.³¹ This effect is more pronounced as the number of projections decreases.³³ In addition, limited projections limit the resolution achieved along the projection axis.³⁴ Our approach, herein, consisted of investigating whether automatically computed regularization factors can lead to improvements in these artifacts and yield more accurate images. For this purpose, we examined how different regularization parameters $\lambda$ would influence the computation of regularization factors in the step 1 inversion. Our objective was to find an ideal $\lambda$ for the automatic estimation of regularization factors for the step 2 inversion. This was initially tested on phantom data containing fluorescent insertions.

During the step 1 inversion described above, Tikhonov reconstruction using the identity matrix was performed and an initial regularization parameter was chosen by $L$-curve analysis at the $L$-curve corner.³⁵ Subsequently, we increased the value of $\lambda$ until we created an over-regularized state represented by smoothing the entire reconstruction toward a single bright lesion in the middle of the phantom. The motivation for this $L$-curve analysis was to find a value for $\lambda$ where the regularization is high enough to suppress surface artifacts, but still enable a representative overall fluorescence distribution. The best $\lambda$ value identified in this process was then used in the step 1 inversion to derive segment-specific regularization factors, the latter employed in the $L$-matrix of the step 2 inversion. The segment-specific regularization factors were determined by computing the mean reconstructed intensity $I_{\mathrm{seg}}$ per segment

2.5.

Phantom Characteristics

The phantoms and corresponding acquisition data employed in this study have been reported in Ref. 25. The phantoms were composed of a mixture of intralipid, ink, agar, and water according to the known spectral properties of ink and published scattering properties of intralipid, in order to attain tissue properties of ${{\mu }_{\mathrm{s}}}^{\prime }=12\mathrm{c}{\mathrm{m}}^{-1}$ and ${\mu }_{\mathrm{a}}=0.2\mathrm{c}{\mathrm{m}}^{-1}$. They were shaped to semicylinders resembling the shape of a mouse torso, with a maximal diameter of 30 mm. Two tubes filled with a fluorescence dye with the peak excitation at 679 nm and peak emission at 702 nm (Alexa Fluor 680, Life Technologies Ltd, Paisley, UK) were inserted into the phantom to mimic fluorescently labeled tumors.

The phantom data were used to develop and test the automatic regularization factor estimation from coregistered limited-projection-angle FMT-XCT on clearly defined and well-known tissue before moving on to the more heterogeneous mouse models where the fluorescence biodistribution is typically unknown prior to ex vivo validation.

2.6.

Animal Studies

Three mouse models were employed in this study, two for studying superficial tumors and one exhibiting deeper-seated tumors. Optical signals attenuate nonlinearly with the source depth. Consequently, the selection of different depths is important in the investigations of optical imaging to examine the ability of the method to reliably perform as a function of depth.

Deep-seated tumors were investigated using a transgenic mouse model spontaneously developing tumors in the lung (Kras, Ref. 36). Mice at the age of approximately 15–18 weeks were injected intravenously before imaging with, respectively, 2 nmol of two different fluorescence probes, one targeting ${\alpha }_{\mathrm{v}}{\beta }_3$-integrins (IntegriSense680, Perkin Elmer, Waltham, Massachusetts) and a blood-pool agent (AngioSense750 Perkin Elmer, Waltham, Massachusetts). The excitation/emission maxima of IntegriSense680 and AngioSense750 are approximately at $680/705\mathrm{nm}$ and $750/770\mathrm{nm}$, respectively. AngioSense can be used in oncology to study angiogenesis and was, therefore, chosen to provide information on vascularization and perfusion, in addition to the molecular information granted by the ${\alpha }_{\mathrm{v}}{\beta }_3$-integrin overexpression. AngioSense was injected in different mice both 1.5 and 24 h prior to imaging but did not show significant biodistribution differences between these two time points as confirmed by ex vivo validation studies. We hence only show, herein, results from a mouse injected 24 h prior to FMT-XCT measurements, i.e., at the same time point for which IntegriSense data was also acquired. This study examined, therefore, a physiological parameter (vascularization/perfusion) and a molecular parameter (${\alpha }_{\mathrm{v}}{\beta }_3$-integrin overexpression).

The second mouse model was a xenograft breast cancer model. We subcutaneously injected Balb/C nude mice with ${10}^6$ 4T1 cells in the neck region and let the tumor grow for 10 days. Twenty-fours hours prior to imaging, mice were intravenously administered scVEGF/Cy (SibTech Inc., Brookfield, Connecticut), a fluorescence agent binding to VEGF receptor 2 (VEGFR-2) and emitting at around 700 nm. This agent was employed to study the ability to visualize a second molecular target in one channel, whereby the second channel again imaged IntegriSense750. In this way, we could examine the relative distribution patterns of VEGFR-2 and ${\alpha }_{\mathrm{v}}{\beta }_3$-integrin.

The third mouse model was employed to demonstrate the capacity of hybrid FMT-XCT and to combine the findings with results from PET imaging. In this pilot study, $6\times {10}^6$ Lewis lung carcinoma (LLC) cells were injected subcutaneously in the right shoulder region and $4\times {10}^6$ cells in the left shoulder region of a nude mouse. The tumors were allowed to grow for 9 days. Twenty-four hours before FMT-XCT imaging, the mice was injected with IntegriSense680. Additionally, 45 min before PET imaging the mice were injected with approximately 13 MBq of [${}^{18}\mathrm{F}$]-FDG tracer.

For imaging, mice were anesthetized (Isoflurane 2%, ${\mathrm{O}}_2$ $0.9\mathrm{L}/\min$), placed in an imaging cassette, and consecutively imaged by XCT, PET (only for LLC mice), and FMT. After imaging, they were euthanized and frozen at $-80°\mathrm{C}$ for ex vivo validation.

2.7.

Imaging Performance Validation

For validation, the frozen mice were cryosectioned in a cryotome (CM 1950, Leica Microsystems GmbH, Wetzlar, Germany). During slicing, we employed an imaging system³⁷ to acquire images of the real fluorescence distribution in every tissue slice at both wavelengths to be compared to our in vivo results. Images were acquired every 200 μm.

We compared the relative contrast of the fluorescence strength between tumors and different tissues in vivo and ex vivo in order to assess the performance of our reconstruction method. For that, we computed the relative contrast $C_{T1,T2}$ between any two regions $T1$ and $T2$ analogously to standard contrast-to-noise calculations as:

For Kras mice, we chose regions of interest in the tumor, muscle, lung, and heart and for 4T1 mice, we chose ROIs in the center and margin of the tumor, as well as in muscle tissue near the tumor. We also quantified the relative contrast reconstructed in vivo in each of the whole tissue segments.

The accuracy of the localization of the reconstructed fluorescence signal in the phantom was evaluated as a function of the regularization parameter $\lambda$ used for respective regularization factor computation. To assess the reconstruction error, we computed the percentage of the reconstructed signal in an area delineated by a distance of 2 mm around the fluorescence tubes (as described in Ref. 25). We further evaluated the accuracy of the localization in Kras, 4T1, and LLC experiments based on intensity profiles through the in vivo and ex vivo images. The threshold set for assuming that a fluorescence signal originates from a location of true probe accumulation was set as the mean signal intensity in the entire mouse plus two standard deviations of background noise (as previously described in Ref. 38).

3.1.

Evaluation of Regularization Factor Estimation

Figure 1(a) depicts four phantom reconstructions with different values of $\lambda$ using standard Tikhonov regularization, and the corresponding $L$-curve. The shown representative values of ${\lambda }_n$ included ${\lambda }_1$ and its $2^{n-1}$ multiples (${\lambda }_2\approx {\lambda }_1·2^1$; ${\lambda }_3\approx {\lambda }_1·2^2$; ${\lambda }_4\approx {\lambda }_1·2^3$). Reconstruction for ${\lambda }_1$ shows an elongated appearance of the circular objects (fluorescent tubes) with the peak intensity biased toward the surface. As the $\lambda$ values increase from ${\lambda }_1$ to ${\lambda }_4$, the reconstructed image appears more diffusive in nature but the elongation is retained, due to the limited angle projection data collected by the system. Figure 1(b) lists the values of ${\lambda }_1$ to ${\lambda }_6$ in its rows and the corresponding regularization factors for the two tubes and the background in its columns. The left tube was assigned the lowest regularization factor of 1 for all cases while the regularization factor for the right tube changed with $\lambda$. The step 2 reconstructions resulting from using the regularization factors computed with ${\lambda }_1$ to ${\lambda }_6$ are evaluated in the rightmost two columns. The normalized relative contrast (i.e., contrast values divided by the largest contrast value, which was 17.5) between right tube and tissue and the percentage of reconstructed signal lying up to 2 mm outside the tube margin are given. Figure 1(c) shows the computed regularization factors for the right tube as a function of all $\lambda$’s starting at the $L$-curve corner with ${\lambda }_1$ as a continuous line. We observed that the regularization factors between ${\lambda }_1$ and ${\lambda }_3$ are in a similar value range, with a local minimum at ${\lambda }_3$, but rise quickly for higher values of $\lambda$. The dashed line shows the change of the normalized relative contrast in the right tube after step 2 inversion as a function of the used regularization parameter for regularization factor estimation. The dotted line depicts the change of the percentage of the reconstructed signal lying within a distance of 2 mm from the tube. Figure 1(d) finally shows the examples for the fluorescence distribution after step 2 inversion using the regularization factors computed from step 1 inversion with ${\lambda }_3$, ${\lambda }_5$, and ${\lambda }_6$. The results from Figs. 1(b) and 1(c) suggest that the relative contrast decreases for regularization factors computed with $\lambda >{\lambda }_3$ and the localization accuracy for $\lambda >{\lambda }_4$. This becomes obvious in the presented examples. The top two phantom slices show the same reconstruction with regularization factors computed with ${\lambda }_3$, but with different scaling to bring out the maximum signal in the right tube. The two reconstructions on the bottom of Fig. 1(d) for ${\lambda }_5$ and ${\lambda }_6$ are also scaled to the maximum value of the signal in the right tube. Here, it becomes obvious that similarly high signals are also reconstructed outside the tube area (arrows). Red circles highlight the region that was used for the localization evaluation, i.e., tube radius $+2\mathrm{mm}$. The independent regularization parameter $\lambda$ used in step 2 was chosen to be the commonly used optimal solution lying at the $L$-curve corner. In contrast to the phantom reconstruction using priors shown in Ref. 25 where the regularization factors were manually set, the result in Fig. 1(d) was achieved by automatic computation based on experimentally determined regularization factors.

Fig. 1

(a) Tikhonov reconstruction (step 1 inversion) of fluorescence inclusions (red circles) in a tissue-mimicking phantom for different regularization parameters $\lambda$, marked on the corresponding $L$-curve. (b) Regularization factors computed from the respective reconstructions using six different $\lambda$ values and resulting contrast ratios and localization accuracy. (c) Computed regularization factor of the right tube as a function of $\lambda$ (continuous line), normalized contrast between right tube and tissue (dashed line), and percentage of signal localized within 2-mm distance of the right tube (dotted line). (d) Reconstructions using regularization factors calculated for ${\lambda }_3$, ${\lambda }_5$, and ${\lambda }_6$ (step 2 inversion). The red circles represent the area for localization evaluation defined by the tube radius $+2\mathrm{mm}$. The chosen regularization parameter for step 2 lies near the point of maximum curvature of the step 2 $L$-curve, at the $L$-curve corner.

3.2.

Dual-Wavelength Experiments

Table 1 lists the computed regularization factors after step 1 inversion for both Kras and 4T1 mouse models at the two employed wavelengths and represents the mean reconstructed fluorescence strength per $\lambda$ (columns) and segment (rows).

Table 1

Regularization factors computed for different values of λ after Tikhonov reconstruction of Kras and 4T1 mouse models. Columns contain values for each respective λ and rows contain values for each respective segment. For both wavelengths, the model-specific segments are allocated a factor between 1 and 3 representing the mean reconstructed fluorescence strength in this segment. A value of 1 means highest fluorescence signal, a value of 3 lowest fluorescence signal.

For Kras mice, step 1 inversion with ${\lambda }_1$ suggests that the fluorescent probe uptake at both wavelengths was highest in the bones. This would result in a segment-specific regularization factor of 1 in the $L$-matrix of Eq. (2), representing the step 2 inversion. The other segment-specific entries into the $L$-matrix, i.e., for tissue, lung, and heart, at 680 nm would for instance be 1.6, 2.21, and 3, respectively. The computation of these segment-specific regularization factors changes as a function of increasing $\lambda$, as can be seen in Table 1 for the values computed with ${\lambda }_2$ and ${\lambda }_3$. In contrast to the results obtained with ${\lambda }_1$, step 1 inversion with ${\lambda }_3$ suggests that the highest fluorescence probe uptake at 680 nm is in the lung. In this case, the elements of the $L$-matrix corresponding to the lung segment would be assigned a value of 1.

Figure 2 shows the results after step 2 inversion of the Kras mouse model using the respective regularization factors from Table 1 for ${\lambda }_3$. Figure 2(a) depicts the two in vivo reconstruction slices at different locations in the chest of the mouse and Fig. 2(b) contains the corresponding ex vivo cryosections for validation. An example for the choice of regions of interest for the determination of the relative contrast between tissues is given in the top in vivo and ex vivo slices by dashed white ellipses for tumor (t), muscle (m), lung (l), and heart (h). The bottom slice in Fig. 1(a) additionally shows the segmentation of lung and heart in the in vivo data as white dashed contour lines. Figure 2(c) depicts a three-dimensional (3-D) rendered image of the whole reconstructed volume. Here, the lung is shown in transparent light yellow in order to allow the evaluation of the location of the reconstructed fluorescence distribution. In all panels of Fig. 2, the cyan signal represents the distribution of IntegriSense680 and magenta the distribution of AngioSense750. Cyan arrows highlight the tumors and magenta arrows the heart. A clear difference in the biodistribution of the two probes is observable in the ex vivo slices. This is reflected in the in vivo reconstruction, where IntegriSense680 is mainly reconstructed in the tumor area and AngioSense750 is more broadly distributed in the lung and heart area.

Fig. 2

Reconstruction of Kras tumor model using two different fluorescence probes and the two-step inversion approach. (a) Overlay of two representative slices of in vivo reconstructions of IntegriSense680 (cyan) and AngioSense750 (magenta) of the Kras mouse. Regions of interest in tumor (t), lung (l), muscle (m), and heart (h) for contrast evaluation are shown in the top slice. White dashed contours in the bottom slice show the segmentation of lung and heart. (b) Overlay of the corresponding fluorescence cryosection images. (c) Three-dimensional (3-D) representation of the reconstructions throughout the volume. (d) Intensity profiles through the straight white dashed lines in (a) and (b) of IntegriSense680 (cyan) and AngioSense750 (magenta) for in vivo (dashed lines) and ex vivo (continuous lines), respectively. (3-D rendering was implemented using AMIRA software).

Figure 2(d) shows the normalized intensity profiles through the dashed straight white lines in the in vivo and ex vivo slices in Figs. 2(a) and 2(b). Dashed lines in the graph represent in vivo and continuous lines ex vivo data. An arrow highlights the position of the biggest tumor in the ex vivo slice [Fig. 2(b)] and the graph to facilitate correlation of corresponding data points. The ex vivo intensity profile shows high IntegriSense680 uptake in the tumor compared to the surrounding lung as well as to heart and the other tissue. Ex vivo fluorescence distribution of AngioSense750 in tumor, lung, and heart on the contrary is of a similar intensity level. The in vivo profiles qualitatively mirror the ex vivo profiles when it comes to recovering the highest fluorescence peaks. For AngioSense750, though, a descent between tumor and heart area can be observed that is higher than observable in the ex vivo validation slices. We further examined the agreement between in vivo reconstructions and ex vivo cryosections by comparing the relative contrast between tissue segments and between selected regions of interest, as described in Sec. 2. Table 2 lists these values where rows contain the different contrast ratios, columns the respective fluorescent probe. The compared tissues are given in the “Regions” column, where the first tissue goes as ${\mu }_{T1}$ and the second tissue as ${\mu }_{T2}$ into Eq. (4), respectively. Highest whole segment contrast for IntegriSense680 was obtained for the lung region and accordingly for ROIs in the tumor, both in vivo and ex vivo. Although the contrast between the whole lung and other tissues is still rather low due to the contribution of both diseased and normal lung tissue, contrast ratios rise significantly when focusing on the contribution of the tumors alone. For AngioSense750, the small absolute value of the lung/heart contrast ratio indicates little contrast between those segments. Similarly high values were also obtained when comparing localized tumor signals to both heart and lung.

Table 2

Comparison of in vivo and ex vivo contrast ratios in Kras mouse model. Left: Computed relative contrast for whole segments for both wavelengths. Right: Relative contrast of model specific regions of interest. For Kras: tumor-to-lung, tumor-to-muscle, and tumor-to-heart contrast.

Figure 3 shows the results after step 2 inversion of the 4T1 mouse model using the respective regularization factors listed in Table 1 for ${\lambda }_3$. Figure 3(a) depicts two in vivo reconstruction slices at different locations in the neck/shoulder region of the mouse and Fig. 3(b) contains the corresponding ex vivo cryosections for validation. An example for the choice of ROIs is given in the top in vivo and ex vivo slices by dashed white ellipses for tumor margin (${\mathrm{t}}_{\mathrm{m}}$), tumor center (${\mathrm{t}}_{\mathrm{c}}$), and muscle (m). The bottom slice in Fig. 3(a) shows the contour of the tumor segment used for in vivo reconstruction. In all panels of Fig. 3, the cyan signal represents the distribution of scVEGF/Cy and magenta the distribution of IntegriSense750. Both in vivo and ex vivo slices show IntegriSense750 accumulation in the center of the tumor whereas scVEGF/Cy is accumulated at the tumor margin (cyan arrows). Figure 3(c) depicts a 3-D rendered image of the reconstructed fluorescence at both wavelengths. The reconstruction of scVEGF/Cy uptake throughout the mouse is equivalent as for the single slices shown in Fig. 3(a), namely on the tumor boundary. This probe mainly accumulates in the margins (note the donut shape of the reconstruction) and IntegriSense750 is in its center, with some overlap of both probes in the peripheral regions of the tumor.

Fig. 3

Reconstruction of 4T1 tumor model using two different fluorescence probes and the two-step inversion approach. (a) Overlay of scVEGF/Cy (cyan) and IntegriSense750 (magenta) in two representative slices from in vivo reconstruction. Regions of interest in tumor margin (${\mathrm{t}}_{\mathrm{m}}$), tumor center (${\mathrm{t}}_{\mathrm{c}}$), and muscle (m) for contrast evaluation are shown in the top slice. The white dashed contour in the bottom slice shows the segmentation of the tumor. (b) Overlay of the corresponding fluorescence cryosection images. (c) 3-D representation of the reconstructions throughout the volume. (d) Intensity profiles through the straight white dashed lines in (a) and (b) of scVEGF/Cy (cyan) and IntegriSense750 (magenta) for in vivo (dashed lines) and ex vivo (continuous lines) data, respectively. (3-D rendering was implemented using AMIRA software).

The same quantitative analysis as for the Kras model was performed on the 4T1 model. Figure 3(d) depicts the intensity profiles through the dashed white straight lines in ex vivo and in vivo slices in Figs. 3(a) and 3(b). The graph shows in cyan the normalized intensity of scVEGF/Cy emission and in magenta the normalized intensity of IntegriSense750 emission. Dashed lines represent the in vivo and continuous lines ex vivo data. An arrow highlights the position of the tumor boundary in the ex vivo slice [Fig. 3(b)] and the graph to facilitate correlation of corresponding data points. The intensity profile of scVEGF/Cy shows the highest intensities at the tumor margins while the highest peak for IntegriSense750 is in the tumor center, both for ex vivo and in vivo data. Contrast ratios for the whole tissue segments and for the selected regions of interest are listed in Table 3. The whole segment contrast ratios show that both probes are accumulated in the tumor segment. Contrast between ROIs in the tumor center and margin furthermore confirm the exact localization of the accumulation of each probe, i.e., in the margin for scVEGF/Cy and in the center for IntegriSense750. In vivo and ex vivo contrast ratios were in qualitative agreement, i.e., highest contrast was achieved for margin/tissue for scVEGF/Cy and center/tissue in both reconstructions and cryosections.

Table 3

Comparison of in vivo and ex vivo contrast ratios in 4T1 mouse model. Left: Computed relative contrast for whole segments for both wavelengths. Right: Relative contrast of model specific regions of interest. For 4T1: tumor-margin-to-center, tumor-margin-to-tissue, and tumor-center-to-tissue ratios.

3.3.

FMT-XCT-PET Study

Figure 4 shows the reconstruction results of the LLC mouse model including FMT, XCT, and PET data. In this case, the expression of ${\alpha }_{\mathrm{v}}{\beta }_3$-integrin was tracked with an optical probe and metabolic activity with an [${}^{18}\mathrm{F}$]-FDG PET radiotracer. The reconstruction of IntegriSense680 was performed analogously to the phantom, Kras, and 4T1 studies, i.e., using ${\lambda }_3\approx 2^2·{\lambda }_1$, with ${\lambda }_1$ being the regularization parameter at the $L$-curve corner of the step 1 regularization of the LLC mouse. Figure 4(a) depicts the different distribution of the two tracers, the in vivo reconstruction of IntegriSense680 being shown in orange and the [${}^{18}\mathrm{F}$]-FDG-PET tracer in green. Figure 4(b) shows the ex vivo validation of the optical probe distribution where the fluorescence signal is shown as contrast-enhanced orange overlay on a white light image. Figure 4(c) visualizes the 3-D distribution of IntegriSense680 and [${}^{18}\mathrm{F}$]-FDG reconstructions and Fig. 4(d) indicates the locations of the subcutaneous LLC tumors on the 3-D rendered mouse skin by orange arrows. Figure 4(e) depicts the normalized intensity profiles through the white dashed lines in Figs. 4(a) and 4(b). Continuous curves represent the profiles through line 1 and dashed curves through line 2. The graph shows in orange the normalized intensity of IntegriSense680 emission in the ex vivo slice [Fig. 1(b)], in yellow the in vivo IntegriSense680 reconstruction [Fig. 1(a)], and in green the normalized intensity of [${}^{18}\mathrm{F}$]-FDG. A gray arrow highlights the position of the left tumor in the ex vivo slice [Fig. 4(b)] and the graph to facilitate correlation of corresponding data points.

Fig. 4

Reconstruction of LLC mouse model and coregistration of FMT-XCT-PET. (a) In vivo reconstruction of IntegriSense680 (orange) and overlay of FDG-PET reconstruction (green). The green arrow highlights a region where the PET reconstruction indicates higher FDG uptake. FMT reconstruction in the same area does not show any fluorescence signal. (b) Ex vivo validation slice of IntegriSense 680 fluorescence (orange). The fluorescence image was enhanced for better contrast. (c) 3-D representation of both FMT (orange) and PET (green) reconstructions. Orange arrows point to the tumors as reconstructed by FMT and PET. The green arrow points to a third region of increased FDG uptake. (d) 3-D rendering of mouse skin and highlighted tumor areas (arrows). (e) Intensity profiles through the white dashed lines in (a) and (b) of ex vivo IntegriSense680 (orange), in vivo IntegriSense680 (yellow), and [${}^{18}\mathrm{F}$]-FDG (green) for line 1 (continuous) and line 2 (dashed), respectively. (3-D rendering was implemented using AMIRA software).

We observed that IntegriSense680 is only reconstructed in the tumor area, in accordance with the ex vivo slice image. The size of the left tumor, though, was overestimated and a slight dislocation between ex vivo and in vivo fluorescence locations could be observed. As can be expected, FDG (green) shows the metabolic activity and is therefore not only distributed in the tumors but also in heart and brain. An additional aggregation of FDG can be observed, though, between the two tumors, as highlighted by green arrows in Figs. 4(a), 4(c), and 4(e). This is neither visible after IntegriSense680 reconstruction nor is a third tumor distinguishable in the ex vivo slice.