2.1.

System Development

A single-channel depth sensing system was constructed using components for fiber coupling to maximize light efficiency for fluorescent sensing, to measure porphyrin production and ICG flow kinetics. The system hardware comprised a source-detector pair setup with electronic control and acquisition from the detector to the computer via Labview software. The source unit had two laser modules: an FC-coupled fiber optic–driven 780-nm laser diode with a maximum 80-mW average power (model: TECIRL-80G-780-TTL-A, World-StarTech, Toronto, Canada) for ICG imaging and an FC-coupled fiber optic–driven 635-nm laser diode with a maximum 50-mW average power (model: TECRL-50G-635-TTL-A, World-StarTech, Toronto, Canada) for PpIX imaging, respectively. Here, the 780-nm laser was directly controlled by the USB communication port of the computer for pulsed laser modulation and triggering using control commands executed in the NI-VISA function via LabVIEW. The 635-nm laser was pulsed modulated and triggered by using TTL modulation via analog output pins of the NI DAQ board. Herein, the square wave–based pulsating signal was generated from this laser, and signal detection was based on the amplitude of the modulated signal at a desired frequency. This was coupled to a modified $200\text{-}\mu \mathrm{m}$ diameter, 0.22 NA, fiber optic cable with FC-connector (Model: M122L05, Thorlabs, Newton, New Jersey, United States), which had the tissue contact end custom-made by gluing on a polylactic acid (PLA) filament-based 3D-printed (Creator 3 Pro, FlashForge, Zhejiang Flashforge 3D technology Co., Zhejiang, China) cylindrical tube to the cut fiber and polishing for a small round ferrule of 3-mm diameter, providing the necessary rigidity for positioning during imaging studies.

The detector end included a splitter fiber to split the light signal into two channels, for detection of the excitation light and the fluorescent light. Each detector was an avalanche photodiode module, with the light first going through a notch filter block. The SMA-coupled bifurcated fiber bundle had 19 fibers in the Bundle with $Ø200\mu \mathrm{m}$, SMA coupling, and 2-m length (BF19Y2LS02, Thorlabs). The excitation channel included a notch filter that blocked the PpIX excitation light and passed just the ICG excitation light (NF633-25 - Ø25 mm Notch, Thorlabs), as well as a neutral density filter (ND10B, Thorlabs) to prevent saturation of the excitation channel. The emission channel passed the fluorescence from ICG with a notch filter (NF785-33 - Ø25 mm Notch, Thorlabs) to block excitation light. Two avalanche photodiodes were used as a photodetector (C12703-01, Hamamatsu Corp, Bridgewater, New Jersey, United States) for capturing the fluorescence-to-excitation ratio simultaneously. APD modules were powered by $\pm 12\mathrm{V}$ (E3630A 35 W, Keysight Tech, Santa Rosa, California, United States). The output of the APD module was fed to a DAQ I/O card to transform the acquired intensity values to a voltage signal, (NI-USB DAQ 6002, $50\mathrm{KS}/\mathrm{s}$ sampling rate, 16-bit resolution, NI, Austin, Texas, United States) measured by LabVIEW software (Version 2022 Q3 (64-bit) 22.3.1f8). Figure 1 illustrates the schematic representation of the prototype system hardware showing individual components, including source/detector pair, optical filters, power supply, and data acquisition.

Fig. 1

Schematic representation of the prototype system hardware showing individual components including lasers, detectors, optical filters, and data acquisition devices. Imaging with a surface imaging system was performed in parallel in the same mice for comparison.

2.2.

Protocol for Animal Studies

Animal studies were conducted in accordance with protocols approved by the University of Wisconsin School of Medicine and Public Health Institutional Animal Care and Use Committee (Protocol M006554). All efforts were made to minimize animal suffering. All surgical procedures and ICG injections were performed under isoflurane anesthesia.

Four ($n=4$) Athymic Nude Mice (cat # 6905M, Envigo, Indianapolis, Indiana, United States) 6 to 8 weeks of age were placed on a low-chlorophyll diet to minimize tissue autofluorescence. Mice were anesthetized using a SomnoSuite Low-Flow Anesthesia system (Kent Scientific, Torrington, Connecticut, United States). A 3% isoflurane concentration was used for induction, and a 2% to 2.5% concentration was used for maintenance, using a flow rate of $500\mathrm{mL}/\min$ and 100% oxygen as the carrier gas. Once anesthetized, each mouse received a 0.1-mL injection containing $1\times {10}^6$ cells into the right flank.

To examine how the signal changes with tumor growth, mice were imaged on various days after implantation. ICG (cat # 099555 Matrix Scientific, Columbia, South Carolina, United States) was suspended in sterile distilled water and diluted to a concentration of $0.7\mathrm{mg}/\mathrm{mL}$. Mice were individually placed in a prone position on a sheet of opaque black plastic with an infrared warming pad underneath, which was connected to and monitored with SomnoSuite’s onboard warming system. A surgical plane of anesthesia was verified via toe pinch, and the hind legs of each mouse were secured to the imaging bed using 3 M Transpore medical tape to avoid excessive movement during imaging.

For ICG imaging, mice were given a retro-orbital injection of $1\mathrm{mg}/\mathrm{kg}$ ICG for fast IV uptake and continuously imaged for 10 min thereafter. The procedure for injecting ICG and measuring with the probe for 10 min was conducted before tumor implantation for each mouse and subsequently on multiple separate days after the tumor reached 3 mm in diameter (once the tumor location became visibly distinguishable in diameter); similarly, this was followed by the next few days after, where the tumor growth size reached $\sim 6\mathrm{mm}$ in diameter. ICG was freshly prepared on the day of each injection, and the same dose was injected each time.

For PpIX imaging, mice were given a retro-orbital injection of $250\mathrm{mg}/\mathrm{kg}$ aminolevulinic acid (ALA) in sterile saline, the pro-drug contrast that develops PpIX over the course of several hours.¹⁷ The imaging was limited to being done with just a few hourly time points as the mice had to be re-anesthetized during each imaging session.

2.3.

Procedure for Tumor Cell Line Development

AsPC-1 pancreatic adenocarcinoma cells were obtained from the American Type Culture Collection (ATCC) and grown in Roswell Park Memorial Institute (RPMI) 1640 Media supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. These were stored in a 5% ${\mathrm{CO}}_2$ incubator. For injection, cells were re-suspended in a 50/50 mixture of PBS and Corning Matrigel (cat # CB-40234A, Fisher Scientific, Hampton, New Hampshire, United States), drawn up into U100 insulin syringes, and placed on ice until injection.

3.1.

Phantom Mimicking and Testing Using ICG and PpIX Fluorophores

To assess the expected signal magnitudes and transient responses during imaging beyond superficial tissues, the fluorescence depth sensing system was first validated for the fluorescence-driven signal to determine the linearity range using different concentrations such as $0\mu \mathrm{M}$ as a control, followed by 0.1, 0.2, 0.3, 0.5, 1, and $2\mu \mathrm{M}$ of ICG and PpIX tissue phantom mimicking solutions. The recipe used for the preparation of the phantom solution includes 1% intralipid (Sigma Aldrich, St. Louis, Missouri, United States) and 0.1% ink (Carbon Black Pigment, Speedball Art Products, Statesville, North Carolina, United States) diluted in deionized water added with different ICG concentrations and, similarly, 2% tween diluted in deionized water added with different PpIX concentrations.¹⁸ The excitation signal specifies a means to standardize variations in laser intensity and tissue variations simultaneously during the imaging of the fluorescence signal. The fluorescence-to-excitation ratio was calculated based on the data points obtained from the fluorescence and excitation channels that displayed a linear response tracking the fluorescence signal. Furthermore, the fluorescence-to-excitation ratio offers equalization against fluctuations in laser or excitation intensity. Herein, the significance of the fluorescence-to-excitation ratio can be traced back to the origins of the born normalization strategy introduced by Ntziachristos and Weissleder in 2001 in context to fluorescence image reconstruction.¹⁹ Furthermore, this study can be utilized to correct the fluctuations that persisted in the input signal due to several different reasons, for instance, laser power fluctuating over time or imperfect fiber contact on tissue surface due to motion artifacts. In tomographic approaches, the F/E ratio is particularly more useful because the sources may be subject to different powers and the detector to different gains, and thus, the measured raw signal at each detector may not be directly comparable if no normalization is performed.

Figure 2 illustrates the different ICG concentration studies ranging from control ($0\mu \mathrm{M}$), 0.1, 0.2, 0.3, 0.5, and $1\mu \mathrm{M}$ to observe the increase in fluorescence signal. Although the fluorescence signal was low, it showed a linear response with incremental concentration levels, as shown in Fig. 2(a). The excitation signal was steady and independent of the fluorophore, as shown in Fig. 2(b). Furthermore, the fluorescence-to-excitation ratio, depicted in Fig. 2(c), showed a linear behavior with the ICG concentration.

Fig. 2

Fluorescence intensity (a), laser excitation intensity (b), and fluorescence-to-excitation ratio (c) were measured in tissue-mimicking phantoms with varying ICG concentrations between 0 and $1\mu \mathrm{M}$. Both the fluorescence and fluorescence-to-excitation ratio showed stable linearity within these concentrations, whereas the excitation intensity was largely independent of the fluorophore concentration.

Similarly, Fig. 3 illustrates the different PpIX concentration studies ranging from control ($0\mu \mathrm{M}$), 0.1, 0.2, 0.3, 0.5, and $1\mu \mathrm{M}$. Figure 3(a) shows the linear fluorescence response. Figure 3(b) represents the excitation signal, which is independent of the fluorophore and appears to be steady. Figure 3(c) shows the fluorescence-to-excitation ratio, showing a linear response as well.

Fig. 3

Fluorescence intensity (a), laser excitation intensity (b), and fluorescence-to-excitation ratio (c) were measured in tissue-mimicking phantoms with varying PPIX concentrations between 0 and $1\mu \mathrm{M}$. Similar to the ICG measurements in Fig. 2, both the fluorescence and fluorescence-to-excitation ratio showed stable linearity within these concentrations, whereas the excitation intensity was largely independent of the fluorophore concentration.

Finally, it can be noted that the F/E ratio strategy implemented here does not necessarily improve the stability of the signal either in Fig. 2 or 3; however, the current sensing system presented in this work is an intermediate step toward a more complex fluorescence tomography system,²⁰ so, as stated above, the F/E approach will definitely have a more important effect. In any case, the most vital results in this study are that (i) the excitation channel is independent of the concentration and (ii) the linear behavior of the fluorescence signal remains linear after normalizing.

3.2.

Fluorescence-Based Tissue Depth Sensing

The sensitivity to depth of the fluorescence depth sensing system introduced in this work was carried out following an experimental setup, as schemed in Fig. 4(a). A cylindrical vial (diameter of 5 mm and length of 10 mm) was filled with a liquid solution of $1\text{-}\mu \mathrm{M}$ ICG in deionized water. This vial was placed inside a plastic container ($10\times 14\times 60{\mathrm{cm}}^3$) with a liquid phantom made of deionized water, 1% Intralipid, and 0.1% India Ink; this proportion aimed at achieving typical living tissues optical properties [i.e., absorption coefficient ${\mu }_{aM}=0.01{\mathrm{mm}}^{-1}$ and reduced scattering coefficient ${\mu }_{s^{\prime }}=1{\mathrm{mm}}^{-1}$ (Ref. 21)]. The vial was suspended at 10 mm above the bottom surface of the container using a thin white thread; this method helped place the vial in a fixed position without floating, sinking, or displacing it in the lateral directions.²² The depth, $d$, of the vial was modified from 4.5 to 20.5 mm by adding liquid on top of it, and the fluorescence and excitation signals were measured with the source and detector fibers in contact with the free surface of the liquid phantom; in addition, the detector fiber was positioned on one side of the vial and the detector fiber on the opposite side, being the source–detector distances used in this study $\rho =5\mathrm{mm}$ and $\rho =20\mathrm{mm}$. Finally, the system was isolated from ambient light by placing it inside a sealed black plastic box.

Fig. 4

(a) Scheme of the depth analysis experimental setup using a spacing $\rho$ between the laser source (Src) and the detector (Det), which was measured using a tube of ICG submerged in an intralipid-based solution. (b) The log of the fluorescence intensity was plotted as a function of the inclusion’s depth for $\rho =5\mathrm{mm}$ (blue) and $\rho =20\mathrm{mm}$ (red).

From Fig. 4(b), it can be seen that both sets of measurements ($\rho =5\mathrm{mm}$, blue; and $\rho =20\mathrm{mm}$, red) show a decrease in the fluorescence signal with the vial’s depth; naturally, the shortest source–detector distance measured higher intensities than the largest source–detector distance for the full depth range (with differences of more than one order of magnitude). The experimental data for $\rho =20\mathrm{mm}$ present an abrupt change of slope at depth values between 8.5 and 10.5 mm; this may indicate that the photon paths begin to intercept the inclusion at this depth. This behavior is indicative of a change in optical properties sensed by the fluorescence depth sensing system.

A more thorough analysis can be performed by fitting the data shown in Fig. 4(b) with an appropriate theory describing fluorescence in turbid media. To this end, we used the analytical model introduced by Patterson and Pogue²⁴ for the continuous-wave (CW) fluorescence signal [equation (13) of the referenced paper]:

With the help of these two assumptions, it is possible to derive a fairly simple analytical expression to fit the data measured shown in Fig. 4(b). First, we can explicitly write ${\mu }_{aM}^x={\mu }_{aM}^x$ (${\lambda }_x$) and ${\mu }_{aM}^m={\mu }_{aM}^m({\lambda }_m)$. As the fluorescence process requires that ${\lambda }_m>{\lambda }_x$, ${\mu }_{aM}^m$ can be rewritten as ${\mu }_{aM}^m={\mu }_{aM}^x+\mathrm{\Delta }{\mu }_{aM}$, where the sign of $\mathrm{\Delta }{\mu }_{aM}$ depends on the absorption spectrum of the fluorophore in the range $[{\lambda }_x,{\lambda }_m]$. With all this in mind, we can take the limit of Eq. (1) for $\mathrm{\Delta }{\mu }_{aM}\to 0$:

Figure 5 summarizes the results from the fitting process. Panel (a) shows the retrieved effective attenuation coefficient, whereas panel (b) shows the retrieved amplitude factor, both as a function of the vial’s depth. The behavior of $\kappa$ indicates that the combination of source–detector separation used in this study maximizes the sensitivity of the system at a depth of inclusion equal to 18.5 mm. To see this, we could imagine a scenario where there is no ICG in the vial, in which case $\kappa$ should be independent of its depth, with the consequence that the fitted values should present a completely flat and horizontal behavior around $\kappa \sim 0.12{\mathrm{mm}}^{-1}$; the fact that this value changes with the depth of the vial makes us believe that the system is detecting its presence as it goes deeper and deeper into the medium (as least up to a depth of 18.5 mm). This trend can be explained as follows: although the excitation photons (i.e., photons coming from the source at ${\lambda }_x$) do not encounter the object in their trajectory from source to detector, the retrieved effective attenuation coefficient resembles the one from the surrounding liquid phantom, i.e., $\kappa ={(3{\mu }_{aM}{\mu }_{s^{\prime }})}^{-1/2}$; on the other hand, when the excitation photons begin to sense the vial (at a depth between 8.5 and 10.5 mm), more emission photons are produced, and hence, $\kappa$ increases because the absorption coefficient tends to be the sum of ${\mu }_{aM}$ and ${\mu }_{af}$.^24,25 Beyond 18.5 mm, the value of $\kappa$ significantly decreases, although not to the baseline level, suggesting that deeper inclusions could still be detectable by the measurement system. Regarding the baseline level, the chosen optical properties for the host medium (${\mu }_{aM}=0.01{\mathrm{mm}}^{-1}$ and ${\mu }_{s^{\prime }}=1{\mathrm{mm}}^{-1}$) should give a value of $\kappa \sim 0.17{\mathrm{mm}}^{-1}$. As it can be seen from the first retrieved value in Fig. 5(a), this is not the case; this difference could be attributed to errors during the phantom preparation or to an inaccurate choice of the model from Ref. 21.

Fig. 5

Results from the fitting process using the model given by Eq. (1): (a) effective attenuation coefficient, $\kappa$; (b) amplitude factor, $C$.

The exponential decay of the amplitude factor $C$ in Fig. 5(b) is probably due to the fact that this parameter is a function of the product between ${\mu }_{af}$ and $\eta$ [as expressed by Eq. (3)]. Hence, as the object moves further away from the source and detector, the liquid phantom without ICG dampens the absorption and the quantum yield of the ICG solution in the vial, with the consequent impact on $C$ just discussed. Here, we must stress that we are using a homogeneous model to test a heterogeneous medium, so what the model actually retrieves is an “effective” absorption coefficient of the fluorophore together with an “effective” quantum yield, which will vary according to the position of the object. This claim is further supported by the form of $C$ given in expression Eq. (3) derived above.

According to these results, the sensitivity of the system could be possibly extended to depths larger than 20 mm; nevertheless, a more thorough study must be performed, for example, by varying the optical properties of the host phantom and the size of the inclusion, or by refining the model used to fit the experimental data.

4.1.

Tracking ICG Kinetics on Different Days with Increasing Tumor Size

During this study, fluorescence depth sensing was used to image through the skin and into the tumor. The ICG kinetic curve for normal tissue had a very rapid uptake and fast clearance with a bi-exponential clearance curve, whereas the tumor was expected to be slower in both. This was imaged for successive days after tumor implantation, including days 0, 10, 15, and 21 with gradual growth of tumor size measured at 0, 3, 4, and 5 mm, respectively, on these days. For all of the in vivo imaging experiments here, the source–detector pair was positioned 5 mm apart located gently on the surface of both the legs of the mouse, and the background light was suppressed by a black covering. Furthermore, the obtained data points were normalized to have the tail regions match both tumor and normal tissues. Figure 7 illustrates the ICG kinetics sampled on four mice on different days with increasing growth of tumor size wherein all the peak values were normalized to 1 for better comparison, analysis, and processing. From this, it is clear that the tumor tissue uptake and clearance are slow in the kinetics curve and vary based on the tumor tissue sampled.²⁵

Fig. 7

Fluorescence depth sensing system was used to track ICG kinetics for 10 min in each mouse at multiple time points, which corresponded to different tumor sizes. Analysis showed that the system is able to resolve kinetic differences as a tumor becomes larger, with larger tumors showing delayed uptake and longer retention of ICG or the system autogain.

4.2.

Fluorescence Depth Sensing Versus Surface Imaging System for ICG

The radiometric target served as a reference to compensate for the EleVision system’s autogain function so that we could normalize the ICG curves. Normalizing to an ICG-equivalent reference target has been previously validated.²⁶ These normalized ICG curves of both the tumor tissue and the normal tissue were compared with the curves obtained by the depth sensing approach to compare the ability of the systems to observe kinetic differences in the ICG kinetics between the tissue types. Usually, surface imaging is highly weighted to the most superficial tissues, i.e., skin layer, and cannot penetrate deep within the tissue and provide much information beyond the superficial tissues. Thus, this is one of the major drawbacks of the surface imaging system in the field of optical imaging and restricts it from being used for high penetration depth studies, especially beneath superficial tissue. Figure 8 shows the ICG kinetics for tumor and normal tissues imaged on the surface of the mice using (a) a depth sensing system and (b) a surface imaging system.

Fig. 8

Comparison of ICG kinetics in normal and tumor tissues sampled on the (a) fluorescence depth sensing system and (b) surface imaging system. The deeper imaging of the fluorescence depth sensing system allows it to better resolve kinetics differences between the tissue types.

4.3.

Analysis of ICG Kinetics

Based on the measured slopes of the ICG signal with fluorescence-to-excitation intensity ratio, kinetic modeling simulations were conducted to promote the hypothesis that feature tumor pathophysiology, such as capillary or vascular leakage, could be responsible for the observed differences across tumor types. Fitted uptake and clearance kinetics show that tumors are slower, both in uptake and clearance. Furthermore, the obtained data points from all the investigated days, both tumor and normal tissues, were taken into consideration, and the data were interpreted to fit the modeling theory of ICG kinetics. The ICG kinetics included two exponential curves, including the rising and falling parts over time for pre-peak and post-peak, respectively, fitting to exponential-type equations for separate parts of the rising and falling parts of the kinetics. It was noticed that the falling off curve, after the peak, was a bi-exponential functional form, fitted using Eq. (4)

Table 1

Curve fit parameters for rising and falling parts of the ICG kinetics.

At this point, it must be noted that we lack a physiological model to compare with the experimental data, and hence, no physiological parameters are retrieved in this fitting process. Instead, a semi-quantitative analysis was performed to notice the fast uptake and clearance of the kinetics curves, which easily differentiates between tissue types. This is a common practice, and although quantitative physiological parameters cannot be studied, a qualitative behavior can be inferred by analyzing the exponential constants.²⁸

4.4.

Fluorescence Depth Sensing Versus Wide Field Imager System for PpIX

Herein, a comparison was made for PpIX imaging using a laboratory setup, which utilizes a 650-nm long pass-filtered color camera for capturing PpIX emission. The wide-field imager collects at an $8\times 8\mathrm{cm}$ field of view (FOV) covering the body of the mouse. Excitation of PpIX is accomplished through a 635-nm LED with a power density of $5\mathrm{mW}/{\mathrm{cm}}^2$ in the FOV. PpIX prompt fluorescence is measured at an overall integration time of 50 s. Acquisition parameters are kept the same across the measurements.

After ALA injection, four mice were imaged with both a fluorescence depth sensing system and a wide-field imager. Mice were imaged at time points of 1-, 3-, and 6-h post-5-ALA injection, with limited sampling due to the need for re-anesthesia each time. For control, images and measurements were also acquired before 5-ALA administration and right after 5-ALA administration.

For measurements, mice were under isofluorane-based anesthesia, and measurements were acquired first with the wide field imager and subsequently with the fluorescence depth sensing system for both tumor and non-tumor thighs. For quantification of the wide-field images, ROIs are selected at both the tumor and opposite leg regions of the mouse, and mean and standard deviation are calculated per each time point. Example fluorescence images acquired at the different time points for one mouse are displayed in Fig. 9 in correlation to the fluorescence depth sensing system measurement quantification.

Fig. 9

Both wide-field imager (WFI) and fluorescence depth sensing (FDS) systems were used to measure PpIX fluorescence in both tumor and normal tissues at multiple time points after 5-ALA administration. The measurements taken by the fiber optic-based FDS system correlated with the images of the wide-field imager, which showed stronger fluorescence in tumor signal in the later time points and fluorescence peaking at 1-h post-injection.

The results shown in Fig. 10(a) describe an analysis obtained through fluorescence depth sensing, and Fig. 10(b) displays images acquired with the PpIX wide-field imager. When calculating the mean and standard deviation of the group, as measured with the fluorescence depth sensing system and wide-field imager [Figs. 9(a) and 9(b)], it can be observed that the overall trend correlates with the acquired fluorescence intensity. However, an exact matching of results is not expected as the fluorescence depth sensing system samples a smaller region of interest in comparison to the wide-field imager, where the ROIs cover the whole tumor and a similar area on the opposite side of the mouse. In comparison to ICG, the kinetics of PpIX are much slower with both systems as described as a significant change was not observed within a single time point. Notably, the detected PpIX with both systems can be a combination of both PpIX skin fluorescence and PpIX accumulation in the tumor.

Fig. 10

Mean PpIX fluorescence for both tumor (mean $T$) and normal (mean $N$) tissues measured with (a) the fluorescence depth sensing (FDS) and (b) the wide field imager (WFI) system both before 5-ALA injection and up to 6 h after injection.