2.1.

Paired-Agent Model

The development and validation of the paired-agent approaches for quantifying concentrations of targeted biomolecules in tissue have been extensively demonstrated in previous publications for both systemically delivered imaging agents^25–33 and for direct administration (topical) applications.^34–38 Briefly, paired-agent approaches require coadministration of a control, “untargeted” imaging agent with the targeted agent(s), signal from which is used to account for variability in targeted imaging agent delivery and nonspecific retention, such that the concentration of biomolecules targeted can be estimated. Based on compartment model approximation of imaging agent kinetics in tissues, the simplest method of estimating biomolecule concentration is through the following equation:

Fig. 1

Paired-agent model for quantification of binding in CNs. (a) Illustration of the distribution of targeted and control imaging agents in tissue. It is presumed that the control agent (green) is dispersed throughout one compartment, the “free” space; and the targeted agent (red) is dispersed throughout two compartments: the “free” space and the “bound” space. It is also assumed that both imaging agents can have nonspecific binding. (b) Illustration of the estimation of BP based on targeted and control imaging agent uptake. If it is assumed that the concentration of control agent in the region of interest is similar to the concentration of unbound targeted agent, and that the adiabatic approximation holds, BP can be approximated, which is directly proportional to receptor concentration. MBP: myelin basic protein, BP: binding potential.

2.2.

Animal Experiments

Animal studies were performed in accordance with approved institutional protocols (IACUC #2013-008). Fifteen male Fischer rats weighing 200 to 250 g were obtained from Envigo (Indianapolis, Indiana) and separated into three groups. Two targeted agents were used in this study, Oxazine 4 and Rhodamine 800, both of which have demonstrated an affinity for myelin basic protein.¹ In the first two groups, Oxazine 4 (Exciton, Dayton, Ohio) was used as the targeted imaging agent. Direct application of the fluorophore was conducted in the first group ($n=5$), whereas intravenous administration was done in the second ($n=5$). For the third group, the targeted agent, Rhodamine 800 (Sigma-Aldrich, St. Louis, Missouri), was directly applied to the specimen ($n=5$). In all cases, indocyanine green (ICG) (Chem-Impex International, Wood Dale, Illinois) served as the control imaging agent. Targeted and control imaging agents were dissolved in solutions containing 10% dimethyl sulfoxide (Amresco, Solon, Ohio), 5% Kolliphor EL (Sigma-Aldrich, Saint Louis, Missouri), 65% fetal bovine serum (FBS) (GE Healthcare Life Science, Logan, Utah), and 20% HEPES buffer (ThermoFisher Scientific, Waltham, Massachusetts), at 2 mM and $2\mu \mathrm{M}$ concentrations, respectively. ICG is known to bind with high affinity to serum proteins.^40,41 By adding FBS to the solution, the ICG was prebound in solution to minimize nonspecific binding in tissue.

For direct application studies [Fig. 2(a)], rats were sacrificed and brains dissected, keeping as many CNs intact as possible. Control images were taken to account for background fluorescence. Fresh brain tissues were stained ex vivo for 10 min with $500\mu \mathrm{L}$ of the fluorescent cocktail. This was followed by three rinse steps of phosphate-buffered saline (PBS): three flushes per rinse. Each flush consisted of pipetting 1 mL of PBS onto the brain surface, and each rinse was completed with the removal of any residual PBS. Fluorescence images were then acquired.

Fig. 2

Experimental procedure for paired-agent imaging of CNs. (a) Using the direct method of imaging agent application, staining was done ex vivo on brains from sacrificed rats. After obtaining background images in the targeted and control fluorescence channels, as well as white light, a mixture of targeted and control imaging agents was applied to the brains. This was allowed to stain for 10 min, followed by three rinse steps. Fluorescence and white light imaging was then repeated. (b) For intravenous administration, the mixed solution of imaging agents was injected via the tail or penile vein. Four-h postinjection, brains were dissected from sacrificed rats, and imaging was done ex vivo. (c) Illustration of the ventral view of a rat brain. CNs used for analysis were the olfactory tract (CN I), optic nerve (CN II), and trigeminal nerve (CN V).

Nerve staining via systemic administration was achieved via tail or penile vein injection [Fig. 2(b)]. Rats were given the fluorescent cocktail at a dose of $1\mu \mathrm{mol}/250\mathrm{g}$. Previous studies using Oxazine 4 to highlight nerves demonstrated optimal contrast 4 h after a single intravenous injection;¹ thus, 4-h postinjection, the rats were euthanized, brains dissected, and imaged.

2.3.

Fluorescence Imaging Systems

A custom-built multiwavelength fluorescence imaging system was used to acquire Oxazine 4, as well as ICG images, to allow for application of the paired-agent approach. The system consisted of a WheeLED wavelength-switchable light source (Mightex Systems, Pleasanton, California). Emitted light was passed through a filter wheel with bandpass filters designed to match each light channel. The light was then split into two liquid light guides to provide even illumination of the surgical field from two sides. A Basler ace NIR monochrome camera (Basler, Ahrensburg, Germany) with an 8-mm fixed focal length lens (Navitar, Rochester, New York) and appropriate emission filters were used to collect fluorescence images. All filters were obtained from Omega Optical (Brattleboro, Vermont) unless specified otherwise. Oxazine 4 fluorescence was acquired using a 617 nm LED filtered with a $615\pm 5\mathrm{nm}$ bandpass excitation filter and a $655\pm 25\mathrm{nm}$ bandpass emission filter. For ICG, a 780 nm light was filtered with a $785\pm 12.5\mathrm{nm}$ bandpass excitation filter (Chroma Technology, Bellow Falls, Vermont), and the resulting emission fluorescence was collected with an 825 nm longpass filter (Edmund Optics, Barrington, New Jersey). White light images were acquired unfiltered. Total imaging time to acquire a targeted fluorescence, control fluorescence, and white light image was 50 s; however, this was just a research-based system used for proof-of-concept, and a more advanced system could collect all images in video-rate if necessary.⁴²

A Pearl Trilogy Imaging System (LI-COR Biosciences, Lincoln, Nebraska) was used to take white light, Rhodamine 800, and ICG fluorescence images. Acquisition of all three images took 38 s. Rhodamine fluorescence was collected in the 700 nm channel, whereas ICG fluorescence was collected in the 800 nm channel.

2.4.

Image Analysis

In order to compare performance of the different experimental groups, nerve contrast was quantified in two ways. First, nerve-to-brain ratio (NBR) was simply calculated as a signal-to-background ratio: average signal in a nerve region over the average signal in a brain region of similar size ($\mathrm{NBR}=S_n/S_b$). The second measure used was contrast-to-noise ratio (CNR), which can be expressed as follows:

For qualitative analysis, and to simulate the potential utility of nerve-specific imaging agents in conjunction with a control agent to highlight CNs, raw fluorescence data were thresholded and overlaid onto white light images. A logistic function⁴³ was used to determine the optimum lower threshold that maximized nerve contrast for each individual image. A univariate color map where transparency increased as the lower threshold approached was applied to replicate how fluorescence might actually be visualized with overlay on a white light image for the surgeon. Furthermore, overlay of the white light image would provide surgeons with anatomical context, and the transparent to opaque color map avoids changing the native FOV.

2.5.

Simulation

To test the utility of NBR and CNR as metrics for image quality, theoretical nerve images with different signals were simulated, and ratios were calculated using the equations above. Both images had a large and small nerve contained within a background. It was assumed that the theoretical images were acquired with a camera that was optimally exposed, and the first image had amplified signal compared to the second due to increased gain. Since image sensors measure signal as the number of photons detected over a given time, photon counting follows Poisson statistics with signal-dependent noise. Thus, Poisson noise was independently added to the signals (in nerve, ranging from 40% to 100% of the detector dynamic range; and in background, ranging from 7% to 33% of the detector dynamic range) from each simulated image using the function poissrnd() in MATLAB, at $1\times$ and $2\times$ simulated gain settings. On average, the second image had $\sim 7$ times more noise than the first. Since the truth in classifying nerve and background was known, sensitivity and specificity were calculated for both images using 50 different thresholds ranging from the minimum image intensity to the maximum. Sensitivity was calculated as the true positive rate (proportion of times nerves were correctly identified as nerves), and specificity as the true negative rate (proportion of times the background was correctly identified as such). A receiver-operating characteristic (ROC) curve was constructed by plotting the true positive rate (sensitivity) versus the false positive rate (1—specificity), and the area under the curve (AUC) was calculated. Corresponding NBR and CNR values were plotted against AUC values to compare ratios and their probability to accurately identify a pixel as nerve or background.

2.6.

Statistical Analysis

All statistical analyses were performed using SPSS (IBM^®, Armonk, New York). A paired-samples $t$-test was used to compare contrast between single-agent and paired-agent imaging methods within each experimental group. Significance among the groups was evaluated with independent samples $t$-test and a Bonferroni correction to account for multiple comparisons. Analyses were done for each of the three CNs separately, and then averaged together. The level of statistical significance was set at $p<0.05$. Results are presented as $\text{mean}\pm \text{standard deviation}$ unless otherwise noted.

3.1.

Ex Vivo Nerve-Specific Fluorophore Cranial Nerve Staining in Rats

Direct application of Oxazine 4 and Rhodamine 800 onto freshly excised rat brains was evaluated for nerve-specific fluorescence of CNs. Figure 3(a) shows representative images of control and targeted fluorescence uptake, as well as BP maps after applying the paired-agent model. Qualitatively, as expected, the control agent demonstrated similar uptake in both groups where fluorescence can be seen across the surface of the brain. Targeted fluorescence images of Oxazine 4 compared to Rhodamine 800, however, showed greater uptake in each of the nerve regions than in adjacent brain tissue located within the midbrain [Fig. 3(b)]. Direct application of Rhodamine 800 resulted in similar signal within CN I ($0.12\pm 0.15$) and the brain tissue ($0.13\pm 0.15$). Paired- and independent-samples $t$-tests indicated that there was no significant difference between signal from each nerve and the brain tissue, nor among test groups. Thus, neither of the agents could be quantitatively identified as the ideal candidate alone.

Fig. 3

(a) False-colored raw fluorescence images of ex vivo rat brains from the ventral view. Each row is representative of independent experiments for different experimental groups. First row: Oxazine 4 direct, second row: Oxazine 4 systemic, third row: Rhodamine 800 direct ($n=5$ in each group). The first column presents residual fluorescence of the control agent, ICG, across the brain and nerve tissue. The second column displays targeted fluorescence uptake (either Oxazine 4 or Rhodamine 800) in the same brains. The third column illustrates BP maps, resulting from application of the paired-agent model. The fourth column shows thresholded BP maps overlaid onto white light images of the same brains. For each image of the last column, the maximum value was set to the maximum intensity of the color scale, and the minimum value was independently chosen as the optimum lower threshold to maximize nerve contrast. Units of fluorescence are arbitrary, and BP is a unitless value. Images from the Oxazine 4 groups were acquired using a custom-built imaging system (FOV: $14.4\mathrm{cm}\times 14.4\mathrm{cm}$), where ICG and Oxazine 4 images were collected at 785 and 655 nm, respectively. In the third experimental group, using Rhodamine 800, control images were taken in the 800 nm channel, whereas targeted images were captured in the 700 nm channel using a pearl trilogy imaging system (FOV: $11.2\mathrm{cm}\times 8.4\mathrm{cm}$). (b) Targeted fluorescence signal in nerve regions and a region of brain tissue located adjacent to nerves within the midbrain. (c) NBR and (d) CNR versus experimental group for both single- and paired-agent methods. Values are presented as the average of all three nerves evaluated. Results from all bar plots are the mean + standard deviation.

To better quantify the data, targeted fluorescence intensities from each region of interest were used to calculate NBR. Ratios were taken as the average among all nerve regions. Between groups, there was no significant difference when evaluating NBR with a single targeted agent, however, use of the paired-agent approach resulted in significant increases ($p<0.005$) for both Oxazine 4 and Rhodamine 800 [Fig. 3(c)]. From inspection of their respective BP maps [Fig. 3(a)], nerves were much easier to locate with the application of paired-agent imaging using Oxazine 4 (NBR: $174.8\pm 103.9$) as the targeted agent, as opposed to Rhodamine 800 (NBR: $71.1\pm 88.3$) as the targeted agent. Contrast quantified by NBR was improved $61.2\pm 31.0$ and $50.7\pm 62.6$ times for Oxazine 4 direct and Rhodamine 800 direct groups, respectively [Fig. 3(c)].

Strong signal seen along the brainstem can be attributed to the presence of CNs as well. However, since it was difficult to preserve all of them consistently during the brain dissections, or track locations of specific CNs, they were not analyzed in this study.

3.2.

In Vivo Oxazine 4 Cranial Nerve Staining in Rats

In order to compare nerve specificity between direct application and systemic administration, in vivo nerve staining using an intravenously injected Oxazine 4/ICG mixture was conducted. The second row of Fig. 3 displays a representative image of control and targeted fluorescence, in addition to BP maps, with rats euthanized 4 h after intravenous injection, prior to brain excision. Although there appeared to be some delineation of nerves in the Oxazine 4-targeted fluorescence image, there was also strong signal throughout the background (NBR: $1.4\pm 0.1$), resulting in no statistically significant difference between the two regions [Figs. 3(a) and 3(b)]. Similar to the ex vivo results presented above, nerves were easier to identify relative to the brain when using the BP map images compared to the control or targeted fluorescent images alone. On average, contrast was improved $30.6\pm 10.2$ times. Previous studies investigating the uptake of Oxazine 4 in the central nervous system of mice demonstrated no uptake in the brain;¹ however, the same results were not obtained in this work, with brain signal being $3.0\pm 2.2$ times higher than autofluorescence. Nonetheless, application of the paired-agent approach accounted for much of the nonspecific uptake. Little to no signal was observed using the control agent likely because of the large size of ICG molecules upon binding to plasma proteins^40,41 that prohibit it from crossing the blood-brain barrier (BBB) and blood-nerve barrier (BNB). Targeted agents were not affected by the BBB and BNB because they do not bind to large plasma proteins, and are inherently smaller molecules (Rhodamine 800: $495.95\mathrm{g}/\mathrm{mol}$, Oxazine 4: $395.84\mathrm{g}/\mathrm{mol}$) compared to ICG ($774.96\mathrm{g}/\mathrm{mol}$), the latter of which binds to proteins in blood (typically albumin; 66.5 kDa) that are over 100 times larger than Rhodamine 800 and Oxazine 4. In the case of direct application where ICG uptake could be visualized, size of the molecule was not a problem because the agent did not have to exit the vasculature.

Compared to the direct method of administration, nerve-specific fluorescence signal from intravenous administration of Oxazine 4 was not as strong. Again, although the intensity of nerve signal was visually brighter in the targeted image of the systemic group, the rest of the brain was also correspondingly bright. The difference in contrast was more obvious when comparing the BP map images, where BP in nerve regions is stronger in the Oxazine 4 direct group.

3.3.

Quantification of Fluorescence Signal Using Contrast-to-Noise Ratio

The CNR values provided significant differences between Oxazine 4 direct, and both Oxazine 4 systemic and Rhodamine 800 groups when using targeted fluorescence alone ($p<0.005$). If paired-agent imaging was used, significance among the groups decreased ($p<0.05$). Figure 5 illustrates this in more detail. Within groups, noteworthy increases in CNR were observed when comparing single-agent versus paired-agent imaging. The CNR improved $2.1\pm 0.7$ times when applying Oxazine 4 directly; there was a $5.2\pm 1.5$ time increase for intravenous administration of Oxazine 4; and CNR using direct application of Rhodamine 800 increased by a factor of $4.0\pm 2.9$ [Fig. 3(d)].

Fig. 4

Comparison of metrics to assess contrast from simulation. (a, b) Theoretical image of background tissue and two different-sized nerves with varied levels of noise. (c) Comparison of NBR and CNR against AUC of the ROC for the simulated situation of image (a).

Both NBR and CNR support the enhanced delineation of nerves in BP map images using the paired-agent approach as opposed to a single targeted agent. The relationships correlate well with what is depicted in the last three columns of Fig. 3(a).

However, it should be noted that the values obtained for NBR and CNR have rather drastic differences despite being measures of contrast for the same images. The NBR calculations reported values as high as 278, whereas the maximum value returned for CNR was 74. Both ratios correspond to paired-agent imaging with Oxazine 4 applied directly; thus, such stark differences demonstrate the challenge of what metric best represents an image.

3.4.

Simulation to Demonstrate the Utility of Contrast-to-Noise Ratio as a Metric for Image Quality

Theoretical images of a large nerve and smaller branch with varying levels of noise in the background are presented in Figs. 4(a) and 4(b). In the first image [Fig. 4(a)], both the large and smaller nerves were easily identified, with a NBR of 3. On the other hand, the same image with a higher NBR of 7.6 had a less distinguishable smaller nerve because of a noise level $\sim 7$ times higher [Fig. 4(b)]. Despite the suggested increase in contrast, resolution was clearly diminished in the second situation, where the smaller nerve was no longer clear and got lost in the background. This demonstrated the danger of relying on signal-to-background ratios as a quantitative measure of image quality because they do not account for noise. Alternatively, the CNR value was 3 times greater for the first image compared to the second. This difference more accurately reflected what was seen qualitatively, therefore suggesting that CNR is more informative as a single metric of image quality. This concept was further verified by plotting NBR and CNR versus AUC of the ROC [Fig. 4(c)]. For the simulated images, sensitivity and specificity were calculated for different thresholds, and an ROC curve was constructed. The plot illustrates wide variability of NBR values for single AUC values, whereas data points for CNR demonstrated a stronger correlation with AUC values and a consistent correlation approaching that of an ideal observer. This simulation further supports CNR as a more accurate quantitative representation of image quality in fluorescence-guided surgery compared to NBR or targeted fluorescence alone.

Fig. 5

Comparison of CNR versus targeted agent type and method of application between single- and paired-agent imaging for (a) the average of all nerves analyzed and (b) CN I, (c) CN II, and (d) CN V, independently. Each data point represents nerve CNR from one mouse. *$p<0.05$, **$p<0.01$, ***$p<0.005$.

3.5.

Direct Application of Oxazine 4 Improves Contrast-to-Noise Ratio

As discussed, CNR provided more meaningful information on image quality, and so a more in-depth analysis of the various groups was carried out using this metric. Figure 5 presents box plots of both single- and paired-agent nerve CNR for all three experimental groups. On average [Fig. 5(a)], nerve CNR was significantly greater in paired-agent imaging for Oxazine 4 direct and Oxazine 4 systemic groups ($p<0.05$ and $p<0.005$, respectively). The Rhodamine 800 direct group also showed an increase, although not statistically significant ($p>0.1$, NS). A similar pattern was revealed when looking at each nerve individually [Figs. 5(b)–5(d)], except for CNR of CN V for Oxazine 4 direct where there was no significant difference. The group that appeared to benefit most considerably from the paired-agent approach was the Oxazine 4 systemic group ($p<0.005$ for average of all nerves). It demonstrated increases by factors of $7.4\pm 4.5$, $5.2\pm 2.5$, and $5.6\pm 2.3$ for CN I, CN II, and CN V, respectively. The lack of significance in the Rhodamine 800 group can be attributed to the outliers. Nonetheless, these results support the benefits of paired-agent imaging.

Between groups, the box plots show rather consistent increases in CNR between the Oxazine 4 direct and Oxazine 4 systemic group, and Oxazine 4 direct and Rhodamine 800 direct group. This was evident for both single- and paired-agent imaging for the average of all nerves ($p<0.05$). In single-agent CNR [Fig. 5(a), red dots], Oxazine 4 direct showed a $6.2\pm 2.6$ time increase over Oxazine 4 systemic, and a $4.8\pm 1.7$ time increase over Rhodamine 800 direct. Similarly, paired-agent CNR [Fig. 5(b), blue dots] for all nerves also displayed improvements with a $2.7\pm 1.8$ time increase in Oxazine 4 direct over systemic, and a $3.3\pm 2.1$ time increase over Rhodamine 800.