2.1.

QD and Fluorophore Preparation

(CdSe)ZnS QDs emitting at peak wavelengths of 500, 550, 600, and 650 nm and dispersed in chloroform were prepared using a well-established organometallic procedure.¹¹ Qdot 705, AlexaFluor488, AlexaFluor532, AlexaFluor568, AlexaFluor633 and AlexaFluor680 were used as purchased (Q21361MP A-20000 A-20001 A-20003 A-20005 A-20008, Invitrogen, Carlsbad, California). QD excitation and emission fluorescence spectra (Fig. 1), and absorption spectra were acquired using a scanning spectrofluorometer (Fluorolog 3, Horiba Jobin-Yvon, Edison, New Jersey, USA) and a UV-Vis spectrometer (Cary 300, Varian, Inc. Palo Alto, California, USA), respectively. The quantum yield of each was calculated as the total emitted fluorescence divided by the absorption coefficient at 385 nm for the QDs and at 488, 532, 568, 633, and 680 for the organic dyes. The values were then normalized to the known (0.92) quantum yield of AlexaFluor488.

Fig. 1

Fluorescence excitation (thick) and emission (thin lines) spectra for (a) the QDs and (b) the corresponding AlexaFluor dyes. The concentrations were adjusted to match the absorption spectrum at 385 nm for the QDs and the peak absorption of the dyes (488, 532, 568, 633, 680). The calculated quantum yields were 0.16, 0.75, 0.58, 0.24, 0.39 for the QDs and 0.92, 0.52, 0.77, 0.53, 0.17 for the dyes.

The concentrations were determined from the absorption spectra, following a method developed by Yu et al.¹² for the QDs and using Beer’s law (with molar extinction coefficients published by Invitrogen) for the dyes.

2.2.

Tissue Preparation

The previous study used porcine liver tissue for proof-of-principle validation of the contrast optimization approach.¹⁰ Here, in order to validate the method further using tissues with a range of optical properties, fresh (not previously frozen) porcine kidney and liver, and bovine brain and lung were used. Each tissue was homogenized and diluted following the previous protocol.⁸ All bulk tissues were processed and imaged within 12 h of acquisition.

In addition, to add clinical relevance, four different hollow organs—bladder, colon, esophagus, and stomach—were included. These were excised immediately post mortem from a healthy, 70 kg female Yorkshire Cross pig. For logistics reasons, these were collected some weeks before the imaging experiments and stored at $-70°\mathrm{C}$. Prior to imaging, the organs were gently cleaned with saline and a $1{\mathrm{cm}}^2$ flap of mucosa plus submucosa was surgically separated from the underlying muscle layer. This allowed both surface and subsurface fluorescence imaging within the same region of interest, as depicted in Fig. 2. The thickness of the $\text{mucosa}+\text{submucosa}$ was estimated mechanically and optically for each tissue using a Vernier caliper and confocal microscopy as $700\pm 50$, $400\pm 50$, $800\pm 100$, $2000\pm 200\mathrm{\mu m}$ for the bladder, colon, esophagus, and stomach, respectively. These values are in reasonable agreement with the literature,^13–16 considering that the thickness of the colon depends on the degree of stretching.

Fig. 2

Schematic of a hollow organ after separation of the mucosa from the underlying muscle or cartilage layer. For subsurface imaging, the capillary target is placed on the muscle layer and the mucosa and submucosa are folded back. For surface imaging the capillary is placed on top of the mucosa.

To avoid degradation, all tissues were refrigerated between manipulations and intact tissues were regularly re-hydrated with isotonic saline during the imaging procedure.

2.3.

Tissue Optical Property Measurements

Tissue optical properties were acquired using a 1 mm diameter diffuse reflectance fiber-optic spectrometry probe that has an accuracy of $\pm 10\%$ in this spectral range.¹⁷ The diffuse reflectance at 3 fiber source-detector separations were processed using an inverse Monte Carlo algorithm, combined with a priori knowledge of the spectral signature of the main absorbers, namely oxy- and deoxyhemoglobin and a simple inverse dependence for the scattering. The measured absorption and reduced scattering spectra for the 100% homogenates and hollow organs are shown in Fig. 3. The uncertainty on these measurements, including both the probe’s precision and tissue spatial variations, is approximately $\pm 15\%$ and $\pm 20\%$ for the homogenized and intact tissues, respectively. Note that the pigment packaging effect²⁴ was partially taken into account by feeding the diffuse reflectance model with measurements of oxygenated and de-oxygenated whole blood absorption spectra rather than hemoglobin spectra from the literature. However, better optical property estimates could likely be obtained by including the pigment packaging effect to our model.

Fig. 3

(a) Absorption (thick) and reduced scattering (thin lines) spectra measured for the 100% homogenized bulk organs, and (b) for the intact hollow organs (measured from the mucosal side).

2.4.

Fluorescence Imaging and Image Analysis

Fluorescence imaging was performed using a custom-made multi-spectral imaging system consisting of an epifluorescence stereomicroscope (MZFLIII, Leica Microsystems, Richmond Hill, Ontario, Canada), cooled CCD camera (CoolSNAP K4, Photometrics, Tucson, Arizona, USA) and automated excitation and emission filter wheels (AB304-T, Spectral Products, Putnam, Connecticut), as previously described in detail.¹⁰ All measurements were done under high signal-to-noise conditions, using 12 excitation filters ranging from 385 to 620 nm (average bandwidth 20 nm) and five emission filters corresponding to the QD emission peaks at 500, 550, 600, 650, and 700 nm (bandwidth 50 nm). Images were corrected for the camera noise, excitation lamp spectrum, exposure time and spatial illumination profile, and analysed as previously described.¹⁰ The measured image contrast, or target-to-background ratio (TBR), is defined as the ratio of fluorescence signals emitted by the target (in this case the QD-filled capillary) and background regions of interest:

3.1.

Threshold Concentration ($c_{\mathrm{th}}$)

We showed previously¹⁰ that the TBR may be expressed as follows:

One challenge of this work was to define a criterion for the minimum detectable contrast. The Rose criterion,²⁵ stating that the signal-to-noise ratio (SNR) of 5 is needed to detect a signal with 100% certainty, was considered but could not be directly applied to our situation since our images are typically acquired with high SNR (~50 to 200 in most cases). Thus, in the absence of a generally-accepted contrast standard for high SNR images, an arbitrary value of ${\mathrm{TBR}}_{\mathrm{th}}=2$ was selected as the minimum to achieve detectable contrast. Note that selecting a higher value does not substantively alter the conclusions reached below, except for scaling of the QD concentrations, by a factor ${\mathrm{TBR}}_{\mathrm{th}}-1$. Thus,

3.2.

Wavelength Optimization Gain (G)

As a performance metric, the gain due to optimizing the wavelength is defined as the ratio between the threshold concentrations for the suboptimal and optimal illumination cases:

Similarly, the maximum gain between the worst wavelength choice and the optimized case is:

For the purpose of validating the model, the $c_{\mathrm{th}}$ and $G$ predictions were compared to values extrapolated from the measurements, obtained by substituting $x_{\mathrm{QD}}$ by $(X_{\mathrm{TR}}-X_{\mathrm{BR}}/c_{\mathrm{QD}})$ and $F_{\mathrm{MC}}$ by $X_{\mathrm{BR}}$ in Eqs. (5)(6)–(7). Also, since the autofluorescence predictions were poor in some cases (see Sec. 4.1), a simplified model in which the autofluorescence was measured directly instead of being calculated, was used for all data presented here, except for Sec. 4.1. Thus, $F_{\mathrm{MC}}$ is simply replaced by $X_{\mathrm{BR}}$, with $x_{\mathrm{QD}}$ unchanged.

4.1.

Autofluorescence Versus Tissue Dilution

First, the autofluorescence predictions were tested using four different homogenized tissues and five different emission wavelengths. To quantify the agreement between the measured and predicted autofluorescence versus dilution curves, the average reduced ${\chi }^2$ coefficient (deviation to error ratio) for each tissue and emission wavelength was calculated, using a relative error of $\pm 20\%$ for the autofluorescence measurements (an upper boundary that includes all sources of noise as well as intra- and inter-sample variations). Most values were below 2, except for some of the lung homogenates and the brain tissue at 650 and 700 nm. Fig. 4 shows an example of good agreement between measured and predicted autofluorescence.

Fig. 4

Typical case (liver, ${\lambda }_{\mathrm{em}}=700\mathrm{nm}$) of good agreement between the measured (symbols) and predicted (dashed lines) autofluorescence versus concentration of homogenized tissue. The data for intact tissue are immediately to the right of 100% homogenized tissue points.

Further analysis of the problematic cases suggested that the discrepancies are related to experimental conditions rather than being limitations of the model itself. For example, in lung tissue, steep variations in the experimental data are likely due to ex vivo tissue manipulations, such as dilution errors, large temperature variations,¹⁸ bacterial contamination of some samples or inter-sample variations due to differential blood oxygenation or blood retention. The model autofluorescence predictions were previously validated under fully controlled experimental conditions (using liquid phantoms of known optical properties). However, the accuracy of the predictions depends strongly on experimental inputs, such as the optical properties, being well determined. Since that was not the case for all tissues here, it was decided to use the directly measured AF data as being overall more accurate.

4.2.

Accuracy of the Model’s Contrast Predictions

To summarize the validation of the model predictions of image contrast, statistics were compiled for the optimal wavelength ${\lambda }_{\text{best}}$ best, threshold concentration for optimal-wavelength illumination $c_{\mathrm{th},\text{best}}$ threshold concentration for broadband illumination $c_{\mathrm{th},\text{broad}}$, and wavelength optimization gain $G_{\text{broadband}}$. For each bulk tissue, measurements and predictions were made for 40 different cases: five emission wavelengths for both QDs and matching dyes for each of four tissue configurations (25% concentration, surface; 25%, 480 µm subsurface; 100%, surface; intact, surface). The hollow organ measurements were also included (four organs at five emission wavelengths for two contrast $\text{agents}=40$ cases), both to increase the range of test cases and to assess the model predictions under more clinically (i.e., endoscopically) relevant conditions. For the subsurface configuration, 15 cases, including all 10 stomach measurements, were discarded because of low fluorescence signal. The validation results are summarized in Table 1.

Table 1

Percent of cases for each parameter in which the predictions were within a factor of 1.6 from the measurements.

The model was judged accurate if the error was $<10\mathrm{nm}$ for the wavelength predictions and if the predicted value was within a factor of 1.6, arbitrarily chosen to yield a target overall success rate of $\sim 80\%$. The poorest agreement was for the hollow organ/subsurface situation (due to the uncertainty on the measured thicknesses and weak target signals), for which the tolerance factor needed to be increased to 2.1 to achieve an 80% success rate. Thus, there is reasonable confidence that the model can predict the threshold concentration within a factor of 2 over a wide range of tissues and wavelengths.

5.1.

Calculation of Threshold Concentrations

Fig. 5 shows a typical example of measured and calculated TBR spectra for a set of tissue samples, and how the variations in TBR translate to significant differences in the $c_{\mathrm{th}}$ values. The uncertainty in the threshold concentration stems from 1. the variability in the AF measurements taken at multiple regions of interest on the tissue sample, and 2. the uncertainty on the $c_{\mathrm{QD}}$ value, estimated to $\pm 10\%$, as suggested by Yu et al.¹²

Fig. 5

(a) Measured (symbols) and modeled (dashed lines) TBR for 25%, 100%, intact and 25% subsurface (480 µm) imaging in kidney tissue. The true $c_{\mathrm{QD}}$ value was 4.0 µM. (b) $c_{\mathrm{th}}$ values calculated from Eq. (5) with ${\mathrm{TBR}}_{\mathrm{th}}=2$, using broad (green) or narrow band illumination centered at the worst (red) or optimal (blue) excitation wavelength. The worst wavelength was 465 nm, and the optimal was 405 nm for surface and 510 nm for subsurface imaging. The $c_{\mathrm{th}}$ values were significantly higher when using broadband or worst wavelength illumination, with $p<0.01$ (**) and $p<0.05$ (*).

In the example of Fig. 5, a simple change of excitation filter can reduce the threshold QD concentration by a factor 3.3 (intact tissue) to 6.6 (25%, subsurface). Similarly, using an optimized excitation filter instead of broadband illumination can reduce the required QD concentration 2- to 3-fold. These are straightforward instrumentation improvements that would translate into reduced cost and potential toxicity of the QD-based contrast agent. Note that the 25% subsurface case in Fig. 5(a) is a rare ($\sim 14\%$) instance of the model inaccurately predicting the optimal wavelength, likely due to underestimating the absorption at 405 nm and failing to predict the shift of optimal wavelength from 405 to 510 nm. This red-shift effect of target depth is discussed further in Sec. 7.3.

5.2.

Excitation Wavelength Optimization

Here, the contrast optimization is summarized in terms of optimal excitation wavelength. First, for each tissue sample, $c_{\mathrm{th}}$ was calculated at all excitation and QD emission wavelengths. Fig. 6 shows a typical example of $c_{\mathrm{th}}$ as a function of ${\lambda }_{\mathrm{ex}}$ and ${\lambda }_{\mathrm{em}}$, and it is clearly observed that the QD emission wavelength has much greater influence on the threshold concentration values than does the excitation wavelength. This is due to the properties of endogenous fluorophores¹⁹ and was observed for all tissue types, as shown below.

Fig. 6

Estimated threshold concentration ($c_{\mathrm{th}}$) versus ${\lambda }_{\mathrm{ex}}$ for QDs emitting at different wavelengths. This particular data set corresponds to surface imaging of undiluted (100%) homogenized kidney tissue. The lines are simply to guide the eye. Typical error bars are shown.

To remove the strong ${\lambda }_{\mathrm{em}}$ dependence, we normalized each curve by applying the transformation:

Fig. 7

Inverse gain (normalized contrast) versus excitation wavelength for different QD emission wavelengths (symbols), for surface imaging in homogenized tissue (left) and intact hollow organs (right). The shaded zones represent optimal excitation windows. The lines are simply to guide the eye.

The normalized contrast spectra were closely related to the excitation spectrum of the contrast agent and the tissue autofluorescence, which is itself strongly affected by the tissue absorption spectrum. Moreover, the tissue backscattering ratio had negligible influence on the shape of these spectra. Most tissues had a narrow optimization band, for all emission wavelengths, in the 395 to 425 nm region, corresponding to the main absorption peak of hemoglobin. This is important for clinical imaging applications, as it confirms that multiple QDs emitting at different wavelengths could be simultaneously excited optimally using a single excitation wavelength, which greatly facilitates multiplexed imaging of several biomarkers.

The lung, brain, and colon tissues had qualitatively different behavior than the other tissues. The colon had the lowest blood content, as confirmed by its low absorption spectrum [Fig. 3(b)], which translated into much less variation in autofluorescence and normalized contrast spectra. Although the lung and brain tissues had extremely different optical properties [Fig. 3(a)], their inverse gain spectra behaved similarly, exhibiting a shift in optimal wavelength from 405 towards 450 nm when using orange (600 nm), red (650 nm), and near infrared (700 nm) QDs. On average they had much lower AF intensity than the other tissues and the proportion of autofluorescence excited in the 420 to 550 nm region was less than the other tissues at 600 nm and negligible at 650 and 700 nm emission. This suggests perhaps not surprisingly that there are major differences in the intrinsic AF of brain and lung compared to the other tissues. Hence, although the wavelength optimization is driven by the hemoglobin absorption in most cases, the spectral signature of endogenous fluorophores and their relative concentration in tissues can also play a role.¹⁹

5.3.

Threshold QD Concentration Versus Emission Wavelength

Fig. 8 shows how the optimized threshold concentrations vary with emission wavelength. It is a common claim²⁰ that it is better to use near-infrared imaging contrast agents, since the tissue autofluorescence background is lower, but this has not been validated quantitatively to date. The results here show that the dependence is indeed strong. Thus, for example, changing from 500 to 700 nm emitting QDs reduces the threshold concentration by up to four orders of magnitude for the hollow organs (from 2800-fold for the colon up to 8000-fold for the bladder), and 3 orders of magnitude for the bulk tissues (from 550-fold for the brain to 2400-fold for kidney). This is also true for the subsurface experiments, as discussed in Sec. 7.3. Another important finding is that all tissues behave similarly with respect to the emission wavelength, and that in general the threshold concentration exhibits stronger dependence on the emission wavelength than on the excitation wavelength or tissue type.

Fig. 8

Surface optimal QD threshold concentration versus emission wavelength for (a) intact bulk tissues and (b) hollow organs. Note the logarithmic concentration scale.

6.1.

Gain Versus Tissue Optical Properties

In order to examine how the excitation wavelength optimization gain behaves with tissue optical properties, Fig. 9 shows the QD600 threshold concentration and corresponding wavelength optimization gain for all tissue types: note that $G=1$ corresponds to no improvement and that, due to the relatively high uncertainties for the estimated concentrations (up to $\pm 30\%$), gain values lower than 2 are not statistically significant in most cases. Similarly, due to the large uncertainties, comparing different gain values yields conclusions lacking statistical significance but still worth exploring here.

The threshold concentration and gain vary markedly with tissue type. As discussed in Sec. 5.2, the optimization gain depends on the tissue autofluorescence, which in turn is spectrally modulated by tissue pigmentation. Thus, highly fluorescent, strongly pigmented tissues exhibit the highest optimization gain values. It is also not surprising that the tissues with the lowest blood content (brain, colon) exhibit the lowest optimization gain. Finally, the lung, despite its strong pigmentation, is among the tissues with the lowest optimization gain, due to its very low AF at 600 nm.

Fig. 9

(a) Estimated threshold QD600 concentration values and the corresponding (b) maximum and (c) broadband contrast optimization gain for the bulk tissues. In (a), the statistical significance of the wavelength optimization is indicated: $p<0.01$ (**) and $p<0.05$ (*).

6.2.

Gain Versus Emission Wavelength

Figure 10 shows how the optimization gain varies with the QD emission wavelength for different tissues. The bladder, colon and brain tissues were chosen as representative of high pigmentation, low pigmentation, and spectrally variable intrinsic autofluorescence, respectively. In general, the optimization gain stays relatively constant with emission wavelength for all tissue types, except for slightly higher gain values at 700 nm. This is particularly true when looking at $G_{\max }$ values, as shown in Fig. 10. For example, Fig. 10(b) shows how the shift in optimal excitation wavelength in brain tissue (see Fig. 7) results in a minimum gain value at 600 nm (corresponding to a relatively flat $1/G_{{\lambda }_{\mathrm{ex}}}$ spectrum). However, Fig. 10(c) shows that, since highly pigmented tissues have $1/G_{{\lambda }_{\mathrm{ex}}}$ spectra of similar shape for all emission wavelengths, the gain stays relatively constant, except at 700 nm. The gain increases at 700 nm due to the sudden shift of the “worst excitation wavelength” from 465 (at ${\lambda }_{\mathrm{em}}<700\mathrm{nm}$) to 620 nm (see Fig. 7). This is true for most highly pigmented tissues (liver, kidney, bladder, esophagus, and stomach). However, this effect is due to the increasing range of excitation wavelengths for longer emission wavelengths, and so is specific to the particular experimental setup rather than being a real spectral feature of the tissues.

Fig. 10

(a) Estimated $G_{\text{broadband}}$ values for the intact bladder, colon and brain as a function of emission wavelength. (b) Estimated $G_{\max }$ values for the intact brain tissue, showing a drop at 600 nm that is likely due to a spectral shift in autofluorescence. (c) Estimated $G_{\max }$ values for the intact bladder, showing low variation except at 700 nm.

6.3.

Dose Reduction and Cost Implications

A clear advantage of the approach presented here is that both the threshold concentration and gain metrics allow, in principle, direct extrapolation to address issues in QD dosimetry. Thus, significant cost and toxicity reductions could be achieved if the findings translate into reduced dose in vivo. For example, a 200 µL vial of 2 µM QTracker565 solution currently costs $450 (Invitrogen, Carlsbad, California, USA), and it is recommended to inject $\sim 20\mathrm{\mu L}$ in a 25 g mouse for vascular imaging. Hence, if an equivalent dosage was used in a 75 kg human, then 60 mL would be required, at a cost of $>\$100,000$. However, based on our results for surface imaging of the esophagus, if the same procedure was done with QTracker705 using the optimal excitation wavelength, this dose could be reduced 250-fold, for a clinical cost of only $500. Clearly then, if QDs are ever to translate to the clinic, wavelength optimization and dosimetry will be defining criterions of feasibility.

6.4.

Quantitative Comparison Between Fluorescent Contrast Agents

Another potential application of the contrast optimization metrics is to establish quantitative comparison between different fluorescent contrast agents, such as QDs and organic fluorophores. Figure 11 shows two typical examples of QDs compared to AlexaFluor for surface imaging of bladder and colon tissue. Note that, although the threshold concentrations are similar at ${\lambda }_{\mathrm{em}}=500\mathrm{nm}$, they are always lower for the QDs at higher emission wavelengths. This is consistent for all tissues and, at ${\lambda }_{\mathrm{em}}=650\mathrm{nm}$ where the difference is the greatest, the threshold QD concentration is on average 17 times lower than that of the AlexaFluor. The likely reason is that more autofluorescence is excited at wavelengths that maximize the AlexaFluor contrast in comparison to the blue-shifted wavelengths that are optimal for QD imaging. Here, the peak contrast for the AlexaFluor imaging was always obtained at the longest excitation wavelength used (450, 510, 546, 600, and 620 nm for AlexaFluor488, 532, 568, 633, and 680, respectively).

Fig. 11

Example of surface QD and AlexaFluor threshold concentration versus emission wavelength for (a) bladder tissue and (b) colon tissue. Note the logarithmic concentration scales. The lines are simply to guide the eye.

7.1.

Experimental Limitations

This work demonstrates, in a wide range of different tissues ex vivo, how the tissue optical properties markedly affect wavelength optimization for QD-based imaging. Although the results are self-consistent, caution should be taken before applying them directly in vivo, due to the inherent difficulties associated with ex vivo tissue manipulation, due to altered hydration, blood content, oxygenation and temperature. In part, these factors are reflected in the high intra-sample variability of the autofluorescence measurements. Likewise, despite making all the measurements in the hollow organs as rapidly as possible, there were still intra-sample variations over time and optical property measurements made on a different set of similar freshly excised organs indicated some blood loss during storage and/or the imaging. Another limitation of the experimental setup was the use of a single capillary size, so that the optimal concentration estimates are specific to that target geometry. However, at least over the range where optical self-attenuation of the QDs is not significant, we would expect $c_{\mathrm{th}}$ values to scale inversely with capillary diameter.

In terms of the accuracy of the experiments, since the analysis involves taking several ratios of measured quantities, the errors propagate and reduce the statistical significance of the performance metrics. The main sources of errors were in the AF measurement (up to $\pm 20\%$), the absorption and reduced scattering measurements (up to $\pm 20\%$), and the QD concentration measurements ($\pm 10\%$). These also reduce the accuracy of the model predictions, as reflected by the tolerance factor of 2 derived in Sec. 4 (expected to be higher for subsurface geometry). Considering the uncertainties in the target depth of $\sim \pm 10\%$), the impact on the predicted threshold concentration is $\sim 15\%$ to 40% across the emission wavelengths, depending on the tissue optical properties. Despite these uncertainties, the approach still provides valuable guidance for applications, considering the significant excitation wavelength optimization gain values (up to 20-fold) and the even larger benefits of using optimum emission wavelengths.

7.2.

Modeling

Although the Monte Carlo model predictions were accurate across a wide parameter space, there were some problematic cases. As discussed, it is likely that the discrepancies are due to experimental factors, rather than the model breaking down. However, one clear limitation is that the tissue is modeled as a single homogenous layer. Nevertheless, reasonable agreement was still obtained even for the hollow organs, given the tolerance factor used and considering the multiple sources of experimental error, but greater precision could likely be achieved using a multi-layered Monte Carlo simulation^21,22.

As a simpler alternative to using the threshold concentration gain as the performance metric of choice, the contrast gain (i.e., a ratio of TBR values) could be used. These are equivalent as long as the signal is much higher than the autofluorescence, but when the signal is low and comparable to the AF background, the TBR values tend towards 1. This is problematic, since it then becomes impossible to differentiate between truly unstructured TBR spectra, and spectra that simply appear flat due to low signal: e.g., spectra such as those of Fig. 5 would show minimal structure if the QD concentration was significantly lower. The chosen optimization metrics do not suffer from this limitation: no matter what QD concentration $c_{\mathrm{QD}}$ is used for the measurements, $c_{\mathrm{th},\text{best}}$,$c_{\mathrm{th},\text{broad}}$, $G_{\text{broadband}}$, $G_{\max }$ are the same. It is, however, still necessary to know $c_{\mathrm{QD}}$ to properly scale the results, as indicated by Eq. (4). Note that all the surface imaging data presented above were acquired under the high signal conditions, and thus the TBR spectra would be equivalent to the presented normalized contrast. However, this is not the case for the subsurface data, mostly acquired in low signal conditions.

7.3.

Impact of Target Depth

As mentioned in Secs. 2.2 and 4.2, subsurface imaging was performed for all tissue samples, but, for the sake of brevity, the results are not reported above, since this was examined in detail in phantoms and homogenized tissues previously¹⁰ and, as noted, most subsurface measurements suffered from lack of QD signal due to strong tissue attenuation, as considered elsewhere.²³ Despite this suboptimal data quality, it was still possible to identify and confirm general trends on the effect of depth on wavelength optimization. Thus, in our previous work, we reported a local maximum at 500 to 550 nm (corresponding to a second hemoglobin absorption peak) in addition to the main 395 to 425 nm optimization window. It was suggested that the latter would not be sustained at depth due to the high QD signal attenuation, resulting in red shift of the optimal excitation wavelength in favor of the secondary peak. This trend was confirmed here: the 500 to 550 nm local maximum was observed in almost all cases for surface imaging (Fig. 7), and the red shift towards 500 to 550 nm was also seen in most subsurface cases, as shown in Fig. 12.

Fig. 12

Normalized contrast spectra for subsurface imaging in (a) 25% homogenized kidney and (b) intact bladder. The plain arrow illustrates suppression of the 395 to 425 nm peak and the dashed arrows highlight the red shift of the optimal excitation wavelength towards the 500 to 550 nm region (highlighted).

The subsurface measurements performed on hollow organs also allowed the effect of depth on the estimated threshold QD concentrations to be quantified. In general, the subsurface $c_{\mathrm{th},\mathrm{best}}$ versus ${\lambda }_{\mathrm{em}}$ curves were parallel to the surface imaging data, suggesting little influence of the emission wavelength on signal attenuation, especially at longer wavelengths (data not shown). The attenuation factor was mostly driven by the tissue thickness and absorption coefficient at the optimal excitation wavelength. The increase in threshold concentration from surface to subsurface (for both QDs and AlexaFluor) ranged from $\sim 10$ for the colon (the thinnest and least pigmented tissue) to $\sim 100$ for the stomach (thickest).

7.4.

Other Factors

In order to minimize the required dose of contrast agent, the fluorophore-to-background autofluorescence ratio must be maximized. Here, we have demonstrated that this can be achieved experimentally by selecting different excitation and emission filters. However, this exploration was limited to a specific set of filters of fixed bandwidth. In principle, imaging could be further optimized by tuning the bandwidths. Narrower filters may result in enhanced contrast if the excitation or emission spectrum of the fluorophore exhibits a sharp maximum and/or the tissue autofluorescence spectrum exhibits a sharp minimum. For example, using 20 nm rather than 50 nm bandwidth emission filters to detect the QD signal would result in $\sim 40\%$ increase in contrast, given the 40 nm FWHM QD emission spectrum. However, this would be at the expense of total brightness, so that the exposure times would have to be increased by $\sim 80\%$ to obtain the same integrated signal. Due to the relative flatness of the QD excitation spectra, there would not be significant gains in narrowing the excitation filters, unless the autofluorescence spectrum exhibits a sharp minimum for some specific tissue. In contrast, due to their narrower excitation spectra and smaller Stokes shift, fine-tuning the filter bandwidths and central wavelengths would improve the image contrast by up to 250% for the AlexaFluor dyes.

Finally, another way to further reduce the threshold concentration would be to increase the quantum yield of the fluorophores. Since the quantum efficiency of the QDs used here ranged from 0.16 to 0.92, the potential improvement would not exceed a factor 6. A summary of the different parameters and their estimated impact on dose reduction is shown in Table 2.

Table 2

Summary of photophysical parameters with respect to their impact on QD dose reduction. The arrows represent the change of parameter value (before→after) that result in QD dose reduction. For each parameter, the main factor responsible for variation of dose reduction values is included in the last row.