2.1.

Phantoms Simulating Tissue Optical Properties

Several photon propagation parameters and system specifications can be examined using phantoms. A polyurethane-based phantom has been suggested to assess the sensitivity of fluorescence cameras.³⁰ The phantom comprised six cylindrical wells that contained ${\mathrm{TiO}}_2$ particles to impart photon scattering and varying quantities of NIR fluorescent QDots dispersed in polyurethane [Fig. 1(a)]. The camera sensitivity was then assessed from an image of the phantom by observing the number of visible wells on the image. In addition, epoxy–resin-based phantoms that combine titanium dioxide (${\mathrm{TiO}}_2$) and aluminum oxide (${\mathrm{Al}}_2{\mathrm{O}}_3$) can be employed to control for photon scattering anisotropy and the phase function.³¹

Fig. 1

(a) Phantom of fluorescent wells of different strength employed to assess the sensitivity of far-red and near-infrared fluorescence imaging systems. Adapted from Ref. 30 with permission. (b) Cylindrical phantom constructed to assess fluorescent capillaries immersed at different depths under the surface within a medium simulating tissue optical properties (cross-sectional view) reproduced from Ref. 33 with permission. (c) Schematics of a bilayer (up) and an inclusion (down) phantom. Adapted from Ref. 34 with permission.

A different phantom, made by spraying a polyurethane–QDot solution onto a reflective surface, has been suggested for determining the leakage of excitation light into the fluorescence channel of a camera.³² Images from the QDot distribution containing both fluorescence and light-leakage signals are compared to images from an area that does not contain QDots to separate the strength of fluorescence contributions compared to excitation light cross talk into the fluorescence channel. Zhu et al. further suggest the use of varying QDots solutions to infer the dynamic range of the camera.

In another study,³³ a customized cylindrical structure with fully submerged thin tubes containing a solution of (CdSe)ZnS QDots in chloroform was constructed for assessing the imaging quality achieved as a function of fluorochrome depth. The medium surrounding the tubes consisted of a water-based homogeneous solution containing intralipid-20%, naphthol green B powder, and distilled water. A small amount of AlexaFluor 568 (0.2 mg in a total volume of 320 mL) was added to simulate tissue autofluorescence [Fig. 1(b)]. The use of phantoms that establish a geometry with varying thickness and optical properties Fig. 1(c)] was suggested for characterizing tissue autofluorescence and validating fluorescence tissue analysis.³⁴ The phantoms were made of porcine skin gelatin and employed rhodamine B (RhB) or FITC to impart fluorescence. The optical properties were selected to correspond to normal and diseased brain tissues. Rectangular phantoms containing variable-diameter cylindrical inclusions placed in parallel were also suggested.³⁴

2.2.

Shape-Maintaining and Organ-Mimicking Fluorescent Phantoms

A different class of tissue-mimicking phantoms focuses on simulating the geometrical features of tissue organs and can be used for clinical training in intraoperative imaging. A lymph-node mimicking phantom was constructed²¹ to assess the efficacy of a real-time fluorescence imaging system [Fig. 2(a)]. The phantom was made of thin-walled polyoxymethylene cones (15-mm long and $300\text{-}\mu \mathrm{m}$ thick) filled with scattering fluid (water and whole ultraheat treated milk of 3.5% fat) and fluorescent dye (ICG-Pulsion). Five lymph node-mimicking cones were mounted on a holder and immersed in a glass reservoir containing a water and milk scattering fluid.

Fig. 2

(a) Fluorescence lymph node phantom: cross-sectional view, reflectance image, and normalized epi-fluorescence image. Adapted from Ref. 21 with permission. (b) Complex phantom containing four compartments (a–d) with different optical properties, each of them contains a fluorescent 1-mm bead (indicated by arrow) that is detected using NIRF reproduced from Ref. 35 with permission. (c) Preoperative detection of fluorescent tumor-simulating inclusion using NIRF; figure reproduced from Ref. 36 with permission.

Rectangular- or breast-shaped phantoms were also considered as a tool to train surgeons in intraoperative fluorescence imaging in breast cancer.^35,36 A phantom made of Tris-buffered saline (TBS, buffer), sodium azide, gelatin, and varying amounts of intralipid and haemoglobin was employed to simulate the geometry, optical properties, and the mechanical properties of the human breast.³⁵ ICG-tagged polystyrene divinylbenzene beads (AG1-X8) or Pam78-labeled hydroxyapatite crystals (simulating breast cancer microcalcifications) were used as fluorophore inclusions [Fig. 2(b)]. In Ref. 36, varying-shaped ICG-agarose inclusions were embedded in a mixture of TBS, gelatin, ${\mathrm{NaN}}_3$, haemoglobin, and Intralipid^©, and the final structure was visualized using a customized NIRF system optimized for real-time surgical imaging³⁶ [Fig. 2(c)].

2.3.

Requirement for a New Class of Phantoms

Table 1 summarizes a nonexhaustive list of fluorescent phantoms and demonstrates a great variability in approaches and materials used. The phantoms listed have been developed to address different specifications and needs, e.g., the characterization of a specific property (e.g., sensitivity) or the study of a physical property (e.g., effect of depth). Typically, only a single or few specifications are addressed.^30,32,37 However, there is a large range of parameters that may require validation in the clinical use of fluorescence cameras. Parameters such as the illumination homogeneity, the resolution, the dependency of fluorescence intensity on tissue optical properties, or depth are not generally comprehensively addressed in the phantoms built so far. Therefore, there is a need for a new class of phantoms that simultaneously assess multiple parameters. This is particularly a requirement in intraoperative settings, where a single image calibration can longitudinally confirm system operation. Also, a comprehensive phantom that concurrently enables the measurement of multiple camera parameters is necessary for the comparison of different camera systems or operational parameters. The use of a comprehensive phantom, therefore, eliminates the need to develop a large number of different phantoms, each testing a single or few parameters only.

Table 1

Examples of fluorescent phantoms found in the literature (nonexhaustive list).

3.1.

Phantom Design and Fabrication

The solid phantom designed is depicted in Fig. 3. Figure 3(a) shows a color image of the fabricated phantom and Fig. 3(b) shows a fluorescence image of the phantom. The phantom is made of transparent rigid polyurethane (WC-783 A/B, BJB Enterprises, Tustin) and contains different structures interrogating different aspects of fluorescence camera specifications and imaging performance. The outer dimensions of the phantom are $10\times 10\times 2.2{\mathrm{cm}}^3$. The background absorption was set to $\sim 0.5{\mathrm{cm}}^{-1}$ at 750 nm, using alcohol soluble nigrosin (Sigma Aldrich, St. Louis). Scattering was imparted by using $1\mathrm{mg}/\mathrm{g}$ ${\mathrm{TiO}}_2$ particles (Titanium IV Oxide, Sigma Aldrich, St. Louis), resulting in a reduced scattering coefficient of $\sim 10{\mathrm{cm}}^{-1}$ at 750 nm. The high-background absorption coefficient in this phantom offers the additional advantage of low photon diffusion from the different structures selected into the background medium. The absorption coefficient was determined by measuring the extinction coefficient of nigrosin using an Ocean Optics USB4000-UV-VIS spectrometer and then selecting the appropriate concentration to impart the desired absorption coefficient. The reduced scattering coefficient was determined by fitting the diffusion equation in a measurement of the spatial transmittance profile of a point source through a 1-cm slab of cured polyurethane containing ${\mathrm{TiO}}_2$ particles and a known absorption coefficient, as determined by measuring the solvent by photospectrometry prior to the addition of the particles. All the inlets in the phantom, except for the ones used for depth dependence evaluation, were made 12-mm deep. The different phantom compartments are schematically explained in Fig. 3(c).

Fig. 3

Top view of the phantom. (a) Color image of the phantom. (b) Fluorescence image (765 to 855 nm) of the phantom excited at 750 nm. (c) Schematic of the phantom.

The upper-right quadrant of the phantom interrogates the sensitivity and fluorescence intensity variation as a function of optical properties through an array of nine fluorescent wells (10-mm diameter). The wells contain a mix of cured polyurethane with QDots (800-nm emission; Thermofisher Scientific, Waltham) at varying concentrations (1, 5, and 10 nM) across the columns and varying absorber (20, 20, and $40\mu \mathrm{g}/\mathrm{g}$ Hemin) and varying ${\mathrm{TiO}}_2$ amounts (0.33, 0.66, and $1\mathrm{mg}/\mathrm{g}$ ${\mathrm{TiO}}_2$) across the rows. Hemin (Sigma Aldrich, St. Louis) was selected to simulate an absorption coefficient $0.25{\mathrm{cm}}^{-1}$ for $20\mu \mathrm{g}/\mathrm{g}$ and $0.5{\mathrm{cm}}^{-1}$ for $40\mu \mathrm{g}/\mathrm{g}$.

The bottom-right quadrant comprises nine fluorescence wells; whereby 10-mm diameter cylinders made of 10 nM 800 nm QDots $0.66\mathrm{mg}/\mathrm{g}$ ${\mathrm{TiO}}_2$ particles and $20\mu \mathrm{g}/\mathrm{g}$ Hemin were embedded at varying depths within the phantom, i.e., at an increasing depth from the phantom surface at 0.2, 0.4, 0.6, 0.8, 1, 1.33, 1.66, 2, and 3 mm. After the QDot mix was poured into the well to reach the appropriate depth, it was allowed to cure and then the cavity was filled with the same material employed in the construction of the phantom body. This quadrant examines the fluorescence sensitivity to depth and can be used to study the ability of a system or method to account for the effects of depth on the fluorescence signal.

The upper-left quadrant examines the camera dark-current offset and the camera cross talk, i.e., the excitation light leakage into the fluorescence channel. Light leakage could be caused due to inadequate excitation light rejection from system filters. This area comprises a circular well of very high absorption coefficient ($3.74\mathrm{mg}/\mathrm{g}$ nigrosin) and one of low absorption and high scattering coefficient ($10\mathrm{mg}/\mathrm{g}$ titanium oxide) offering reflecting characteristics. The circular elements were embedded in the background material and do not contain fluorescence substances. Inspection of the dark area approximates the dark-light camera photon count “dc-offset” (although a separate measurement with the excitation light off also can be performed). Inspection of the reflective nonfluorescent area reveals the camera cross talk with better signal-to-noise ratio characteristics over observing the cross talk seen from the phantom background material. This cross talk measurement can be utilized in different ways as discussed in Sec. 3.2. In addition, signals captured in the fluorescence channel from the highly scattering well of the upper-left quadrant can be employed in a differential way to evaluate stray light, e.g., using two measurements, one with ambient light on and one with ambient light off.

Finally, the lower-left quadrant tests for the resolution of the optical system and cameras employed. A standard 1951 United States Air Force resolution test chart is fixed on the top of the phantom to interrogate the white-light optical resolution, useful for characterizing the color or the fluorescence camera employed. In addition, an L-shaped diffusive fluorescence structure, made of polyurethane, $0.66\mathrm{mg}/\mathrm{g}$ ${\mathrm{TiO}}_2$ particles, $20\mu \mathrm{g}/\mathrm{g}$ Hemin, and 10 nM QDots, was embedded for the examination of photon diffusion and fluorescence resolution. The outer dimensions of the L-shaped structure are $30\mathrm{mm}\times 30\mathrm{mm}\times 12\mathrm{mm}$, whereas the width of each branch is 5 mm.

There are five identical reflective circular areas (5-mm diameter) made of $10\mathrm{mg}/\mathrm{g}$ titanium oxide in polyurethane, four at the corners of the phantom and at the center sample the homogeneity of the light illumination employed by the camera system.

The phantom was produced by molding a cuboid made of the base material. After curing, the cuboid was treated by face milling of the surfaces. Openings were milled for each of the desired wells and shapes. All milling procedures were performed on a cartesian computer numerical controlled three-axis milling router with $\mu \mathrm{m}$ resolution.

3.2.

Imaging System

A home-made EagleRay-V3 imaging system previously developed for intraoperative imaging³⁸ was employed for the experimental measurements herein. The EagleRay-V3 system is a hybrid color-fluorescence system employing an iXon-3 electron multiplying charge-coupled device (DU-897 EMCCD; Andor Technology, Belfast, Ireland) for fluorescence detection and a 12-bit-CCD camera (pixelfly qe, PCO, Kelheim, Germany) for color detection. A 300-mW continuous laser diode (BWF2-750-0, B&W Tek, Newark) at 750 nm was used for fluorescence excitation and a 250-W halogen lamp (KL-2500 LCD, Scott, Mainz, Germany) was employed for white-light illumination.

4.1.

Sensitivity and Intensity Variation Due to Optical Properties

System sensitivity can be assessed by observing the fluorescence collected from the upper-right quadrant wells. Figure 4(b) shows the fluorescence profiles across the nine wells, plotted from right to left as shown in Fig. 4(a). The profiles demonstrate variations in intensity since each row of the $3\times 3$ matrix of cells contains a different combination of optical properties and each column contains a different QDot concentration at 1, 5, and 10 nM from right to left, respectively. The plot allows the evaluation of sensitivity not only under different fluorochrome concentrations, but also for varying optical properties. Even though each column contains exactly the same amount of QDots, different fluorescence intensities are recorded from the three wells in that column due to the variation of optical properties. Importantly, therefore, the upper-right quadrant can also be employed to evaluate algorithms that account for the effects of optical properties on the fluorescence intensity.

Fig. 4

Cross sections of the fluorescent image of the phantom. (a) Phantom schematic with arrows painted with colors corresponding to the cross sections. (b) Fluorescence intensity across each column (different fluorophore concentration). (c) Intensity across the highly reflecting and absorbing area. (d) Fluorescence intensity for different depths. (e) Fluorescence intensity versus the depth distance.

As observed in Fig. 4(d), the system can detect 1 nM of fluorochrome concentration only in the low-attenuation well. Therefore, it appears that under the operational conditions applied herein, the sensitivity of the system is between 1 and 5 nM, depending on the background optical properties.

4.2.

Light Leakage

Figure 4(c) depicts the cross section across the two wells contained in the upper-left quadrant of the phantom, which examines the light leakage and dark count. The light-leakage measurement can be utilized in different ways. It can be related to fluorescence intensity measurements (e.g., from the upper-right quadrant of the phantom) to provide a relative metric of light-leakage recorded from a low-absorption and nonfluorescent lesion to measurements from lesions containing varying amounts of fluorochrome. The ratio of fluorescence measurement for a certain well in the upper-right quadrant over the light-leakage signal is a quality metric; the higher the value the better the ability of a camera to discriminate a fluorochrome over cross talk. Another method utilizing the light-leakage measurement is to observe the intensity of the excitation light reflected off the reflective well using the color camera [i.e., from Fig. 3(a)] and generate a ratio of the light-leakage measurement [from Fig. 3(b)] over reflected light in Fig. 3(a). This ratiometric measure can also be used to compare the performance of different cameras or examine the performance of algorithms correcting for light leakage.

The measurement shown in Fig. 4(c) was intentionally performed using filters offering medium excitation light rejection to illustrate that it is possible to obtain light-leakage signals that are stronger than fluorescence signals, depending on the system design. For example, when comparing the signal intensity collected in the leakage channel to the fluorescence intensity seen in the left-most column of the upper-right quadrant, one can observe that the light leakage is stronger than the fluorescence signal obtained from low (1 nM) QDot concentrations and highly absorbing lesions.

4.3.

Depth

Figure 4(d) shows the fluorescence intensity profiles as a function of fluorochrome depth across the nine varying depth entities, plotted in the order that is indicated by the arrow in Fig. 4(a). As expected, a decrease of fluorescence intensity is observed for increasing depth. The rate of signal drop with depth is depicted in Fig. 4(e). This figure shows at least a twofold fluorescence intensity reduction over the first 3 mm of lesion depth and for the optical properties selected. This measurement can be employed to examine the performance of cameras and algorithms that account for depth-related fluorescence signal variations.

4.4.

Fluorescence Resolution

Figure 5 shows the fluorescence image from the lower-left quadrant, which can be employed to evaluate the diffusive resolution achieved in the fluorescence channel. The manufactured L-shape has a sharp border. However, due to photon diffusion, the border appears of low resolution. To evaluate the diffusive resolution, we plotted fluorescence intensity profiles [Fig. 5(b)] along the diagonal double-arrow lines plotted in Fig. 5(a). The diagonal double-arrow line indicates the $x$-axis in Fig. 5(b), whereas the $y$-axis is the fluorescence intensity recorded along the profile shown and the position on the $d$-axis indicates the distance from the inner vertex of the L-shaped structure. The minimum step between the lines could be 1 pixel diagonal, but for better visualization, a step of 20 diagonal lines was used. The profiles confirm that even though the L-shape has been constructed with a hard border between fluorescence and nonfluorescence material, the resulting reading has a diffusive appearance. These readings can be useful for evaluating cameras and algorithms that improve upon photon diffusion. One resolution improvement metric that could be implemented with this phantom would be the comparison achieved between the L-shaped edge (border) resolved on a diffusion-corrected image and the location of the known manufactured edge of the L-shaped structure. Another resolution metric could be implemented by reporting the distance on the $x$-axis [Fig. 5(b)], whereas the plotted fluorescence intensity profiles start to be resolved. The L-shaped phantom allows for a gradual distance increase between the two edges, so that different distances between the edges can be seamlessly evaluated.

Fig. 5

Resolution assessment. (a) L-shaped structure and the lines drawn across the $x$-axis. (b) Cross section profiles of the green and red line from figure (a) and their distance from the inner vertex of L-shaped structure is shown.

4.5.

Illumination Homogeneity

The illumination homogeneity can be evaluated by plotting the intensity measurements recorded for the five reflective wells placed in the corners and center of the phantom. Figures 6(a) and 6(b) depict the intensity profiles as cross sections across the five reflective wells. These profiles serve as a quick estimation of the illumination field homogeneity. Using five sampling points, we can define a three-dimensional surface that approximates the illuminating field. Since a single illumination source was employed in this study, we fitted a cubic spline curve to the five points, as shown in Fig. 6(c). More generally, a larger number of sampling points on the phantom and higher order polynomial surface fits can be employed for describing more complex illumination fields. Tracking the illumination pattern, as shown in Fig. 6(c), can help in correcting the fluorescence images by normalizing them for the excitation field variation or help to adjust the illumination to make it more homogeneous.

Fig. 6

Assessment of the homogeneity of the illumination field of the camera system. (a) Homogeneity profiles across the left-upper-corner to right-bottom-corner. (b) Homogeneity profiles across the right-upper-corner to left-bottom corner. (c) Comparison of the five reflective spots surface profile with the surface profile of a white reflectance sheet.