1.

Introduction

In today’s clinical practice, optical monitoring and imaging techniques are indispensible in disease management. Morphological and functional information, ideally down to the cellular level, is needed for diagnosis, for monitoring response to therapy, and for follow-up after treatment. Optical imaging modalities range from the visualization of intracellular and intercellular processes through (confocal) microscopy to macroscopic imaging by optical tomography. In between, techniques such as optical coherence tomography (OCT) image down to $2\mathrm{mm}$ in depth at micrometer scale resolution. The interaction of light with tissue can also be used to acquire functional information of the tissue under study, such as oxygenation,¹ perfusion,² blood content, tissue viability, and chemical characterization of malformations.³ Alongside instrumental developments, the integration of different techniques and the use of novel contrast agents such as gold nanoparticles⁴ are investigated to enable quantitative functional and molecular imaging of tissues.^{5, 6}

In vivo validation of these novel optical approaches is often difficult. Clinical acceptance requires proof of reproducible, device-independent, quantitative functional information in spite of the biological variability in the sought-after functional parameters. Moreover, for nanoparticle contrast agents, toxicity questions can exist. In such cases, the minimal detectable concentration of nanoparticle contrast agents needs to be determined ex vivo before their clinical use can be explored in vivo. For all these applications, tissue mimicking phantoms can play an essential role.^{7, 8, 9, 10, 11}

Tissue often possesses structural inhomogeneity. Layers of different cell types can be formed, and cavities containing fluid and blood vessels can disrupt uniform tissue structures, all resulting in a structural complex geometry. Disease often manifests itself as a change in the morphology of the tissue (e.g., blood vessels in a tumor causing increased hemoglobin contrast or loss of layered architecture). It is therefore paramount that structural variations such as layers or inclusions can be incorporated in the phantom. An “ideal phantom” has a number of other properties,¹² of which controllable tissue-like scattering and absorption,¹³ homogeneity,¹⁴ tissue-like refractive index,¹⁵ and durability¹⁰ have been extensively investigated. Within individual layers or inclusions of a structured phantom, these desirable properties should still be applicable.

In this work, we describe the development of easy-to-manufacture, low-cost phantoms using $50\text{to}300\text{-}\mu \mathrm{m}$ thin layers as building blocks, that fulfill the aforementioned desirable criteria. To our best knowledge, it is the first time that geometrical variations such as wavy structures mimicking the boundary between dermis and epidermis in skin and small capillary channels are demonstrated. The phantoms are characterized by OCT, transmission spectroscopy (TS), and confocal microscopy. To show that our approach enables construction of anatomically realistic phantoms that can be used in optical coherence tomography applications, images of a model eye/retina are presented. The latter is built from $50\text{-}\mu \mathrm{m}$ thin layers with different scattering properties. This model can be used to compare segmentation methods of clinically used OCT systems and can test their reproducibility. In addition, we describe protocols to include gold nanoparticle contrast agents, and present layered phantoms to quantify the contrast caused by these particles in OCT images.

2.

Materials and Methods

2.2.

Phantom Protocol

Individual phantom layers are made by mixing the desired concentration of absorber and/or scatterer with the curing agent component of the silicone elastomer. To obtain a homogeneous mixture, the particles or dyes are forced to mix with the curing agent by using a tissue homogenizer with a very small spacing between tube and pillar (VWR Labshop, Batavia, Illinois). Next, the mixture is placed in an ultrasonic bath for $10\min$ at $30\mathrm{kHz}$ to break residual clusters. The curing agent is then mixed with the silicone (1 to 9 weight ratio) under careful stirring using a standard laboratory mixer for $30\min$ . Remaining air bubbles are removed by using a vacuum pump that keeps the mixture under low pressure condition for $5\min$ . As shown in Fig. 1 , a small portion $\left(0.5\mathrm{ml}\right)$ of the final mixture is placed between two thick glass plates, separated by two spacers of the desired phantom thickness (ranging from $50\text{to}300\mu \mathrm{m}$ ). Finally, curing at $60°\mathrm{C}$ for $6\mathrm{h}$ or at room temperature for $24\mathrm{h}$ results in a thin single-layered phantom building block. Multiple layered phantoms are created by stacking the desired phantoms on top of each other. Accidental air bubbles between layers are removed using a vacuum pump. Electrostatic forces keep the phantom layers together.

Fig. 1

Final steps in the phantom-making process. (a) Standard thin-layer phantoms are molded between two glass plates of $1\text{-}\mathrm{cm}$ thickness. (b) One glass plate can be replaced with a machined plate with desired groove sizes. (c) Channels can be made by placing a wire within the phantom matrix material.

Geometrical complex phantoms like skin models with wavy structures are prepared by replacing one of the glass plates with a machined plate with the desired structured interface. This plate is then placed on top of the viscose matrix material before curing. After curing, both plates are removed, resulting in a structured phantom [Fig. 1]. A second layer can also be molded on top of the first structured phantom and can be cured again, resulting in a multiple layered phantom model with incorporated structure. Flow channels can be made by placing thin (electrical) wires in the layer, which can be removed after curing by carefully pulling the wire out of the phantom [Fig. 1].

Inclusion of gold particles in the phantom building blocks is challenging due to their high surface charge, which tends to lead to clustering when embedded in silicone elastomer. To avoid this, the gold particles are coated with polyethylene glycol (PEG 6000, Sigma Aldrich), a procedure used to couple antibodies to the gold particles when they are used as targeted contrast agents.²⁴ The PEG was added in equal amounts directly to $2\text{-}\mathrm{ml}$ gold particle-water solution $(4.1\times {10}^9\text{particles}∕\mathrm{ml})$ at room temperature for at least $1\mathrm{h}$ . Subsequently, the particles were centrifuged at $6500\mathrm{RPM}$ for $20\min$ . The PEG-containing liquid was removed with a pipette. The pallet of gold nanoparticles was dispersed in the curing agent of the phantom material as described before for obtaining a homogeneous mixture with other particles. We constructed four layered phantoms consisting of $\mathrm{Ti}{\mathrm{O}}_2$ -based, nonabsorbing phantoms ( ${\mu }_t=4{\mathrm{mm}}^{-1}$ , 50, 100, 150, and $200\mu \mathrm{m}$ thickness) placed on top of a $200\text{-}\mu \mathrm{m}$ thin nanoparticle-containing phantom $(\sim {10}^9\text{particles}∕\mathrm{ml})$ using the procedures described before.

3.

Results

OCT imaging of nonabsorbing phantoms of $300\text{-}\mu \mathrm{m}$ thickness with varying concentrations of $\mathrm{Si}{\mathrm{O}}_2$ and $\mathrm{Ti}{\mathrm{O}}_2$ scattering particles was performed. From the OCT data, fitting the attenuation coefficients using Beer’s law (Fig. 2 ) showed linear relations between attenuation coefficient and concentration, ${\mu }_t=4.67\pm 0.31{\mathrm{mm}}^{-1}$ /(weight percent) with an $R^2=0.986$ for the $\mathrm{Si}{\mathrm{O}}_2$ phantoms, and ${\mu }_t=21.56\pm 0.38{\mathrm{mm}}^{-1}$ /(weight percent) with an $R^2=0.998$ for $\mathrm{Ti}{\mathrm{O}}_2$ phantoms. The intercepts of both curves, ${\mu }_t=-0.34\pm 0.22{\mathrm{mm}}^{-1}$ and ${\mu }_t=-0.11\pm 0.11{\mathrm{mm}}^{-1}$ , show that there is no significant absorption or scattering by the matrix material in these phantoms at $850\mathrm{nm}$ . Figure 3 shows transmission spectra of $1\text{-}\mathrm{cm}$ -thick nonscattering phantoms containing the green dye as an absorber with five different concentrations (0.002 to 0.01 weight percent), measured with TS. In Fig. 3 attenuation spectra are shown for $517\mathrm{nm}$ [ ${\mu }_t=8.01\pm 0.87{\mathrm{mm}}^{-1}$ /(weight percent); $R^2=0.95$ ]; and $700\mathrm{nm}$ [ ${\mu }_t=14.07\pm 0.58{\mathrm{mm}}^{-1}$ /(weight percent); $R^2=0.99$ ], corresponding to minimal and maximal absorption. Again, a linear relation between attenuation coefficient and concentration was found. The intercepts of $0.016\pm 0.006{\mathrm{mm}}^{-1}$ for the $517\text{-}\mathrm{nm}$ measurement and $0.035\pm 0.004{\mathrm{mm}}^{-1}$ for the $700\text{-}\mathrm{nm}$ measurement indicate there is some residual scattering or absorption by the matrix material at these wavelengths.

Fig. 2

Phantom attenuation coefficient $\left({\mathrm{mm}}^{-1}\right)$ versus concentration (weight percent) of $\mathrm{Ti}{\mathrm{O}}_2$ (◻) and $\mathrm{Si}{\mathrm{O}}_2$ particles (●), measured with optical coherence tomography. Solid line is linear fit and dashed lines are 95% confidence bounds of the linear fit. For $\mathrm{Si}{\mathrm{O}}_2$ particles, the slope of the linear $\text{fit}=4.67\pm 0.31{\mathrm{mm}}^{-1}\mathrm{w}{\mathrm{\%}}^{-1}$ and offset ${\mu }_t=-0.34\pm 0.22{\mathrm{mm}}^{-1}$ $(R^2=0.986)$ . For the $\mathrm{Ti}{\mathrm{O}}_2$ particles, the $\text{slope}=21.56\pm 0.38{\mathrm{mm}}^{-1}{\text{weight}}^{-1}$ and offset ${\mu }_t=-0.11\pm 0.11{\mathrm{mm}}^{-1}$ $(R^2=0.998)$ .

Fig. 3

Attenuation spectra of five different nonscattering phantoms with ABS 551 absorber versus concentration. (a) Uncertainties of the data points are smaller than symbol size. (b) Attenuation coefficient $\left({\mathrm{mm}}^{-1}\right)$ versus absorber concentration (weight %) at $517\mathrm{nm}$ (◻) and at $700\mathrm{nm}$ (●). Solid lines show linear fits; dashed lines indicate the 95% confidence bounds of the linear fits. At $517\mathrm{nm}$ , the slope was $8.01\pm 0.87{\mathrm{mm}}^{-1}\mathrm{w}{\mathrm{\%}}^{-1}$ ; offset: ${\mu }_t=0.016\pm 0.006{\mathrm{mm}}^{-1}$ ; and $R^2=0.95$ . At $700\mathrm{nm}$ , the slope was $14.07\pm 0.58{\mathrm{mm}}^{-1}\mathrm{w}{\mathrm{\%}}^{-1}$ ; offset: ${\mu }_t=0.035\pm 0.004{\mathrm{mm}}^{-1}$ ; and $R^2=0.99$ .

Figure 4 shows size distributions measured inside $300\text{-}\mu \mathrm{m}$ thin phantoms obtained from confocal microscopy. The dashed line shows the size distribution obtained from a phantom containing 1 weight percent $\mathrm{Si}{\mathrm{O}}_2$ particles. According to the manufacturer, the size distribution of the $\mathrm{Si}{\mathrm{O}}_2$ particles is normal, with a mean radius of $500\mathrm{nm}$ and standard deviation of $58\mathrm{nm}$ . Calculations on the measured distribution reveal a mean radius of $730\mathrm{nm}$ with a standard deviation of $600\mathrm{nm}$ . Note that we did not correct these results for transversal resolution of the confocal microscope (FWHM $340\mathrm{nm}$ ). The positive skewness (0.6 versus 0 for a normal distribution) reveals the asymmetry of the distribution toward larger particle radii. This suggests that some clustering has taken place in the phantoms. The solid line shows the distribution obtained from a phantom containing 0.5 weight percent $\mathrm{Ti}{\mathrm{O}}_2$ particles. As for the $\mathrm{Si}{\mathrm{O}}_2$ particles, the data are not deconvolved to account for the transversal resolution of the microscope; the determined distribution therefore deviates from the expected $r=100\mathrm{nm}$ . We found a mean of $220\mathrm{nm}$ , standard deviation of $630\mathrm{nm}$ , and positive skewness of 0.43. This indicates clustering at the microscopic level.

Fig. 4

Estimated size distribution of particle radius inside $300\text{-}\mu \mathrm{m}$ phantoms obtained from confocal microscopy. Dashed line: phantom containing 1 weight procent $\mathrm{Si}{\mathrm{O}}_2$ particles. Mean: $220\mathrm{nm}$ ; SD: $630\mathrm{nm}$ ; skewness: 0.43; solid line: phantom containing 0.5 weight procent $\mathrm{Ti}{\mathrm{O}}_2$ particles. Mean: $730\mathrm{nm}$ ; SD: $600\mathrm{nm}$ ; skewness: 0.6.

Macroscale homogeneity was verified with OCT for the five different concentrations of $\mathrm{Ti}{\mathrm{O}}_2$ scattering phantoms. Attenuation coefficients were measured in three different regions of interest within each phantom ( ${\mu }_t$ roi1, ${\mu }_t$ roi2, and ${\mu }_t$ roi3). From these measurements, the average attenuation coefficient (mean ${\mu }_t$ ) and SD were calculated (Table 1 ). The variation in ${\mu }_t$ (SD/mean expressed as a percentage) was less than 8.5% for these phantoms. Similar results were obtained for the $\mathrm{Si}{\mathrm{O}}_2$ phantoms with a variation in ${\mu }_t$ less than 7.3%.

Table 1

Macrohomogeneity measured with OCT. Each concentration is measured using three different regions of interests within the phantom.

Tissue-comparable refractive indices were determined for five different phantoms using OCT (Table 2 ). Optical path length measurements with OCT showed a mean thickness of $465\pm 29\mu \mathrm{m}$ , and Gauge measurements showed a mean thickness of $328\pm 20\mu \mathrm{m}$ . This resulted in a mean refractive index of 1.42 with standard deviation of 0.01, which is close to the value specified by the manufacturer (1.41).

Table 2

Thickness measurements of 300-μm -thick phantoms measured with OCT and a precision Gauge tool, and corresponding calculated refractive indices.

Durability was verified using OCT and TS. As shown in Table 3 , measured ${\mu }_t$ $24\mathrm{h}$ after curing (t1) and after $6\text{months}$ (t2) showed small changes of less than 15% increase in scattering $\mathrm{Ti}{\mathrm{O}}_2$ phantoms. These variations are larger than the variations in macroscale homogeneity reported in Table 1. Nonscattering phantoms containing only absorber measured with TS $24\mathrm{h}$ after curing (t1) and after $4\text{months}$ (t2) also showed small changes (less than 5%), except for the 0.008% weight concentration phantom (Table 4 ).

Table 3

Durability measurements of TiO2 phantoms; t1 is 24h after curing, and t2 is after 6months .

Table 4

Durability measurements of ASB 551 phantoms; t1 is 24h after curing, and t2 is after 4months .

Structural variations were visualized with OCT. Figure 5 demonstrates a skin model with wavy structures within two layers representing dermis and epidermis. Figure 5 represents the 3-D reconstruction of the skin phantom. Figure 5 depicts a vascular phantom model with a vessel of $200\mu \mathrm{m}$ . This phantom is constructed from a 0.4 and 0.1% $\mathrm{Ti}{\mathrm{O}}_2$ phantom layer with a 20% Intralipid-filled channel within the bottom layer. Figure 5 represents the 3-D reconstruction of the vascular phantom.

Fig. 5

Geometrical variations of $\mathrm{Ti}{\mathrm{O}}_2$ particles containing phantoms obtained with the Santec OCT system. (a) Skin simulating phantom resembling the wavy dermal and epidermal structure of skin. (b) 3-D reconstruction of the skin phantom reconstructed from 250 B-scans. (c) Small $200\text{-}\mu \mathrm{m}$ channel within a $300\text{-}\mu \mathrm{m}$ -thick phantom layer. The phantom is constructed from a $300\text{-}\mu \mathrm{m}$ thin top layer (0.1% $\mathrm{Ti}{\mathrm{O}}_2$ ) and a $300\text{-}\mu \mathrm{m}$ thin bottom layer (0.4% $\mathrm{Ti}{\mathrm{O}}_2$ ), which includes the vessel filled with 20% Intralipid. (d) 3-D reconstruction of the channel phantom reconstructed from 250 B-scans.

OCT images of the model eye/retina taken with the Topcon 3-D OCT system are depicted in Fig. 6 . This phantom model is constructed from alternating 0.5% $({\mu }_t=11{\mathrm{mm}}^{-1})$ and 0.2% $({\mu }_t=4{\mathrm{mm}}^{-1})$ $\mathrm{Ti}{\mathrm{O}}_2$ phantom layers, fixed in a curved holder placed in a water-filled chamber. The $50\text{-}\mu \mathrm{m}$ -thick layers are clearly distinguishable from each other. Also visible is the adhesive tape used to fix the phantom in the eye, which shows up as a relatively transparent layer. Figure 6 is a schematic overview of the different parts of the model eye. Figure 6 part a shows the water-filled cube. A lens is positioned at the position shown by part b, the retina holder shown by part c can be placed inside the cube on a translation stage, shown by part d. The translation stage allows for changing the position of the retina holder. The cube can be closed with a water-sealed transparent lid on top.

Fig. 6

(a) Retinal phantom model imaged with the Topcon 3D, depicting the cross sectional image of the retina phantom and reconstructed 3-D image. Individual layers are $50\mu \mathrm{m}$ thin. A indicates adhesive tape. (b) Inset: Model eye with a water filled chamber; b is $f=20\mathrm{mm}$ achromat; c represents retina phantom; and d is eye length control.

Figure 7 , top row, shows a fused image from four OCT datasets taken with our time-domain OCT system at $850\mathrm{nm}$ . A phantom of $200\text{-}\mu \mathrm{m}$ thickness, containing approximately ${10}^9$ gold nanoparticles per ml, is placed underneath a stack of $50\text{-}\mu \mathrm{m}$ thin $\mathrm{Ti}{\mathrm{O}}_2$ containing phantoms with ${\mu }_t=4{\mathrm{mm}}^{-1}$ . The nanoparticles show up as bright spots in the OCT images, although they become less visible with increasing layer thickness (or OD) of the top layer. Labels in the figure indicate the thickness of the overlying layer. The lower row of Fig. 7 is a processed version of the upper image, in which the visibility of the nanoparticles is enhanced by subsequent median filtering and contrast-to-noise ratio calculation. In Fig. 6 we quantified the decrease of visibility with OD by calculating the contrast-to-noise ratio in five $25\times 25\text{pixel}$ windows in the phantom layer containing the gold nanoparticles (open squares). For reference, CNR was also calculated in the overlying phantom layer (solid circles). With increasing OD of the overlying layer, the CNR in the nanoparticle-containing layer decreases to $\sim 40\mathrm{\%}$ of its starting value, confirming the qualitative observations from the OCT images. Figure 7 shows that for the present combination of nanoparticles, sample, and OCT system, the nanoparitcles cause detectable contrast with the optical density of the overlying layer up to $\mathrm{OD}=0.4$ . For higher OD, the particles are present but their signal is not strong enough to be differentiated from the background. The CNR in the overlying layer is independent of the layer thickness, as expected for a homogeneous sample.

Fig. 7

(a) Compounded OCT images of four double-layered phantoms, with gold nanoparticles in the bottom layer ( $200\mu \mathrm{m}$ , $\sim {10}^9$ gold nanoparticles/ml). Top layers are 50, 100, 150, and $200\text{-}\mu \mathrm{m}$ thin phantoms containing $\mathrm{Ti}{\mathrm{O}}_2$ particles $({\mu }_t=4{\mathrm{mm}}^{-1})$ . Lower image is a filtered version of the top image to enhance the visibility of the nanoparticles. (b) Visibility (contrast-to-noise ratio) versus $\mathrm{OD}={\mu }_{t}d$ of the overlying layer in the phantom layer containing the gold nanoparticles (◻) and in the overlying phantom layer (●).

4.

Discussion

We demonstrated that homogeneous and durable silicone elastomer-based optical phantoms can be constructed using thin layers of $50\mu \mathrm{m}$ as building blocks, with controllable thickness, absorption, and scattering properties, refractive index, and they allow multifaceted structural variations resembling flow channels and wavy skin-like structures. We constructed a curved multilayered human retina phantom as an example. Novel types of contrast agents can also be incorporated. The phantoms are based on affordable materials, are easy to make, and fulfill many of the criteria for an “ideal phantom” as stated by Pogue ¹² We believe that this advanced class of phantoms is important for characterizing new and already existing optical techniques used in the clinic. Layers thinner than $50\mu \mathrm{m}$ can in principle be fabricated, although the curing process may then take longer than $6\mathrm{h}$ .

Our confocal microscopy results indicate that some microscopic clustering of the particles takes place. However, from the macroscale homogeneity study by OCT, we conclude that our phantom protocol is repeatable, and that Figs. 2 and 3 can be used to predict the optical properties as functions of phantom ingredients. Still, individual characterisation of phantoms is recommended.

The feasibility to incorporate molecules of specific interest (e.g., fluorophores) in a phantom building block is largely determined by the chosen matrix material and is closely related to our process of including nanoparticle contrast agents. The silicon matrix used in our phantoms is hydrophobic, which means that inclusion of biological chromophores may be challenging. For example, the absorption spectrum of the ABS 551 dye used in the measurements presented in Fig. 3 changed from its original spectrum supplied by the manufacturer during the curing process, which is probably due to a reaction with the curing agent. The original spectra were measured in methylene chloride, which is significantly different from a silicone elastomer. Nevertheless, after curing, the absorption spectra were stable. This behavior also suggests that mixing in fluorophores will not always be trivial. On the other hand, several absorbers specifically designed for silicone are available.²⁸ The minimal changes in optical properties over time can be contributed to the fact that all components are inorganic and chemically stable. Initial experiments in our laboratory show that it is even feasible to include red blood cells in the phantoms without significantly altering their spectral signature. Therefore, inclusion of specific molecules may be possible using appropriate encapsulation strategies while still maintaining their spectral properties. This might be achieved using a comparable PEGilation process as described with the gold nanoparticles. Although confocal microscopy (Fig. 4) and OCT images showed homogeneity (Table 4), it was suggested by Bisaillon ¹⁰ that better homogeneity might be obtained by mixing hexane with silicone resin, which results in a low-viscose but still-curable silicone mixture.

The phantom material has various avantages. The two-component silicon matrix allows variation of the mechanical properties by changing the component ratios. Silicon phantoms therefore have been used for elastography applications.²⁹ The ability to incorporate Brownian motion or flow in the phantom is possible by including flow channels or compartments containing fluids inside the sample. In this work we have already demonstrated a flow channel. Including a compartment containing fluid in Brownian motion that mimics perfusion will be more challenging, but we believe feasible. Assuming that the thermal properties of the phantom are determined by the thermal conductivity of the matrix material $(0.17\mathrm{W}∕\mathrm{mK})$ , this parameter is also in the same order of magnitude of most values found for soft tissues $(0.1\text{to}0.7\mathrm{W}∕\mathrm{mK})$ .³⁰

The time to manufacture a phantom according to the protocol presented in this work is approximately $8\mathrm{h}$ (including curing). Material costs of a phantom containing $\mathrm{Ti}{\mathrm{O}}_2$ particles and absorber is estimated at approximately €2 per phantom; phantoms with more expensive $\mathrm{Si}{\mathrm{O}}_2$ particles are estimated to be approximately €50 per phantom for the highest concentration. The solid phantoms do not interact with their environment and are consequently easily transportable. Covered transportation is desirable to keep the phantoms clean, since dust is attracted due to high surface tension of the silicone.

5.

Conclusions

The validation of novel biomedical optical imaging techniques requires phantoms that allow creation of complex geometrical structures and inclusion of novel types of contrast agents. We show the design and characterization of silicone elastomer-based optical phantoms. For the first time, thin complex structures approaching real tissue geometry are demonstrated. Moreover, these phantoms exhibit well-defined controllable absorption and scattering properties. We present phantoms with attenuation coefficients ranging from $2\text{to}11{\mathrm{mm}}^{-1}$ , scaling linearly with scatterer concentration. The phantoms are characterized using optical coherence tomography, confocal microscopy, and transmission spectroscopy. The phantoms demonstrate good microscopic and macroscopic homogeneity, and have a tissue-comparable refractive index of $1.42\pm 0.01$ . The phantoms, tested by repeated OCT and TS attenuation measurements after $6\text{months}$ , show small changes in the optical properties over time. We believe our phantoms fulfill many of the requirements for an “ideal” tissue phantom, and will be particularly suited for novel optical coherence tomography applications.