Contrast-enhanced proton radiographic sensitivity limits for tumor detection

Abstract.

Purpose: Proton radiography may guide proton therapy cancer treatments with beam’s-eye-view anatomical images and a proton-based estimation of proton stopping power. However, without contrast enhancement, proton radiography will not be able to distinguish tumor from tissue. To provide this contrast, functionalized, high-$Z$ nanoparticles that specifically target a tumor could be injected into a patient before imaging. We conducted this study to understand the ability of gold, as a high-$Z$, biologically compatible tracer, to differentiate tumors from surrounding tissue.

Approach: Acrylic and gold phantoms simulate a tumor tagged with gold nanoparticles (AuNPs). Calculations correlate a given thickness of gold to levels of tumor AuNP uptake reported in the literature. An identity, $\times 3$, and $\times 7$ proton magnifying lens acquired lens-refocused proton radiographs at the 800-MeV LANSCE proton beam. The effects of gold in the phantoms, in terms of percent density change, were observed as changes in measured transmission. Variable areal densities of acrylic modeled the thickness of the human body.

Results: A $1\text{-}\mu \mathrm{m}$-thick gold strip was discernible within 1 cm of acrylic, an areal density change of 0.2%. Behind 20 cm of acrylic, a $40\text{-}\mu \mathrm{m}$ gold strip was visible. A 1-cm-diameter tumor tagged with $1\times {10}^5$ 50-nm AuNPs per cell has an amount of contrast agent embedded within it that is equivalent to a $65\text{-}\mu \mathrm{m}$ thickness of gold, an areal density change of 0.63% in a tissue thickness of 20 cm, which is expected to be visible in a typical proton radiograph.

Conclusions: We indicate that AuNP-enhanced proton radiography might be a feasible technology to provide image-guidance to proton therapy, potentially reducing off-target effects and sparing nearby tissue. These data can be used to develop treatment plans and clinical applications can be derived from the simulations.

1. Introduction

Proton therapy is now an established method¹ of cancer treatment for certain classes of difficult-to-treat tumors, such as tumors of the head and neck,^2,3 especially in pediatric patients.⁴ It is uniquely able to spare nearby critical structures,^5–7 due to the enhanced dose-deposition accuracy provided by the proton Bragg peak.^8,9 The accuracy may be further increased using relativistic ($\sim 800\mathrm{MeV}$ or greater) protons,¹⁰ because the Coulomb-scattering angle is reduced at higher energies.¹¹ These high energy beams could deliver a highly conformal, stereotactic treatment,¹² as suggested by Durante et al.¹³ However, to take full advantage of the increased precision in dose delivery enabled by a higher energy proton beam, a real-time, in-situ imaging capability is needed to guide the treatment.¹⁴ Relativistic, beam’s-eye-view proton radiography has the potential to provide this sort of image guidance that can be used for proton treatment.^15–17

Lens-refocused proton radiography is fast and provides high spatial resolution¹⁸ but also has an intrinsically low level of soft-tissue contrast. This makes it impossible to differentiate healthy tissue from tumorous tissue because they have approximately the same density and $Z_{\mathrm{eff}}$, or effective nuclear charge. Functionalized gold nanoparticles (AuNPs), designed to specifically target malignant tissue, could enhance the contrast.^19–22

AuNP uptake decreases the transmission through the tumor, thus making the tumor detectable²³ and quantifiable.²⁴ Test phantoms were developed to optimize the configuration of the proton radiography beamline and to confirm our calculations of the expected image resolution. The phantoms consist of varying thicknesses of overlapping gold leaf on an acrylic base. Acrylic was chosen for its radiographic qualities that are like human tissue.

2. Methods

2.1. AuNP-Tagged Tumor Simulation

Proton radiography requires a difference in atomic number $Z$ and/or areal density within an object to visualize an inclusion or detail.²⁵ In this case, inclusion would be a tumor within healthy tissue of the same density and $Z_{\text{eff}}$. Our phantoms mimic the case where contrast is created by tagging the tumor cells with AuNPs. The range of gold thicknesses was determined based on known expectations of the contrast and sensitivity of proton radiography.²⁶ To correlate the gold thicknesses to conditions of tumor tagging, we calculate the thickness of gold in a tagged tumor, assuming cuboidal cell, spherical nanoparticle, and cubic tumor geometries (Fig. 1).

Fig. 1.

(a) Schematic of a cuboidal cell tagged with functionalized AuNPs and (b) a depiction of a volume of tumor comprised of many cells, where all of the gold is compressed into a layer at the bottom. The thickness of this gold layer ($T_{\mathrm{Au}}$) corresponds to the thickness of gold leaf or foil used for the phantoms constructed in this article.

Depending on the cancer type, the cell shape (cuboidal, columnar, squamous, etc.) and the tumor shape, these calculations may vary. For the sake of simplicity, a cuboidal cell and cubic tumor geometry²⁷ give a reasonable estimate that translates easily to our method of phantom construction. Assuming a eukaryotic cell diameter of $10\mu \mathrm{m}$, which is representative of the average cell size,²⁸ and guided by previous studies that showed that as many as ${10}^5$ AuNP can be delivered per cell,²⁹ the equivalent thickness of gold leaf for a given amount of gold material within a tumor would be

$$T_{\mathrm{Au}}=\frac{V_{\mathrm{Au}}}{L_{\text{tumor}}^2},$$ (1)

where $T_{\mathrm{Au}}$ is gold leaf thickness, $V_{\mathrm{Au}}$ is the sum volume of gold within a simulated tumor, and $L_{\text{tumor}}$ is the length of the cubic shaped tumor on a side. The total volume of gold within the tumor is given as

$$V_{\mathrm{Au}}=V_{\mathrm{AuNP}}\times N_{\mathrm{AuNP}},$$ (2)

where $V_{\text{AuNP}}$ is the volume occupied by a single gold nanoparticle and $N_{\text{AuNP}}$ is the total number of gold nanoparticles within a tumor. The number of gold nanoparticles is given as

$$N_{\mathrm{AuNP}}=N_{\mathrm{AuNP}/\text{cell}}\times N_{\text{cells}/\text{tumor}},$$ (3)

where $N_{\mathrm{AuNP}/\text{cell}}$ is the number of nanoparticles assumed to be tagged per cell, and $N_{\text{cells}/\text{tumor}}$ is the number of cells estimated to constitute a tumor of a given size. The number of cells within a tumor then is

$$N_{\text{cells}/\text{tumor}}=\frac{V_{\text{tumor}}}{V_{\text{cell}}}={\left(\frac{L_{\text{tumor}}}{L_{\text{cell}}}\right)}^3,$$ (4)

where $L_{\text{tumor}}$ is the length on a side of the tumor, $L_{\text{cell}}$ is the length on a side of a single cell, $V_{\text{tumor}}$ is the tumor volume, and $V_{\text{cell}}$ is the cell volume.

Thus, using $r_{\text{AuNP}}$ as the radius of the nanoparticles and $V_{\mathrm{AuNP}}=\frac{4}{3}\pi r_{\mathrm{AuNP}}^3$ for the volume of gold nanoparticles, the final expression for the equivalent thickness of gold leaf in a tagged tumor would be

$$T_{\mathrm{Au}}=\frac{\frac{4}{3}\pi r_{\mathrm{AuNP}}^3\times N_{\mathrm{AuNP}/\text{cell}}\times L_{\text{tumor}}}{L_{\text{cell}}^3}.$$ (5)

For example, a $1\times 1\times 1{\mathrm{mm}}^3$ tumor comprised of cells $10\times 10\times 10{\mu \mathrm{m}}^3$, having ${10}^5$ AuNP per cell of 40 nm diameter would have an equivalent gold foil thickness, $T_{\mathrm{Au}}$, of $3.4\mu \mathrm{m}$.

2.2. Au-Leaf Phantom Fabrication

Flat acrylic sheets (due to their tissue equivalent nature),³⁰ $\sim 0.5\mathrm{cm}\times 5\mathrm{cm}\times 5\mathrm{cm}$, support the fragile gold leaf [Fig. 2(a)]. Each set of two has through holes and tapped holes so they can be joined with nylon screws on the four corners. The gold leaf pattern fits within the smallest available $15\mathrm{mm}\times 15\mathrm{mm}$ field of view (FoV) of the $\times 7$ magnifier,³¹ as well as other lenses with larger FoVs available at the LANL proton radiography facility.

Fig. 2.

Phantom fabrication process. (a) Two square acrylic blanks, $\sim 5\mathrm{cm}\times 5\mathrm{cm}$ in cross section and 0.5 cm thick, side-by-side. (b) Acrylic blanks attached with nylon screws in each of the four corners. (c) Crosshatch design of gold-leaf strips, showing the number of layers of gold leaf in each column/row. (d) Assembled gold-leaf phantom encased by acrylic blanks, with $15\mathrm{mm}\times 15\mathrm{mm}$ pattern of gold leaf in the center.

A varying number of sheets of Au leaf (24 kt Genuine Gold Leaf, Barnabas Blattgold, Hong Kong) is used to create a specific thickness of Au on the acrylic support plates. Each sheet of Au leaf is $0.14\mu \mathrm{m}$ thick, as reported by the manufacturer. To build up more layers of gold leaf, a single sheet on tissue paper is folded in half and pressed by hand to seal the layers together, as the gold leaf readily adheres to itself. This process is repeated until the necessary number of layers is achieved. A single sheet can easily be folded into 64 layers, and four sheets of Au leaf can be folded into 128 layers. After it is folded to the appropriate thickness, a sheet of gold leaf is cut into strips $\sim 25\mathrm{mm}\times 2\mathrm{mm}$ with scissors for handling while it is still on the tissue paper backing. Next, the strips are trimmed to $\sim 15\mathrm{mm}\times 2\mathrm{mm}$ to fit the FoV of the $\times 7$ magnifier. Following the Au layer design shown in Fig. 2, the gold strips are placed on the clean, scored square of acrylic [Fig. 2(c)]. The strips of gold leaf are covered with a piece of tissue paper and pressed. The gold adheres to the surface without any adhesive. The tissue paper is then carefully removed, and the second acrylic square is attached with nylon screws to protect the fragile structure [Fig. 2(d)].

2.3. Imaging with Proton Radiography

Proton radiographs of the Au leaf phantoms were acquired at the 800-MeV proton radiography facility¹⁸ at the Los Alamos Neutron Science Center (LANSCE) at Los Alamos National Laboratory. An identity-magnification lens (12-cm FoV, $200\text{-}\mu \mathrm{m}$ spatial resolution),³² $\times 3$ magnification lens (4.0-cm FoV, $80\text{-}\mu \mathrm{m}$ spatial resolution),²⁵ and a $\times 7$ magnification lens (1.5-cm FoV, $25\text{-}\mu \mathrm{m}$ spatial resolution)³¹ were used to acquire the data. To ensure that the same number of protons illuminates a given pixel at the imaging plane for each lens configuration, the size of the proton beam at the object is changed, by adjusting the diffuser thickness. This prevents changes in imaging noise while scaling the dose accordingly between the lenses because the various lenses increase spatial resolution while decreasing the FoV. A 5-mrad collimator was used in all lens configurations. Proton radiographs taken of phantoms behind varying thicknesses of acrylic were visualized using the identity lens. Proton radiography relies on multiple Coulomb scattering (MCS) as a source of contrast, (Fig. 3),^25,31 which scales with the density and $Z$ of the target material. The experimental setup exploits this phenomenon to provide a higher degree of contrast than is provided by nuclear interactions alone. The proton radiographic technique described in this study is different than particle tracking proton radiography,^33,34 in that it relies on the attenuating effect of placing a collimator at the Fourier plane of the imaging lens to reduce transmission proportionately to scatter induced in the proton distribution by the object. In particle tracking proton radiography, protons are detected one at a time, with measurements of entry position, exit position, and residual energy, from which a map of energy loss through the patient can be reconstructed. In lens-refocused radiography (the technique we describe here), by contrast, the proton distribution exiting the patient is refocused by the lens system, which allows for an image to be acquired, and viewed, instantaneously. Transmission is attenuated proportional to the angular scatter accrued by the proton distribution by the collimator at the Fourier plane, with the amount of transmission lost proportional to the amount of material imaged through, as described in more detail below. One limitation of this lens-based technique, where we use integral scatter to infer areal density, is that the order in which protons traverse materials (such as bone then tissue or tissue then bone) has been demonstrated to have an impact on the overall scatter.³⁵ Although this effect is expected to be relatively small at higher energies, where scatter as a function of energy loss through the thickness of the patient remains relatively constant, it is indeed a larger effect at lower energies, approaching the clinically typical energy of 230 MeV. Smaller changes in areal density can be visualized by utilizing contrast agents of much higher $Z$, which impart the most scatter to protons as a function of material path length. Protons transmitted through the collimator opening are refocused at the image plane (Fig. 4). Although the lens is designed to correct for chromatic aberrations to first order, and thus has a relatively broad range of focus, energy loss determines the exact focus settings of the magnetic lens. A collimator may be selected such that contrast is most sensitive to small changes in areal density, where changes in thickness result in the highest proportional changes in transmission.³⁶

Fig. 3.

Proton–material interactions. Coulomb scattering increases with target areal density and $Z$. Tagging a tumor with high $Z$ nanoparticles increases the tumor’s areal density, increasing the amount of Coulomb scattering interactions, making the object visible by proton radiography.

Fig. 4.

Proton magnetic lenses with particle transport diagrams. Each lens comprised of four quadrupole magnets refocuses the scattered proton distribution at the image plane. (a) The identity lens has a 12-cm FoV and $200\text{-}\mu \mathrm{m}$ spatial resolution. (b) A $\times 3$ magnifier provides an $80\text{-}\mu \mathrm{m}$ spatial resolution and a 40-mm FoV. (c) A $\times 7$ magnifier has a 1.5-cm FoV and $25\text{-}\mu \mathrm{m}$ spatial resolution.

Protons at lower energies would accrue a greater amount of scatter while passing through an object and thus provide a more sensitive measure of contrast and density resolution. However, at higher energies, the decreased amount of accumulated Coulomb scattering allows for thicker objects (such as a human body) to be imaged with high spatial resolution, and a lower $\mathrm{\Delta }E/E$ (energy loss in material over lens focus energy) for a broader range of focus across proton energies.¹⁸ Scattering scales inversely with velocity and momentum, the Gaussian distribution of which is defined by ${\theta }_0$ as described as³⁷

$${\theta }_0=\frac{13.6\mathrm{MeV}}{\beta p}\sqrt{\frac{x}{X_0}}[1+0.038\ln (\frac{x}{X_0}\left)\right],$$ (6)

where $\beta$ is the relativistic velocity of the proton, $p$ is the proton momentum, $x$ is the areal density of the material ($\mathrm{g}{\mathrm{cm}}^{-2}$), and $X_0$ is the material’s radiation length.

From the scatter accumulated in the object, the transmission through the lens-based system is then determined as

$$T=\exp \left({\displaystyle \raisebox{1ex}{$-x$}\!\left/ \!\raisebox{-1ex}{${\lambda }_c$}\right.}\right)[1-\exp ({\displaystyle \raisebox{1ex}{$-{\theta }_c^2$}\!\left/ \!\raisebox{-1ex}{$2{\theta }_0^2$}\right.}\left)\right],$$ (7)

where ${\lambda }_c$ is the nuclear attenuation coefficient, ${\theta }_c$ is the collimator cut angle, the first part of the equation is the relatively small contribution from attenuation due to nuclear interactions, and the second part of the equation is the attenuation due to the collimation of the system. Transmission is then inverted to areal density using these equations.

2.4. X-Ray Imaging

X-ray was used as an imaging gold standard to assess the proton radiographs. The x-ray cabinet was a Northstar X25. The x-ray tube was set at 140 kVp with 100 mA exposures, with the phantoms placed flat against the detector. We acquired the x-ray data at typical settings equivalent to chest x-ray with an expected dose of 0.1 mGy.

2.5. Image Calibration and Processing

Image processing was performed in ImageJ.³⁸ All proton radiographs were dark field subtracted. Phantom images were divided by an average of several flat fields to correct for the intrinsic Gaussian distribution of the beam and remove any detector variations. The result is a flattened image. The calibrated image, $I_{\text{calibrated}}$, is given as

$${\mathrm{I}}_{\text{calibrated}}=\frac{({\mathrm{I}}_{\text{raw}}-{\mathrm{I}}_{\text{dark}})}{({\mathrm{I}}_{\text{beam}}-{\mathrm{I}}_{\text{dark}})},$$ (8)

where $I_{\text{raw}}$ is the raw image data, $I_{\text{dark}}$ is the dark field of the detector, and $I_{\text{beam}}$ is the data of the open beam, sometimes also referred to as a flat field image. The contrast range was adjusted for the images, including both the proton radiographs and the x-radiographs displayed in this paper.

2.6. Determination of Percent Density Change

While the Au thickness was an important factor in determining phantom visibility with proton radiography, it was also imperative to determine the percent density change in a simulated tumor once AuNPs have been added to its volume. The percent density change was calculated by multiplying the strip thickness by Au density ($19.3{\mathrm{g}/\mathrm{cm}}^3$). A 1% density change is the known threshold for visibility with proton radiography, within a uniform material.²⁶ The contrast-to-noise ratio (CNR) was calculated using the common metric:

$$\mathrm{CNR}=\frac{(S_{\mathrm{Au}}-S_{\mathrm{bg}})}{{\sigma }_{\mathrm{bg}}}.$$ (9)

The total Au mass per tumor was estimated by multiplying the AuNP mass by the number of AuNPs in the tumor. The tumor mass was calculated from an average density of $1{\mathrm{g}/\mathrm{cm}}^3$. The percent density change in $\mathrm{g}/{\mathrm{cm}}^2$ was then calculated based on these values. The CNR will vary as a function of the proton flux through the region where a measurement was taken. Because the proton beam has a Gaussian distribution in proton flux spatially, and experimental constraints caused the measurements to be taken through different portions of the Gaussian profile, the measured CNR values did not vary linearly with contrast-agent-thickness, as expected. A CNR correction was applied to account for varying proton doses as the proton flux decreases toward the periphery of the proton radiograph. This correction involved fitting a two-dimensional Gaussian to the beam profile of each proton radiograph and estimating the flux through each measurement by the fall-off in the Gaussian in the region where the measurement was taken. The CNR was then divided by $N^{1/2}$, where $N$ is the estimated local proton flux, so the flux normalized CNRs could be compared with the contrast agent thickness.

2.7. Application of Variable Amounts of Acrylic

We used a similar method to create phantoms of gold leaf and gold foil in a range of gold thicknesses (20 to $200\mu \mathrm{m}$) on variable thicknesses (0.15 to 20 cm) of acrylic to simulate various amounts of healthy tissue through which proton radiographs are acquired (Fig. 5). These gold phantoms, in a different fashion than the first set of phantoms, were protected by a single layer of scotch tape and attached to acrylic with tape. The images were calibrated (Sec. 2.5) and the percent density change and CNR were also calculated (Sec. 2.6) based on the new thicknesses of acrylic and gold.

Fig. 5.

Experimental set-up of gold phantoms on variable thicknesses of acrylic to simulate different amounts of healthy tissue. Proton beam moves from left to right. Gold phantoms range in thicknesses from 20 to $200\mu \mathrm{m}$.

3. Results

Results from the experiment with the $\times 3$ magnifier are shown in Fig. 6, which displays a phantom with 128-, 32-, 8-, and 2-layer Au leaf strips, corresponding to gold thicknesses of 18, 4.5, 1.1, and $0.28\mu \mathrm{m}$, respectively, in the nonoverlapping regions. Figure 6(a) shows the x-ray of the phantom, and Fig. 6(b) shows the proton radiograph. The 128-layer strip was most easily distinguishable and the 32- and 8-layer strips were also visible to some degree, with the eight-layer strip near the limit of visibility, in the proton radiograph. The 8-layer strip is $1.12\mu \mathrm{m}$ in thickness, which, for example, corresponds to a 3.5-mm cube tumor with $1\times {10}^4$ 40-nm diameter AuNPs tagged per cell, and $8.0\times {10}^{10}$ total AuNPs tagged in the tumor.

Fig. 6.

(a) Calibrated x-ray radiograph and (b) proton radiograph ($\times 3$ magnification) of a gold-leaf phantom. All Au-leaf strips are visible in x-ray. 128-, 32-, and 8-layer Au-leaf strips, corresponding to gold thicknesses of 18, 4.5, 1.2, and $0.28\mu \mathrm{m}$, respectively, are visible in the calibrated radiograph.

Proton radiographs were obtained of the gold foil phantoms behind varying thicknesses of acrylic (Fig. 7). The 200- and $120\text{-}\mu \mathrm{m}$ layers were the most easily distinguishable. As the thickness of acrylic increased, the range of gold foil visible diminished. All gold foil layers were discernable behind 0.15 cm of acrylic. Most of the layers except for the $20\text{-}\mu \mathrm{m}$ layer, which was approaching the limit of visibility, were visible behind 10 cm of acrylic. The 40- to $200\text{-}\mu \mathrm{m}$ layers were visible behind 20 cm of acrylic, where the $20\text{-}\mu \mathrm{m}$ layer was indistinguishable.

Fig. 7.

Calibrated proton radiographs (magnification with identity lens) of gold-foil phantoms through 0.15, 10, and 20 cm of acrylic.

The percent density change corresponding with every strip of gold leaf in the first phantom (Fig. 6) was plotted against the CNR of the gold leaf strips. A generally linear increase was observed as percent density change and CNR increased, demonstrating the contrast dependence on the amount of gold present in the phantom (Fig. 8). However, the 256-layer strip of gold, at the edge of the FoV (and thus on the edge of the Gaussian spread of the beam), received a lower proton dose, leading to greater noise in the proton radiograph due to decreased statistics. In the FoV of the phantoms with varying thicknesses of acrylic (Fig. 7), the gold strips were situated farther apart from the center of the proton beam, indicating that the same issue with lower proton dose on the peripheral gold strips would be observed at a greater magnitude. The plots of corrected CNR versus percent density change appear in Fig. 9 and display a linear trend, providing the expected relation of CNR to contrast that would be expected if the object were uniformly illuminated by the proton beam. Quantifiably, the CNR of the x-ray images is much larger than the proton radiographs, with a CNR of $\sim 400$ compared with a CNR of 5 when looking at the 128-layer gold leaf strips.

Fig. 8.

CNR as a function of the percent density change of tissue added by gold leaf. The sensitivity threshold was determined by eye to be a CNR of 2, because all steps above that threshold were visible.

Fig. 9.

CNR as a function of the percent density change of tissue (acrylic) added by gold leaf. CNR and percent density change were calculated for gold leaf phantoms on variable thicknesses of acrylic (a) 1.5-mm acrylic, (b) 100-mm acrylic, and (c) 200-mm acrylic. Blue dots represent observed CNR values, and red dots represent CNR values corrected by known proton flux. The red dashed line represents the linear nature of the fixed CNR values versus percent density change.

3.1. Clinical Considerations

The sensitivity limits of this technology allow us to extrapolate to various tumor sizes, nanoparticle types, nanoparticle sizes, and therefore, scale the simulations to the human body. The fractional density change the AuNPs will impose on the area of treatment can be determined given the sizes of the tumor, the tumor cells, and the nanoparticles, along with the thickness of the tissue $T_{\text{tissue}}$ and the number of nanoparticles tagging each cell. The fractional density change is given by the ratio of the areal density of the gold to the areal density of the tissue,

$$F_{\text{density change}=\frac{T_{\mathrm{Au}}\times {\rho }_{\mathrm{Au}}}{T_{\text{tissue}}\times {\rho }_{\text{tissue}}},}$$ (10)

where ${\rho }_{\mathrm{Au}}=19.3{\mathrm{g}/\mathrm{cm}}^3$ is the density of gold, and ${\rho }_{\text{tissue}}\approx 1{\mathrm{g}/\mathrm{cm}}^3$ is the density of tissue. Our observations show that a fractional density change of 0.2% is visible in proton radiographs of current technology. Radiographs of the gold foil phantom of Fig. 7, where no less than $T_{\mathrm{Au}}=40\mu \mathrm{m}$ is visible with an acrylic thickness or $T_{\text{tissue}}=0.2\mathrm{m}$, yield a more conservative value of $F_{\text{density change}}=0.4\%$ for visibility. The fractional density change can also be written as

$$F_{\text{density change}}=\frac{\frac{4\pi }{3}\times r_{\mathrm{AuNP}}^3\times N_{\mathrm{AuNP}/\text{cell}}\times L_{\text{tumor}}\times {\rho }_{\mathrm{Au}}}{L_{\text{cell}}^3\times T_{\text{tissue}}\times {\rho }_{\text{tissue}}}.$$ (11)

Thus, we conclude, for example, with a tissue thickness $T_{\text{tissue}}=25\mathrm{cm}$, a 5-cm tumor tagged with $3\times {10}^4$ gold nanoparticles of 40-nm diameter per cancer cell of $10\mu \mathrm{m}$ diameter, the fractional density change is 0.4%, which is at the limit of visibility with the current state of flash proton radiography technology. Proton radiographs were acquired with $1.5\times {10}^{10}$ protons per image, which corresponds to $\sim 3000$ protons per pixel. This deposited a dose of $\sim 30\mathrm{mGy}$ in the acrylic of the phantom, three orders of magnitude greater than the dose deposited by a typical chest x-ray. This was, however, in a regime optimized strictly for spatial resolution. If pixel resolution is reduced such that the spatial resolution for the largest FoV would be 1.0 mm, with the same number of protons per pixel the same density resolution presented here could be achieved with an imaging dose of $100\mu \mathrm{Gy}$, similar to a typical chest x-ray. Compared to a typical chest x-ray image, this would represent a much lower spatial resolution, however, would also be from a beam’s-eye-view perspective, enabling instantaneous registration between the image acquired, the perspective through which a treatment dose would be delivered to the patient.

The calculations can now help answer the practical question of whether the technique of nanoparticle tagging can reach the levels required to image tumors adequately to guide proton therapy. The tagging levels, in terms of nanoparticles per cell, needed to reach a particular density change can be calculated as follows:

$$N_{\mathrm{AuNp}/\text{cell}}=\frac{F_{\text{density change}}\times T_{\text{tissue}}\times {\rho }_{\text{tissue}}}{\frac{4}{3}\times r_{\mathrm{AuNP}}^3\times \left(\frac{L_{\text{tumor}}}{L_{\mathrm{cell}}^3}\right)\times {\rho }_{\mathrm{Au}}}.$$ (12)

Figure 10 shows an assortment of solutions to reach $F_{\text{density change}}=0.004$, for cancer cells of diameter $L_{\text{cell}}=10\mu \mathrm{m}$, and tissue thickness $L_{\text{tissue}}=25\mathrm{cm}$.

Fig. 10.

The level of nanoparticle tagging per cell of size $L_{\text{cell}}=10\mu \mathrm{m}$ to discern tumors in 25 cm tissue with proton radiography. For different scenarios, Eq. (10) gives the required average number of nanoparticles tagging each cell to produce density contrast of 0.4%, adequate to discern tumors. Studies observing as many as ${10}^5$ AuNP per cell have been reported.²⁹

4. Discussion

Proton radiography has a relatively low standard density detection threshold of 1%,²⁶ which makes imaging a tumor within healthy tissue intrinsically difficult. This difficulty arises because both healthy tissue and tumors have similar densities and $Z_{\text{eff}}$. However, this issue may be overcome using high-$Z$ nanoparticle tagging to enhance tumor contrast. Gold nanoparticles have a high $Z$, selectively bind to certain tissues, are biocompatible, and have been functionalized to certain tumor types.^39,40 With the current research, we have shown that with a high-$Z$ material, such as gold, the proton radiographic density detection threshold can be reduced to 0.2%. This increase in sensitivity is directly correlated to the increased proton scattering provided by high-$Z$ materials. We have applied our findings to predict the level of gold nanoparticle tagging, $N_{\mathrm{AuNP}/\text{cell}}$, that is required to image tumors with the current level of proton radiographic technology.

The proton radiographic technique examined in this paper is different from particle tracking proton radiography in a few ways. The particle-tracking technique tracks individual protons, their entry, and exit positions and then measures the residual proton energy in a range telescope.^33,34 The three-dimensional volume of the patient is reconstructed in units of proton stopping power or water-equivalent thickness. This technique, by contrast, estimates the water equivalence from the transmission equations, Eqs. (6) and (7) to instantaneously estimate stopping power, and, using the characteristic radiation length and nuclear attenuation coefficients for water in the inverse calculation, a map of estimated areal densities.

At higher proton energies, such as the energies available at LANSCE for this study, spatial imaging resolution can be very high, on the order of $200\mu \mathrm{m}$ at the clinical imaging scale. If this technique were to be implemented at lower proton energies, the spatial resolution will be lower due to increased MCS within the patient. However, at these lower energies, the density resolution (stopping power resolution) will be higher, due to an increased $\mathrm{d}E/E$ (energy loss over nominal beam energy) for the same material. The overall effect is that contrast increases along with sensitivity, allowing for the detection of smaller percent density changes, and comes also at the expense of lower spatial resolution.

The proton energy must be high enough to traverse a patient with enough residual energy and angular coherence to make a high-quality image. Thus, proton energies above 300 MeV would provide higher image quality in a clinical setting.⁴¹ This has the benefit of also offering potentially increased dose conformity, as proposed by Durante et al. (2012).¹³ In addition, with more heavy-particle treatment centers, such as carbon therapy centers, coming online worldwide, it is also possible to deliver higher energy proton beams that can be used for proton radiography, concurrently or interleaved with carbon treatment beams. Such concurrent proton radiography could be used to inform the day’s treatment session.

The primary advantage of this technology is that it can be used to provide instantaneous beam’s-eye-view imaging that would be useful for image guidance. In addition, because the images are created with protons, they can provide an accurate measure of proton stopping power that can be used for adaptive therapy once the nanoparticles have cleared the body. This is because the transmission through the system is directly related to the amount of scatter accumulated in an object, with the collimator cutting protons scattered beyond a specified scattering angle. If the distribution of protons traveling through material is a Gaussian with sigma defined by Eq. (6), from measurements of transmission, we get a direct measurement of theta. Any given transmission measured in pixel directly maps to a specific areal density of material. By doing an inverse operation on the formulas for scatter and transmission, the images in transmission space are converted to projection maps of measured areal density. With that known, the expected stopping power through the patient can also be derived. If the treatment particle is the same as the imaging particle, no assumptions are needed to convert proton radiographs to stopping-power maps, as is the case for example, in using an x-ray CT to estimate stopping power. Tumors could be visualized with the targeted tracer delineating their boundaries and stopping power could be quantified with standard proton radiography.

The data gathered in this report can be utilized to develop in vivo murine models of cancer with tracer for tumor detection with proton radiography. For a 5-mm-diameter tumor, tagged with nanoparticles of diameter $L_{\text{AuNP}}=40\mathrm{nm}$, in a mouse thickness $L_{\text{tissue}}=2.5\mathrm{cm}$, a 2% density change can be achieved with $1.6\times {10}^4\mathrm{AuNP}/\text{cell}$. The total mass of gold, $m_{\mathrm{Au}}$, is the total volume of gold times its density:

$$m_{\mathrm{Au}}=\frac{\pi }{6}\times L_{\mathrm{AuNP}}^3\times N_{\mathrm{AuNP}/\text{cell}}\times \frac{L_{\text{tumor}}^3}{L_{\text{cell}}^3}\times {\rho }_{\mathrm{Au}},$$ (13)

which is 1.25 mg. The volume of the stock solution necessary to achieve the desired mass of gold in the tumor $V_{\mathrm{AuNP}\text{stock solution}}$ can be determined by

$$V_{\mathrm{AuNP}\text{stock solution}}=m_{\mathrm{Au}}\times \frac{\mathrm{mL}}{m_{\mathrm{AuNP}}}\times \frac{1}{n_{\text{uptake}}},$$ (14)

where $m_{\text{AuNP}}/\mathrm{mL}$ is the mass of AuNPs per milliliter of the stock solution, and ${\eta }_{\text{uptake}}$ is the tumor nanoparticle uptake efficiency, where ${\eta }_{\text{uptake}}=1$ indicates that all nanoparticles bind to cells within the tumor.

With ${\eta }_{\text{uptake}}=1$, injection of 0.25 mL of a $5\text{-}\mathrm{mg}/\mathrm{mL}$ functionalized AuNP stock solution into the tail vein of a mouse will deliver the 1.25 mg of gold for this example. The authors leave it as an exercise to the reader to determine the specific uptake of their given nanoparticle functionalization/ligand combination. Future clinical applications could utilize the same technique (injecting NPs) to visualize tumors with proton radiography within patients.

While we suggest that gold nanoparticles be used as a high-Z contrast agent to detect tumor location, there will need to be a clearance period where the nanoparticles are allowed to naturally clear from the tumor cells. Gold, or other high-$Z$, nanoparticles will contribute unpredictably to delivered dose through a radiation enhancing effect.⁴² The appropriate clinical throughput then would be that targeted nanoparticles are injected into the patient, then a proton radiograph is taken of the tumor at the time of maximal uptake, and then once the nanoparticles are known to have left the tumor, proton therapy can then be delivered. Nanoparticles are eliminated from the body either through hepatobiliary or renal pathways, where factors, such as the shape, size, and chemical conjugation, determine their fate.⁴³ Significant amounts of the administered dose of nanoparticles are typically eliminated within hours through the renal pathway.^43,44 Nanoparticles are eliminated more quickly through the renal pathway compared with the hepatobiliary pathway, so it will be of interest to choose a nanoparticle design that allows for renal elimination.⁴³ However, it is predominantly important to consider the clearance of nanoparticles from the treatment site into the bloodstream, and gold nanorods were shown to be fully exocytosed from human breast cancer cells after 30 min.⁴⁵ This rapid elimination allows for the reversal to native tumor conditions prior to the radiation treatment within a reasonable time period.⁴⁵ Gold nanoparticles have been shown to be useful in other therapeutic techniques such as radiotherapy because of their intrinsic properties, so it is important that the nanoparticles eliminate from the tumor site prior to treatment to prevent any undesirable phenomena.⁴⁶

Finally, it is important to address the difference in energies between the proton radiographs taken in this study, at 800 MeV, and clinical proton energies of 230 MeV. At 230 MeV, a few things change. Protons accrue significantly more multiple Coulomb scatter (a factor of $\times 3$) from traversing the patient. The increase in MCS can be determined using Eq. (6), where the proton scatter is inversely proportional to the relativistic velocity of the proton ($B$) multiplied by the proton momentum ($p$). At 800 MeV, $\beta =0.842$, $p=1465\mathrm{MeV}/\mathrm{c}$, and at 230 MeV, $\beta =0.596$, $p=696\mathrm{MeV}/\mathrm{c}$. This ratio provides a factor of 2.97. This increase in scatter leads to two effects: one is that this leads to a greater degree of contrast, as changes in patient thickness translate into changes in transmission with the same proportionate effect, and so it is expected that the sensitivity to small changes in areal density would be more readily detected. This also has the effect of contributing more greatly to blur, as the scatter within the patient becomes a bigger contributor to the overall spatial resolution of the image. The other effect that we will see is that at 230 MeV, protons both lose more energy per unit length, as well as the energy lost, as a fraction of the nominal focus energy of the lens system is greater, by having a lower starting energy. These two factors combine to increase the blur associated with the chromatic effects of the lens, and this imposes a practical limit on the amount of material that can be imaged using these lens designs at 230 MeV, to less than the thickness of the human body. However, through the design of lenses specifically designed to reduce the chromatic effects,⁴⁷ it will be possible to image increasingly thicker objects, at lower energies. If the lenses used in this study, which have an overall drift length of 9.4 m, were reduced to something like 4 m, through the use of higher-field strength magnets, the chromatic lengths would be reduced approximately proportionately, reducing the chromatic effects enough to be able to image the human body, or by introducing additional lensing elements to make the lens at least partially achromatic.⁴⁸ We thus suggest that the study shown here be viewed in the context of what could be possible, either through the increasing of the proton beam energy, or the introduction of next-generation, low-chromatic length lenses, in extrapolating to the potential for imaging at the scale of the human body with clinical energy protons.

5. Conclusions

Proton therapy is an established method of cancer therapy, which has demonstrated improved patient outcomes, owing predominantly to a dose reduction to critical structures. As this treatment technique continues to advance, it could benefit from an instantaneous beam’s-eye-view image guidance, such as proton radiography. Proton radiography will require the use of a higher-$Z$ contrast agent, such as targeted gold nanoparticles, to delineate a tumor volume. The goal for this technology is to inform a daily treatment session, especially in the event that a patient’s anatomy has changed due to fluid buildup, radiation-induced pneumonitis, tumor shrinkage, or tissue inflammation. In addition, using 360 deg of projections would allow for the reconstruction of the patient’s anatomy. Gold leaf was used to model the gold uptake that could be seen in cancerous tissue using functionalized gold nanoparticles. A level of contrast sufficient for tumor visualization was achieved at the levels of nanoparticle uptake reported in the literature. Using the sensitivity limits established here, future work will aim to functionalize AuNPs to target tumors, to achieve enough proton radiographic contrast within the tumor to guide proton beam therapy.⁴⁹

Biographies

Rachel B. Sidebottom received her BA degree in biology from Occidental College in Los Angeles, California, with an emphasis in cellular and molecular biology. She is currently a post-baccalaureate researcher with the E-6, Non-Destructive Testing and Evaluation Group at Los Alamos National Laboratory, with a focus on applications of proton radiography toward medicine. Her research focus ranges from the histological study of the acellular linings of hemal spaces in crustaceans to studying the sensitivity limitations of contrast-enhanced proton radiography for image-guided proton therapy.

Biographies of the other authors are not available.

Disclosures

No conflicts of interest, financial or otherwise, are declared by the authors of this paper.