MRI of Radiation Chemistry: First Images and Investigation of Potential Mechanisms

Abstract

Background:

Paramagnetic species such as O₂ and free radicals can enhance T₁ and T₂ relaxation times. If the change in relaxation time is sufficiently large, contrast will be generated in magnetic resonance images. Since radiation is known to be capable of altering the concentration of O₂ and free radicals during water radiolysis, it may be possible for radiation to induce MR signal change.

Purpose:

We present on the first reported instance of x-ray-induced MR signal changes in water phantoms and investigate potential paramagnetic relaxation enhancement mechanisms associated with radiation chemistry changes in oxygen and free radical concentrations.

Methods:

Images of water and 10-mM coumarin phantoms were acquired on a 0.35-T MR-linac before, during, and after a dose delivery of 80 Gy using an inversion-recovery dual-echo sequence with water nullified. Radiation chemistry simulations of these conditions were performed to calculate changes in oxygen and free radical concentrations. Published relaxivity values were then applied to calculate the resulting T1 change, and analytical MR signal equations were used to calculate the associated signal change.

Results:

Compared to pre-irradiation reference images, water phantom images taken during and after irradiation showed little to no change, while coumarin phantom images showed a small signal loss in the irradiated region with a contrast-to-noise ratio (CNR) of 1.0 – 2.5. Radiation chemistry simulations found oxygen depletion of −11 μM in water and −31 μM in coumarin, resulting in a T₁ lengthening of 24 ms and 68 ms respectively, and a simulated CNR of 1.0 and 2.8 respectively. This change was consistent with observations in both direction and magnitude. Steady-state superoxide, hydroxyl, hydroperoxyl, and hydrogen radical concentrations were found to contribute less than 1 ms of T₁ change.

Conclusion:

Observed radiation-induced MR signal changes were dominated by an oxygen depletion mechanism. Free radicals were concluded to play a minor secondary role under steady-state conditions. Future applications may include in vivo FLASH treatment verification but would require an MR sequence with a better signal-to-noise ratio and higher temporal resolution than the one used in this study.

Introduction

Image-guided radiation therapy (IGRT) is a key technique for achieving safe and accurate delivery of conformal dose distributions. Pre-treatment radiographic imaging allows for matching of patient anatomy to planning images, while fluoroscopic, infrared, and cine MRI, on the other hand, enable real-time monitoring of patient motion during treatment.¹ An enhancement to these current techniques would be the ability to differentiate between irradiated and unirradiated tissue either in real-time or immediately following treatment, effectively creating an image of the radiation beam path in the patient. A number of technologies are in various stages of development to aid in this goal. For example, researchers have demonstrated that radioactivity induced by proton beams can be imaged with positron emission tomography (PET) or prompt gamma spectroscopy during treatment.^2,3 Other groups are working toward using radiation-induced thermoacoustic effects for in vivo dosimetry and range verification.^4,5 One technique currently being translated into clinical practice is Cherenkov imaging, which uses cameras to monitor for light generated in irradiated areas of the patient’s skin by high-energy electrons.⁶ These techniques, which have important limitations in energy thresholds, allow for visualization of radiation effects based on physical deposition of dose through induced radioactivity, thermoacoustic effects, and Cherenkov radiation rather than changes in radiation chemistry. To date, no technique has been developed for real-time in vivo radiation visualization based on radiation chemistry changes.

A recent innovation relevant to this discussion is the integration of MRI with linac technology in the form of MR-guided radiation therapy (MRgRT). Introduced clinically in the early 2010’s, MRgRT devices are capable of simultaneous MR imaging and megavoltage x-ray radiation delivery.^7,8 MR imaging during treatment is typically acquired in a cine-mode, where a single slice or a few slices are acquired several times per second to monitor patient motion.⁹ For treatment sites where patient motion can be well-controlled, such as the brain, an opportunity exists to apply a wider range of MR sequences and techniques during radiation delivery, including those that could distinguish between irradiated and unirradiated regions.

One possible mechanism for MR imaging of radiation-induced effects may be in time-varying concentrations of paramagnetic species, namely oxygen and free radicals. These paramagnetic species enhance the relaxation rate of surrounding hydrogen nuclei via dipole-dipole coupling, altering the local T₁ relaxation time – a higher concentration shortens T₁, while a lower concentration lengthens T₁. This paramagnetic relaxation enhancement (PRE) effect is the same mechanism by which commonly used gadolinium-based contrast agents act, due to the seven unpaired electrons present on the chelated Gd³⁺ ion.¹⁰

In this work, we present the first reported MR images that have been directly attributed radiation chemistry changes in water solutions. Water-nullified images were acquired with an MR-linac before, during, and after irradiation of water phantoms. A simulation of radiation chemistry was performed to calculate anticipated changes in oxygen and free radical concentrations by numerically solving a system of ordinary differential equations describing reaction kinetics using Python’s SciPy package.¹¹ Relaxivity values were then used to evaluate how these simulation-determined changes could affect relaxation rates and signal intensity. Simulation results were compared with images acquired on a 0.35-T MR-linac.

Methods and Materials

Radiation Chemistry Simulation

The processes of radiation chemistry can be broadly divided into four temporal stages: physical (t = 0 – 10⁻¹⁵ s), physicochemical (t = 10⁻¹⁵ – 10⁻¹² s), non-homogeneous chemical (t = 10⁻¹² – 10⁻⁶ s), and homogeneous chemical (>10⁻⁶ s).¹² The result of the physical and physicochemical stages is formation of radicals, reactive species, and stable molecules in spurs and blobs along tracks. In the non-homogeneous chemical stage, diffusion and intra-track reactions proceed until a uniform spatial distribution along the track is reached. Non-homogeneous-stage chemistry can be simulated using Monte-Carlo-based computer models that incorporate diffusion and reaction probability distributions, with the output being “escape yields” (G-values), that is, the number of each species remaining after the track structure has dissipated per 100 eV deposited.¹³ Escape yields, which are a complex function of linear energy transfer (LET), dose rate, and target molecular composition, can then be used in a conventional homogeneous-stage reaction kinetics simulation to calculate the time-evolution of species concentrations beyond 1 μs.

For the experimental component of this study, we used a 0.35-T MR-linac with 6-MV x-ray beam manufactured by ViewRay (ViewRay Technologies Inc, Oakwood Village, Ohio). The linear accelerator of this machine nominally supplies beam pulses lasting 5 μs, with 7.7 ms between pulses. This was confirmed by monitoring RF and electron gun waveforms with an oscilloscope. Because the beam-on time for each pulse is >1 μs, a homogeneous-stage radiation chemistry simulation was used to calculate the concentration of paramagnetic species as a function of time.

In this homogeneous-stage radiation chemistry simulation, there were a total of m chemical species and n reactions, with C_i denoting the concentration of the i^th species for i = 1,2, …, m, k_j denoting the reaction rate constant of the j^th reaction for j = 1,2, …, n, and a_ij and b_ij denoting stoichiometric coefficients for reactants and products respectively, as shown in Eqn. 1a. Reaction rates, r_j, were calculated according to Eqn. 1b. The rate of change of each species depended on both radiation action and conventional reaction kinetics, as reflected in Eqn. 1c.

$$\sum _{i=1}^m{a}_{ij}{C}_i \stackrel{k_j}{\to } \sum _{i=1}^m{b}_{ij}{C}_i\text{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}for} j=1,2,\dots ,n$$ (1a)

$$r_j=k_j\prod _{i=1}^m{C}_i^{a_{ij}} \text{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}for} j=1,2,\dots ,n$$ (1b)

$$\frac{d{C}_i}{dt}=G_i\frac{dD}{dt}\rho +\sum _{j=1}^n\left(b_{ij}-a_{ij}\right)r_j\text{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}for} i=1,2,\dots ,m$$ (1c)

On the left-hand side of Eqn. 1c, dC_i/dt is the rate of change in concentration for the i^th species. On the right-hand side, the first term represents production or depletion due to direct radiation action, where G_i is the escape yield for the i^th species, dD/dt is the dose rate, and ρ is the mass density of the medium. The remaining terms in the summation reflect the reaction kinetics, as governed by the law of mass action.¹⁴ The negative sign in front of the stoichiometric coefficient a_ij denotes consumption, while the positive sign in front of the stoichiometric coefficient b_ij denotes production. This system was solved using the solve_ivp function in Python’s SciPy package, implemented using code adapted from a similar radiation chemistry simulation.^11,15 Figure 1 shows a subset of the system of differential equations used in this simulation. The 79 reactions included in this simulation, along with the reaction rate constants, were drawn from another radiation chemistry study and are listed there for reference.¹⁶ No uncertainties were provided for these reaction rate constants.

FIGURE 1. Diagram of simulation workflow and measurement setup.

A homogeneous-stage reaction kinetics simulation was performed to calculate anticipated changes in oxygen and free radical concentrations. Relaxivity values were then used to evaluate how these simulation-determined changes could affect relaxation rates and signal intensity. Simulation results were compared with MR-linac images acquired before, during, and after irradiation.

Escape yields for this study, given in Table S1 of the supplementary material, were drawn from a recent paper in which the authors used a Monte-Carlo-based computational model to simulate non-homogeneous stage chemistry, with a specific focus on how escape yields varied as a function of dissolved oxygen concentration and LET.¹⁷

The initial concentration of dissolved oxygen in our water phantoms was calculated using Henry’s law, [O₂] = k_HP, where [O₂] is the concentration of dissolved oxygen, k_H is Henry’s constant, and P is the partial pressure of O₂. Under atmospheric equilibrium (pO₂= 0.21 atm) and a temperature of 25°C, the dissolved oxygen concentration was calculated to be 0.277 mM.¹⁸

Coumarin and coumarin-3-carboxylic acid (C3CA) were incorporated into the simulation as scavengers to determine how the steady-state hydroxyl radical concentration changed. Coumarin was chosen because it has a high reaction specificity to the hydroxyl radical. Rate constants for reactions with coumarin and C3CA have been established by other groups using pulse radiolysis techniques.^19–21 A comparison of simulation results with experimental data from four studies showed good agreement and is available in the supplementary material Fig. S2.^22–25

Radiation chemistry simulations in water and 10-mM coumarin were run for 26.5 minutes to cover a 30-second pre-irradiation and two ~13-minute image acquisitions, the first occurring during irradiation, and the second occurring after irradiation.

The goal of this simulation was to provide estimates of the steady-state concentrations of each paramagnetic species, for the purpose of identifying candidates for generating MR contrast. To meet this goal and avoid unnecessarily complicating the simulation, prior published and established reaction rate constants and escape yields were assumed and output of our simulations were compared against other prior published results to verify we achieved a reasonable concentration change associated with each species. No uncertainties were assigned to the reaction rate constants or escape yields within the simulation. Consequently, simulation results from this deterministic model are reported without uncertainty values. Such an approach has also been taken by other authors performing similar radiation chemistry simulations.^15,16,26,27 Techniques for assigning uncertainties to rate constants and propagating errors through to simulation results have been made possible by improvements in computational power.^28,29 However, implementing these techniques requires technical expertise and effort that was beyond the scope of this initial, multi-disciplinary work. Should the results of this present work warrant future advancement, it may be appropriate to implement a quantitative uncertainty analysis for radiation chemistry simulation results.

MR Physics Simulation

The relationship between paramagnetic species concentration and the relaxation rate R₁ (=1/T₁) is described in Eqn. 2, where R_1,0 is the initial relaxation rate, r_i is the relaxivity of the i^th paramagnetic species, and ΔC_i is the concentration change of the i^th paramagnetic species relative to unirradiated conditions. When multiple paramagnetic species are present, each contributes to the total R₁ change additively, as the summation in Eqn. 2 reflects.³⁰ In the context of the present radiation chemistry work, R_1,0 is the relaxation rate before irradiating, and R₁ is the relaxation rate after irradiating.

$$R_1=R_{1,0}+ \sum _i{r}_{1i}\Delta C_i$$ (2)

The relaxivity of O₂ has been reported to range from 0.08 – 0.50 s⁻¹ mM⁻¹, depending on the B₀ field strength, temperature, and surrounding medium (e.g., water, saline, plasma, vitreous).³¹ For this study, the empirical model published by Bluemke et al. was used to calculate an O₂ relaxivity of 0.32 s⁻¹ mM⁻¹ for a B₀ field strength of 0.35 T.³¹ The reported uncertainty for this relaxivity value is approximately 20%. The relaxivity of the superoxide radical, ${\text{O}}_2^{•-}$, has been reported to be similar to that of O₂, with a value of 0.29 s⁻¹ mM⁻¹ in water at 9 T and 20° C.³² In stark contrast, the relaxivity of hydroxyl radical, ^•OH, has a reported value of 3.4 × 10⁷ s⁻¹ mM⁻¹ in egg white at 9.4 T and 21° C, which is eight orders of magnitude higher than the reported relaxivity of O₂, ${\text{O}}_2^{•-}$, and typical gadolinium-based contrast agents.³³ Other radicals produced by water radiolysis, including hydrogen radical, H^•, and hydroperoxyl radical, ${\text{HO}}_2^{•}$, do not currently have published relaxivity values. For analysis in this study, the relaxivities of H^• and ${\text{HO}}_2^{•}$ were assumed to be equal to that of ${\text{O}}_2^{•-}$.

T₁ relaxation change due to O₂ and free radicals was calculated using Eqn. 2 by time-averaging concentrations over the length of the MR acquisition. The concentration of some radicals, such as ^•OH, were found to be markedly higher during the beam pulse. To mitigate the effects of this, a long TR was chosen so that >100 beams pulses would be contained in a single repetition cycle. A short TR of <7.7 ms may have confounded data interpretation, as some k-space lines would have been acquired under different free radical conditions.

Signal calculations for the chosen inversion-recovery dual-echo sequence were completed using Eqn. 3, where S is the signal magnitude, K is the signal proportionality constant, TI is the inversion time, TR is the repetition time, T₁ is the longitudinal relaxation time, and TE_last is the echo time of the final echo.^34–36 Note that M₀, the T₂-decay term $e^{-TE/T_2}$, and coil sensitivity are all absorbed into the signal proportionality constant, K.

$$S=K\left(1+e^{-\left(TR-T{E}_{\mathit{\text{last}}}\right)/T_1}-2e^{-TI/T_1}\right)$$ (3)

The T₁ relaxation time for unirradiated water phantoms, denoted as T_1,0, was found by measuring signal from the phantom at varying inversion times. Equation 3 was then fit to this data using a least squares method, where T_1,0 and K were treated as free parameters. The average T_1,0 calculated using both echoes was T_1,0 = 2583 ± 11 ms. The proportionality constants for each echo were K₁ = 1277 ± 8 and K₂ = 1031 ± 7 for the 1^st and 2^nd echo respectively. Signal was found to be mostly linear with respect to inversion time in the measured range, especially for small variations around the nullifying TI value. R² values for fits to the 1^st and 2^nd echo data were both >0.99. A plot of this fitted data is available in the supplementary material Fig. S3.

MR-Linac Images

Water phantoms consisted of a 17-cm-diameter plastic container filled with approximately 1 liter of distilled water or coumarin solution. All solutions were prepared using distilled water passed through a Milli-Q Advantage A10 filtration system.

Images were acquired using an inversion-recovery dual-echo sequence for a single 2-cm-thick axial slice centered at d_max = 1.6 cm and a long TR of 2000 ms. Two spin-echoes at TE = 103 ms and TE = 208 ms were applied to monitor for changes in T₂. An inversion time of TI = 745 ms was chosen such that the signal from unirradiated water was nullified or nearly nullified, that is, producing little to no signal. Nullification of signal was confirmed by comparison with a region of interest encompassing only the air surrounding the phantom. The single scan acquisition time was 3 minutes 12 seconds for a resolution of 1.3 mm ×1.3 mm × 20 mm (168 mm field of view, 128 base resolution). Images were reconstructed as “Real” images, as opposed to the standard “Magnitude” reconstruction, which allowed for retention of the sign of the MR signal. This scan was repeated four times per image and averaged to reduce noise, bringing the total acquisition time to 12 minutes 48 seconds, as shown in Fig. 1. An inversion-recovery dual-echo sequence was chosen for this study because it is a simple sequence with a well understood contrast mechanism and signal equation. The standard sequences used clinically during treatment, the balanced steady-state free precession (bSSFP) or TrueFISP sequence, have a mixed T₁/T₂ contrast mechanism and have not previously been observed to show contrast under irradiation.

Radiation was delivered using the MR-linac’s 6-MV flattening-filter-free photon beam. Immediately prior to starting each image acquisition, a 30-second pre-irradiation was delivered with the intention of allowing radical concentrations to approach a steady state. Radiation was delivered for the duration of the 12 minute 48 second scan, bringing the total beam-on time to 798 ± 5 seconds. Given a dose rate of 615 ± 5 MU min⁻¹, the total number of monitor units delivered was 8180 ± 120 MU. Linac calibration was such that 1 MU = 1 cGy at a depth of d_max = 1.6 cm in water and a field size of 10 cm × 10 cm. Calibration accuracy was confirmed by monthly quality assurance measurements. Since the ViewRay MR-linac does not have a physical jaw, field size is only defined by the MLC. As such, the field size can only be set in specific increments dictated by the MLC design. A square field size of 6.6 cm × 6.6 cm was chosen so that the entire field would easily fit within the cylindrical phantom. The output factor for a 6.6 cm × 6.6 cm field size was 0.970. After applying the linac calibration and the output factor, the total dose delivered at a depth of d_max = 1.6 cm was 79.3 ± 1.2 Gy.

A substantial artifact running through the midline of the image was present in the first echo (TE = 103 ms) of this dual-echo sequence. This artifact was inherent to this dual-echo MR sequence due to the developmental nature of some of the sequences available on the ViewRay Siemens-based MRI console system. Sequences used clinically do not exhibit this artifact, nor did a single-spin-echo sequence with the same TE.

Standardized regions of interest (ROIs) were established and applied to each image to obtain signal values. The primary ROI was defined as all water inside the irradiated 6.6 cm × 6.6 cm field. An additional ROI of water outside the irradiated area was used to verify consistent performance of the image sequence. An ROI of air outside the phantom was used to estimate image noise. The average signal value inside these ROIs was recorded for each image. Experiments were repeated 5 times. From the 5 image sets, signal averages and standard deviations, corresponding to the variation across these 5 measurements, were calculated for each ROI.

Results

Water

Simulation results are presented in Fig. 2. The hydroxyl radical, ^•OH, reached steady-state behavior approximately 30 seconds after beam-on, as seen in Fig. 2c. The average ^•OH concentration was used to estimate T₁ change, as approximately 260 beam pulses were delivered within a single TR of 2000 ms. The integral-average ^•OH concentration over the course of the ~13-minute beam-on acquisition was 20 pM. According to the published relaxivity value for ^•OH, an average concentration of 20 pM should have resulted in a T₁-shortening of 1624 ms, which should cause a large and visually observable signal change.³³ Following beam-off, the concentration of ^•OH dropped to less than 0.1 pM within 3 ms according to our simulations (Fig. 2d).

FIGURE 2. Radiation chemistry simulation for distilled water and 10-mM coumarin under atmospheric equilibrium.

(a) Concentrations before, during, and after irradiation. Bar above each plot indicates when beam is on. Radical concentrations fall quickly after beam turns off. (b) First 300 ms of beam-on in water. (c) First 30 seconds of beam-on in water. After approximately 30 seconds, radicals have approached steady-state concentrations. (d) Zoom in on final two beam pulses in water. Hydroxyl radical concentration rises quickly during each 5 μs beam pulse, then falls to less than 0.1 pM within milliseconds following each beam pulse. (e) Depletion of O₂ during beam-on in both water and 10-mM coumarin.

Within the first several seconds of beam-on, ${\text{O}}_2^{•-}$ grew quickly as solvated electrons reacted with O₂, before reaching a steady-state concentration of 0.3 μM, as shown in Fig. 2c. According to our simulation and to the published relaxivity value for superoxide radical, this steady-state concentration of ${\text{O}}_2^{•-}$ would produce a T₁ change of less than 1 ms, which was likely undetectable for the MR sequence parameters used in this study due to image noise. Beyond the first several seconds, the larger change in O₂ concentration begins to dominate the overall T₁ change and MR contrast. Consequently, superoxide radical, ${\text{O}}_2^{•-}$, would have been most likely to play a role only for short acquisitions, due to the nonlinear rate of formation.

Following beam-off, concentrations of radical species quickly fall while the stable molecular products H₂O₂, H₂, and O₂ persist without change, as shown in Fig. 2a. Among these stable molecular products, only O₂ is paramagnetic. It follows that the relative change in T₁ after beam-off was primarily dictated by how much O₂ had been consumed. Fig. 2e shows that simulated oxygen consumption during an 80-Gy delivery at 0.1 Gy s⁻¹ was linear with respect to time, which is consistent with measurements reported by others.^37,38 The rate of oxygen depletion in this simulation with respect to dose was calculated to be −0.14 μM Gy⁻¹. This value lies at the lower end of reported experimental measurements for low LET radiation at this dose rate, which range from −0.14 μM Gy⁻¹ to −0.22 μM Gy⁻¹.³⁷ The simulated change in O₂ concentration after an 80-Gy delivery was −11 μM, though experimental measurements suggest that this change could have been as large as −18 μM. Using a literature-reported relaxivity value, a change in O₂ concentration of −11 μM to −18 μM would be sufficient to induce a small increase in T₁ of 24 ms to 38 ms.³¹ The approximately 30% uncertainty in the actual oxygen depletion rate and the approximately 20% uncertainty in the relaxivity value for O₂ are expected to dominate the overall uncertainty in the simulated signal change.

MR-linac images of distilled water acquired during and after the delivery of 80 Gy showed no signal change in the irradiated region. Based on relaxivity values currently available in the literature, this result contradicted the expectation that there would be a large T₁ shortening during beam-on due to ^•OH, and a small T₁ increase following beam-off due to O₂ depletion, as shown in Table 1. Based on this observation, it may be the case that the relaxivity of ^•OH reported in the literature was overestimated by at least two orders of magnitude.

Table 1: Simulated signal change for a water medium during beam-on using different relaxivity values for ^•OH.

Measured signal change in water was −0.4 ± 0.2 for TE = 103 ms images and −0.3 ± 0.2 for TE = 208 ms images. Simulated signal change, $\Delta S_{sim}$, calculated using the literature relaxivity value for ^•OH was inconsistent with this measured signal change. If an ^•OH relaxivity value similar to O₂ or ${\text{O}}_2^{•-}$ is substituted, the simulated signal change is much closer to measurements.

Species	Concentration change, ΔC (mM)	Literature relaxivity values				Substituted relaxivity for •OH
		r1 (mM−1s−1)	r1ΔC (s−1)	ΔT1 (ms)	ΔSsim	r1 (mM−1s−1)	r1ΔC (s−1)	ΔT1 (ms)	ΔSsim
O2	−6.0 × 10−3	3.2 × 10−1	−1.9 × 10−3	−1624	299 241	3.2 × 10−1	−1.9 × 10−3	12.3	−0.5 −0.4
${\text{O}}_2^{•-}$	2.8 × 10−4	2.9 × 10−1	8.0 × 10−5			2.9 × 10−1	8.0 × 10−5
${\text{HO}}_2^{•}$	2.2 × 10−7	2.9 × 10−1	6.4 × 10−8			2.9 × 10−1	6.4 × 10−8
•OH	1.9 × 10−8	3.4 × 107	6.6 × 10−1			2.9 × 10−1	5.6 × 10−9
H•	1.6 × 10−13	2.9 × 10−1	4.5 × 10−14			2.9 × 10−1	4.5 × 10−14

One possible explanation for the absence of observable signal change in distilled water following beam-off is that the signal change was too small to be detected. Image noise measured using a region of interest confined to the air outside the water phantom found an average noise level of approximately 1, meaning that the measured contrast-to-noise ratio (CNR) for our water phantom images was less than 1. According to Eqn. 3, a signal change of −0.5 for the 1^st echo was expected for ΔO₂ = −11 μM (ΔT₁ = 24 ms). Given the level of noise in the images, it is likely that any signal change due to O₂ depletion was simply too small to generate visible contrast. It may also be possible that the time resolution of our MR acquisition was too low to observe T₁ changes resulting from the increase in free radical production that occurs within the initial seconds following beam-on.

Coumarin Solution

To isolate any potential paramagnetic relaxation enhancement effects due to ^•OH, a scavenger was introduced to lower the steady-state ^•OH concentration during beam-on. Coumarin was the scavenger of choice to accomplish this, primarily because it has been studied for decades as a chemical probe for ^•OH.^22–24 When coumarin reacts with ^•OH, one of the reaction products is fluorescent 7OH-coumarin. For this reason, rate constants of coumarin with ^•OH and other radicals have been measured by means of competition kinetics and pulse radiolysis, enabling it to be modelled accurately.^19–21 Mechanistic studies of coumarin have also concluded that O₂ is consumed during the formation of hydroxylated coumarin.²⁰ Thus, the paramagnetic relaxation effects of adding coumarin are two-fold – the steady-state concentration of ^•OH is lowered, and the rate of O₂ depletion is raised.

To maximize scavenging effects, a 10-mM coumarin solution was prepared. Simulation indicated that in this coumarin solution, the average ^•OH concentration during beam-on would be 1 × 10⁻¹⁶ mol L⁻¹, approximately three orders of magnitude lower than in water. The O₂ depletion rate in simulation of irradiated coumarin solution was found to be −0.38 μM Gy⁻¹, which was more than double the depletion rate calculated for water alone. For a simulated delivery of 80 Gy, the total O₂ depletion was −31 μM, enough to increase T₁ by 68 ms. For the inversion-recovery dual-echo sequence used in this study, a T₁ increase of this size would be expected to result in a signal change of −2.8 for the 1^st echo and −2.2 for the 2^nd echo. Tabulated simulation results for water and coumarin during both beam-on and beam-off acquisitions are shown in Table 2.

TABLE 2: Comparison of simulated signal with measured signal.

Measured signal change is defined as the difference in mean signal from a region of interest defined by the 6.6 cm × 6.6 cm square field, relative to the reference image taken prior to radiation. Measured signal change uncertainty is the standard deviation for n = 5 images. A value of 0.32 s⁻¹ mM⁻¹ was used for oxygen relaxivity and a value of 0.29 s⁻¹ mM⁻¹ was used for all radical relaxivities. For ΔS and ΔS_sim, rows correspond to 1^st and 2^nd echoes respectively.

Condition	Measured ΔS	Simulated
	TE = 103 ms TE = 208 ms	Species	ΔC (mM)	r1ΔC (s−1)	ΔT1 (ms)	ΔSsim
Water, beam-on	−0.4 ± 0.2 −0.3 ± 0.2	O2	−6.0 × 10−3	−1.9 × 10−3	12.3	−0.5 −0.4
		${\text{O}}_2^{•-}$	2.8 × 10−4	8.0 × 10−5
		${\text{H}\text{O}}_2^{•}$	2.2 × 10−7	6.4 × 10−8
		•OH	1.9 × 10−8	5.6 × 10−9
		H•	1.6 × 10−13	4.5 × 10−14
Water, beam-off	−0.1 ± 0.3 −0.1 ± 0.4	O2	−1.1 × 10−2	−3.6 × 10−3	24.5	−1.0 −0.8
		${\text{O}}_2^{•-}$	3.8 × 10−5	1.1 × 10−5
		${\text{H}\text{O}}_2^{•}$	2.8 × 10−8	8.0 × 10−9
		•OH	5.9 × 10−14	1.7 × 10−14
		H•	9.7 × 10−19	2.8 × 10−19
10-mM coumarin, beam-on	−2.2 ± 1.0 −1.0 ± 0.6	O2	−1.6 × 10−2	−5.3 × 10−3	34.4	−1.4 −1.1
		${\text{O}}_2^{•-}$	5.9 × 10−4	1.7 × 10−4
		${\text{H}\text{O}}_2^{•}$	7.0 × 10−7	2.0 × 10−7
		•OH	2.1 × 10−11	1.5 × 10−11
		H•	5.3 × 10−16	1.5 × 10−16
10-mM coumarin, beam-off	−1.7 ± 0.8 −0.9 ± 0.6	O2	−3.1 × 10−2	−1.0 × 10−2	68.3	−2.8 −2.2
		${\text{O}}_2^{•-}$	4.8 × 10−5	1.4 × 10−5
		${\text{H}\text{O}}_2^{•}$	3.5 × 10−8	1.0 × 10−8
		•OH	8.9 × 10−14	2.6 × 10−14
		H•	1.9 × 10−18	5.4 × 10−19

MR-linac images of coumarin solution acquired during and after the delivery of 80 Gy showed a small decrease in the average signal within the irradiated region, as shown in Fig. 3. Changes in measured signal relative to the reference scan taken before irradiating are presented in Table 2, where the uncertainty reported is the standard deviation across 5 acquired images. CNR ranged from 1.0 to 2.5 for these scans. A small change in signal was observed between the two echoes at TE = 103 ms and TE = 208 ms, as can be seen in Table 2. This could be indicative of a modest T₂ PRE effect as all paramagnetic materials should affect both T₁ and T₂ relaxation rates.

FIGURE 3: Images of distilled water (top) and 10-mM coumarin solution (bottom) showing signal change in irradiated region.

Red square indicates the 6.6 cm × 6.6 cm irradiated region. Total dose delivered was 80 Gy. A prominent artifact runs through the midline of the TE = 103 ms images.

Discussion

The images in Fig. 3 are, to our knowledge, the first reported images demonstrating the feasibility of using x-rays to directly induce MR signal change in a water phantom and, perhaps most importantly, the first demonstrated MR images directly attributed to time-dependent radiation chemistry.

While free radical production may have influenced contrast, it is likely that the observed signal change was primarily due to an oxygen depletion mechanism. This assertion is supported by our simulation results which show that radical concentrations drop within seconds of the beam turning off, as shown in Fig. 2a. Consequently, any signal change attributable to radicals should not appear in images taken minutes after beam-off. Our observation of signal change persisting for minutes after beam-off therefore indicates that radicals did not cause this signal change. Instead, this persistence suggests that an O₂ depletion mechanism is plausible, given that local O₂ concentrations are not expected to change following beam-off (Fig 2e).

Furthermore, simulation results in Table 2 show that O₂ was the paramagnetic species with the largest absolute concentration change over the 80-Gy delivery. By comparison, the absolute change in radical concentrations were at least an order of magnitude lower and therefore would have only a small contribution to the signal change, assuming the relaxivities of all radical species were on the order of 0.29 s⁻¹ mM⁻¹.

Comparing the simulated signal change with the observed signal change in Table 2 shows reasonable agreement. The direction of signal change in all images is negative, which is consistent with the T₁ lengthening that would be induced by O₂ depletion. The magnitude of the simulated signal change agrees well with measurements for beam-on images, but for beam-off images the simulation predicts a larger signal change than was observed. The large relative uncertainty in the measured signal change complicates this comparison. Boosting the signal-to-noise ratio (SNR) by using a larger B₀ field or a larger voxel size may help to reduce the relative uncertainty. Another complicating factor in the comparison of simulated results to measured results is that water from the irradiated area diffused into unirradiated areas, as shown in Fig. 3, which may have been due to thermophoresis along a temperature gradient set up by the RF coil arrangement. This was not accounted for in the simulation as our simulations assume spatial homogeneity. An attempt was made to define a smaller region of interest in the irradiated area to help control for this effect, but the relative uncertainty remained large due to the non-uniform signal change within the irradiated area, which may have been a result of the low SNR. For an example of this non-uniform signal change, see the image in Fig. 3 of 10-mM coumarin during beam-on, TE = 208 ms. Finally, there is the potential that free radical production during beam-on may have a larger contribution to the signal change than modeled, potentially indicating either greater free radical production than simulated or larger relaxivity values than used. As relaxivities tend to increase with decreasing field strength, it is likely that the literature relaxivity of the superoxide radical used in this study was less than the actual relaxivity value at B₀ = 0.35 T.

Additional support for the hypothesis that O₂ depletion was the dominant mechanism responsible for the observed signal change is that images acquired with an elevated level of dissolved O₂ also showed a visible change in signal in the irradiated region (Fig. 4). Due to a lack of radiation chemistry literature contemplating dissolved O₂ levels elevated beyond atmospheric equilibrium, this setup was not simulated. Interestingly, this finding poses the possibility that imaging results would also be dependent on the initial oxygenation levels, as in normoxia and hypoxia.

FIGURE 4: Images of distilled water with elevated dissolved oxygen showing signs of signal change in the irradiated region.

Dissolved oxygen level was raised by bubbling oxygen gas through the sample for several minutes before sealing the phantom with an airtight lid. Based on the altered TI_null of approximately 20 ms relative to atmospheric equilibrium, the dissolved oxygen concentration was estimated to be 0.32 atm (422 μM). Reference image acquired using a single acquisition in 3 minutes 12 seconds.

The primary challenge in detecting this radiation-induced signal change was achieving a sufficient contrast-to-noise ratio. To achieve the proof-of-concept images presented here, CNR was optimized in three ways. First, the dose delivered was raised to 80 Gy, resulting in more O₂ depletion and a larger T₁ change. Second, coumarin was added to distilled water, boosting O₂ depletion. Third, noise was lowered by repeating the sequence four times and averaging the resulting images. This required long scan times that limited the temporal resolution.

With these first images and a potential mechanism to explain the effect in hand, we turn our attention to assessing the feasibility of imaging radiation chemistry changes in vivo. There are two primary constraints to consider when dealing with living tissue, namely, dose and reoxygenation. Conventional radiation treatments use dose fractions of ~2 Gy, typically delivered at a rate of 6 Gy min⁻¹. As demonstrated in a recent study, this dose rate not high enough to deplete oxygen in tissue to a degree meaningful for the application under consideration, presumably because the rate of reoxygenation by blood supply exceeds the rate of oxygen depletion by radiation.

To meaningfully lower the oxygen concentration in tissue, much higher dose rates such as those found in FLASH treatments are required. In the same study, researchers showed that a FLASH treatment of 20 Gy was sufficient to lower the oxygen concentration in tissue by 2 – 4 μM, with the maximum depletion level occurring immediately after the beam pulse.³⁸ Reoxygenation by blood supply restored normoxic conditions within 15 seconds. According to our water phantom results, an oxygen depletion of 16 – 31 μM was required to produce observable contrast changes in the irradiated area, as indicated in Table 2. Imaging smaller O₂ changes of 2 – 4 μM that persist for no more than 15 seconds would require the SNR to be improved by a factor of approximately 5 to 15 times and a higher temporal resolution. The current study was intended to explore the possibility of using x-rays to induce MR signal change and to demonstrate potential mechanisms that are within the realm of what is currently feasible on our MR-linac system. Unfortunately, this required raising the dose to 80 Gy and lowering the temporal resolution, both of which are contradictory to what would be needed for imaging in vivo FLASH O₂ depletion. Significant additional work would be needed to meet the SNR and temporal resolution requirements.

Several options for improving SNR are available. One option would be to raise the B₀ field strength, as SNR increases almost linearly with field strength. The Elekta Unity MR-linac, for example, has a B₀ field strength of 1.5 T. Raising the B₀ field from 0.35 T to 1.5 T should improve SNR by a factor of 4, which may in and of itself be sufficient. Another option to improve SNR is to increase the voxel size. For this study, the largest voxel size available for this specific pulse sequence on the ViewRay MR-linac was chosen. As our sequences on the ViewRay system are more experimental, it may be possible that future investigator-developed sequences may have more options, including alternative resolutions. Finally, higher SNR coils could be used. ViewRay has recently released a new head coil with improved SNR relative to current coils. Future studies should involve this head coil rather than the prior coils.

Improving temporal resolution will require a new MR sequence. Our current ~13-minute acquisition was chosen because it provided easy to interpret results and sufficient SNR for the purposes of this study. One option to improve temporal resolution would be to use a faster T₁-weighted or T₁ mapping sequence with greater scan acquisition efficiency, such as an industry standard Magnetization Prepared-RApid Gradient Echo (MPRAGE). As 2D MPRAGE sequences are very fast scans, it may be possible to image this effect at much higher temporal resolution. Another option is to consider that changes in oxygen concentration can also be measured by $T_2^{*}$ effects. An entire body of literature for functional MRI (fMRI) has been developed for blood oxygen level dependent (BOLD) MRI that may be of use. In the fMRI literature, the BOLD fMRI temporal resolution, using an echo-planar imaging technique, is often acquired in less than 2 seconds.

Conclusion

In this work, we have presented the first reported images demonstrating the feasibility of using x-rays to induce MR signal change in a water phantom due to radiation chemistry changes. Mechanisms involving oxygen depletion and radical production were investigated. Oxygen depletion was found to be the dominant mechanism and was largely consistent with the observed signal change. Potential application of this technique may include in vivo FLASH treatment verification. Realizing this potential would require MR sequences with higher temporal resolution and improvements to SNR.