Contemporary Applications of Cherenkov Imaging in Radiation Therapy

Cherenkov radiation (CR) is produced from all high energy radiation sources and is part of the dose delivery process in tissue. As such, CR is a direct indicator of the dose delivery process and in recent years the ability to image and measure CR has provided a number of ways to help with radiotherapy dosimetry and delivery tracking. This review provides an overview of the fundamental physical principles of CR production and the radiation transport in tissue, along with applications of imaging CR that have seen significant development in the past few years.


Introduction
CR has a broad spectrum of light emission that is from the UV region down to near the infrared region and is generated when a charged particle travels through a dielectric medium at a phase speed greater than the speed of light in that medium.The requirement on the phase speed of the charged particle sets the threshold condition as ௩ 1, and the emission angle between the CR wavefront and moving vector of the charged particle determined as ‫ߠݏܿ‬ = ௩ , where ߠ is the emission angle, ܿ and ‫ݒ‬ are the speed-of-light in a vacuum and the phase speed of the charged particle in the medium respectively, and ݊ is the refractive index of the medium.In liquid water, the threshold energy of Cherenkov radiation is around 0.26 MeV and emission angle about 41 ࡈ for electrons.The intensity and spectroscopic characteristics of CR are described by the Frank-Tamm formula , where ܰ is the number of Cherenkov photons, ݈ is the pathlength of the charged particle, ߙ is the fine-structure constant, ߣ is the wavelength and the angular frequency ߱ = ଶగ ఒ .Therefore, the spectrum of CR is proportional to domain and flat in the energy domain.The polarization properties of CR are described by Jelley et al [9] based on classically electro-magnetic theory.
Theoretical analysis has demonstrated that, under transient charged particle equilibrium, the local intensity of CR is strongly proportional to the dose deposited, for both x-ray and electron beams (Fig. 1, adapted from [7]) [7,[10][11][12].However, that proportionality in spatial domain can be complicated by many factors including the beam quality, spectral changes such as beam hardening and phase space properties of secondary charged particle [7], particularly the anisotropic distribution [13].Corrections or mitigation methodologies are therefore generally warranted to account for those factors in order to use the local CR intensity distribution as surrogate for dose.In transparent medium such as water, the anisotropy of Cherenkov emission can be addressed by applying correction factors based on Monte Carlo simulations of specific experimental setups [7,14].Additionally, Cherenkov emission can be converted to be more isotropic by doping the solution (water) with fluorophores that absorb CR and emit fluorescence [15].By choosing fluorophores with strong absorbance in the relative short wavelength range, large spectral separation between the absorption and emission peaks (to minimize self-absorption and emission), and high quantum yield, complex corrections due to the anisotropy of CR can be greatly mitigated to simplify measurements [15,16].However, note that even though the CR is largely absorbed, a mix of CR and fluorescence will be detected [16].A recent study also showed that extracting polarized signal in combination with Monte Carlo simulated corrections can help address challenges posed by the anisotropic angular distribution of CR for generalized measurements [17].In turbid medium, such as biological tissues or equivalent phantoms, the transport of CR depends on the absorption and scattering coefficients.Studies have shown that the sampling of CR in biological tissue is generally limited to the first several millimeters underneath the surface with the sampling sensitivity decreasing approximately exponentially as the depth increases [18][19][20][21].The angular distribution of CR emitted from the surface is close to the Lambertian distribution with minor deviations affected by the g factor in Mie scattering [21] due to the strong scattering in turbid medium,.Unlike the spectrum of native CR, the emission spectrum is typically peaked in the red wavelength region from biological tissue surface due to the strong absorption in the short wavelength range [19][20][21].
In the context of radiation therapy (RT), the observation of Cherenkov radiation has been reported by multiple groups utilizing various modalities, including megavoltage electron [22], x-ray [15,22,23], proton/particle beams [24] and brachytherapy isotope seeds with the emission energy above the threshold energy [13,[25][26][27].Due the relatively low intensity (on the order of 0.01-1 nW cm í2 per MBq g í1 for radionuclides, and 1-ȝ: FP í2 per Gy s í1 for conventional beams used for external radiotherapy) [7,28], CR was initially considered as noise in fiberoptics based scintillation dosimetry [29,30] which needed to be removed via spectral or temporal filtering.Technological advances in sensitive optical detectors such as PMT, avalanche photodiode and intensified cameras have enabled direct measurements of CR in RT [26,27,[31][32][33][34]. Fiber optic dosimeters, as point detectors or arrays, have shown promises for CR based dosimetry in phantom for various RT modalities [13,[25][26][27].Sensitive camera systems including intensified CCD and CMOS, have enabled real-time 2D capture of CR and spectroscopy when integrated with a spectrometer [34].To further increase the detectability and signal to noise ratio (SNR) with ambient light, gating in the time domain has been explored, particularly by synchronizing the intensifier to radiation pulses with a low duty cycle [35].Spectral filtering and the spectral optimization of ambient light have also been investigated to boost the detection of CR [36].With the theory correlating CR to dose being laid out and advances in detection technologies , the imaging of CR has emerged in RT over the past decade.Various applications have been explored, which can be roughly categorized as 1) dosimetry/QA in phantom, 2) chemical sensing, and 3) in-vivo imaging.
2. Dosimetry/QA in phantom By using camera-based systems, CR 2D projection imaging has been demonstrated to profile electron and x-ray beams in water [14,22].Tomography of CR has demonstrated the reconstruction of volumetric dose distributions [37,38] and been validated for quality assurance (QA) of conventional IMRT and VMAT plans [39,40].Given the versatile and non-contact nature of optical imaging, Cherenkov imaging by camerabased systems has been applied for routine machine QA [19,20,41], the optimization of total skin electron therapy (TSET) [42], commissioning, QA and the investigation of electron return effect on an MR-Linac [43][44][45][46][47].
Perhaps the greatest advantage of Cherenkov imaging is being able to achieve superb spatiotemporal resolution.In modern RT, advances in image-guidance and radiation delivery technologies allow the utilization of small beamlets to achieve optimized dose distributions with steep dose gradients, such as stereotactic radiation surgery (SRS), micro-and mini-beam RT and stereotactic body radiation therapy (SBRT) leading to challenges in dosimetry, beam characterization and QA [39,48,49].Optical imaging such as Cherenkov imaging provides viable solutions to characterizing small fields with sub-millimeter spatial resolution [50] in 2D via projection imaging (Fig. 2) and in 3D via tomography [38,51].
Recently, the temporal modulation of dose was explored to increase treatment efficacy in radiation oncology.RT with ultra-high dose rates (UHDR) termed as FLASH-RT, which uses average dose-rates upwards of 40 Gy/sec as opposed to dose rates of 0.01 Gy/s in conventional delivery, has shown improved therapeutic ratio via enhanced normal tissue sparing while retaining comparable tumor killing in translational studies.However, UHDR conditions cast unmet challenges to conventional dosimeters, particularly the independency on dose rate over a range of several orders of magnitude, high spatial and ultra-high temporal resolution [52] that are desirable for FLASH-RT.Cherenkov imaging in combination with fast electronics and feedback circuits has overcome those challenges under UHDR conditions and has been explored for dosimetry, monitoring, image-guidance, and control in FLASH-RT [53][54][55][56][57] in complement to traditional technologies.

Chemical sensing
Cherenkov spectroscopy and excited luminescence imaging have been explored in sensing chemical/physiological factors that are of interest in RT, particularly the tissue oxygenation which has a critical role to play in the efficacy and sensitivity of tissues to RT.Because of the invasiveness and complexity of conventional oxygen measurement techniques, the sensing of oxygenation in-vivo has remained difficult to find in routine use.Cherenkov emission spectroscopy, with spectral characteristics correlated to the tissue optical properties at various oxygenation levels, has been demonstrated to quantify and track tissue oxygen during RT non-invasively [58,59].It has been shown that CR generated during RT can stimulate various optical (fluorescence or phosphorescence) probes [60,61] that are biologically compatible for chemical sensing [62][63][64] and photodynamic therapy [65][66][67].By introducing optical molecular probes stimulable by CR in the medium, spectroscopy and lifetime measurements of the excited luminescence which is sensitive to the local oxygen partial pressure (pO2) has been applied for oximetry both in-vitro and in vivo [34,68].Diffuse optical tomography with the excitation pattern optimized by radiation beam has allowed 3D reconstruction of pO2 to a couple of cm in depth [51,69,70].This technique is suitable for preclinical studies with small animal and translational sites near the surface.
The sampling depth and resolution can be further improved by Cherenkov-excited luminescence scanned imaging (CELSI) utilizing 2-dimensional sheets of radiation to produce Cherenkov photons, which then excite luminescence of probes distributed in biological tissues (Fig. 3a).By measuring total luminescence signal and then taking into account prior information about position of the scanning beam, the distribution of optical signal in the direction of scanning may be recovered and reconstructed in 3D with high spatial resolution [1-3, 71].While current FLASH-RT studies focused on phenomenological observations have shown enhanced normal tissues sparing, termed the 'FLASH effect', the underlying mechanism is unknown.One of the initial most widespread hypotheses is that the effect is related to substantial oxygen consumption and depletion upon FLASH-RT.Like what has been demonstrated in conventional RT, most recently, Cherenkov excited luminescence lifetime imaging has been utilized for direct sensing of pO2 dynamics during FLASH-RT (Fig. 3b), suggesting that radiologically relevant levels of hypoxia due to oxygen depletion is unlikely to occur in bulk tissue due to FLASH irradiation [6].

In-vivo imaging
The imaging of CR in humans was first carried out in a trial for breast cancer patients receiving external beam RT [72].By synchronizing the camera to the short radiation pulses, ambient, room light was rejected and real-time, background-subtracted CR images have been acquired at the speeds > 10 frames per second.The shapes of imaged CR revealed field segments projected on the patient surface [72,73] while the intensity showed a good correlation with erythema observed on the skin.Soon after the proof-of-concept studies, CR imaging was explored as a method of image guidance to provide real-time assurance, motion tracking, post-treatment recording as well as treatment verification [74].The human trial has been expanded to other sites including TSET via the Stanford technique [75][76][77][78][79][80][81][82], head and neck treatment with VMAT [83], and frame-based intracranial SRS [84], where the purposes of CR imaging have been feasibility-driven and exploratory with qualitative analysis such as motion monitoring and the validation of coverage.
What remains challenging is the quantitative correlation between CR emitted from patient surface and the deposited dose, as the relationship is sensitive to multiple factors including tissue optical properties, beam energy and geometry, imaging angle as well as treatment modality [21].Progress has been made by correcting the in-vivo Cherenkov image with patient specific skin optical properties measured by Spatial   Frequency Domain Imaging (SFDI) [85].X-ray attenuation extracted from patients' simulation CT scans has shown good correlation with the optical absorption and thus has been explored as another promising correction factor correlating CR emission to the superficial dose distribution (Fig. 4) [86][87][88].In future work, by combining geometry related corrections such as the patient surface profile and imaging angle with additional optical corrections, such as that for skin pigmentation, the correlation between the Cherenkov intensity and superficial dose as well as the responses to radiation in the imaged regions can be improved.Despite the challenge of imaging superficial dose quantitatively, CR is a free signal that exists in the presence of any megavoltage RT, and comes at no additional dose or time to the patient.It has been exploited for many unique applications in the domain of quality assurance and quality improvement.In addition to the real-time motion monitoring and coverage validation, it has been shown that CR imaging can be used for the verification of field matching [89].The concept of utilizing an always-on CR imaging system for uninterrupted RT recording to support offline review has been realized by permanently mounting cameras in the therapy bunker.Such a prototype system is now commercialized, providing software and hardware in one solution to complement existing technologies such as surface guided radiation therapy [90].
In a yearlong review of over 600 patients imaged with this system, study members identified 9 treatment anomalies across different stages of RT including simulation, planning and treatment, that would be have gone otherwise unnoticed without CR imaging [91].While none of these incidents were deemed clinically significant by the treating physician upon review, visualizing the beam on a patient provides the unique ability to detect these types of cases.Future work is focused on developing an automated detection algorithm to catch treatment anomalies in real-time through Cherenkov image analysis [92].The increased scale of data which can be made available through always-on CR imaging should enable several important machine learning (ML) applications.Several ML applications which have already been demonstrated in other imaging domains and have high potential in CR imaging include: deep image denoising [93,94], image registration and motion estimation [95], anomaly detection [96,97], and in vivo landmark segmentation (vasculature, for example) for automated patient alignment vertification [98].A novel three-channel camera that was time-gated was used to image color Cherenkov emissions from patients during treatment.
Color shades of Cherenkov emission in radiotherapy may be used to interpret oxygen saturation, tissue blood volume and major vessels within the body [99].Given the nature of instant emission, CR imaging via fast electronics is suitable for ultrafast monitoring and recording of FLASH-RT delivery which typically takes less than tenth of a second.Figure 5 demonstrates imaging CR in real time of a UHDR electron FLASH delivery to a minipig's skin with optically clear bolus resolving the treated field [4,5].

Conclusion
In conclusion, through the review of contemporary applications of CR imaging in RT, the technology has matured over the past decade and is now readily transferable to clinics for scaled and diversified investigations.Despite the challenge in correlating CR intensity to absolute dose, new avenues are being created for CR imaging in emerging and advanced RT regimens, including FLASH-RT.

Acknowledgements
This work has been partially supported by NIH grant R01 EB023909 and the Dartmouth Cancer Center.

Figure 1 .
Figure 1.The proportionality between Cherenkov emission (a) and dose depositon (b) can be visually seen by comparing Monte Carlo point kernels for an 18 MeV x-ray beam in water.Adapted from [7].

Figure 4 .
Figure 4. Patient Cherenkov images (a) can be corrected for subsurface tissue optical properties using routine CT scans.Correct images (c) are better surrogates for the dose distribution (b).
on 3D and Advanced Dosimetry Journal of Physics: Conference Series 2630 (2023) 012011