1.1.

Value of Cherenkov

Delivery of radiation from a linear accelerator as electron or x-ray beams is now widely adopted throughout cancer medicine with extensive control systems and a range of technologically advanced delivery and measurement systems. There are $\sim 4000$ clinical linacs in the USA used for cancer treatment and $\sim \mathrm{11,000}$ throughout the world.² These devices are used to perform the majority of all radiotherapy, and $>50\%$ of all cancer patients receive it as part of their daily fractionated treatment course.³ The ability to deliver precision doses is thought to be exceptionally high, yet the reality is that delivery is not monitored at the patient. Direct validation of treatment doses is replaced by calculated estimates based upon calibration of the machine and computational treatment planning with verified patient setup. The value of Cherenkov imaging is that it shows a visualization of the dose on the patient’s skin,^4–6 providing a real-time intuitive display of these dynamic treatments.^7,8 The technology advances to capture this, and the implementation and interpretation of how these images are used is discussed here.

The observed value of Cherenkov has been in seeing delivery incidents that should be fixed or small errors that need to be compensated for.^7,8 An early prospective study showed 10% incident rate where Cherenkov could be utilized,⁷ although likely more realistic estimates throughout radiotherapy practice may be lower; however, because reliable tools to capture dose delivery do not currently exist, the true rate of errors is simply not well known. Most incidents are discovered through indirect means, and the investigation of them is time-consuming and expensive. Thus there is potential value in Cherenkov imaging for the fast, intuitive capture of all delivery to the patient. As the x-rays enter the tissue and induce soft electron collisions, the Cherenkov light is emitted and scattered within the tissue, and ultimately some fraction of it escapes from the tissue to be detected by the camera.⁹ The fact that tissue is translucent and emissive in the red to near-infrared wavelengths helps this process happen,¹⁰ as illustrated in Fig. 1.

Fig. 1

Image of x-rays entering tissue and the Cherenkov being emitted. Typical dose rate has Cherenkov radiosity of $\approx 1$ to $10\mathrm{nW}/{\mathrm{cm}}^2$ and an internal Cherenkov fluence within tissue of $\approx 1$ to $100\mathrm{nW}/{\mathrm{cm}}^2$.

1.2.

Engineering Ultra-Low-Level Optical Signal Detection

Cherenkov is an extremely low-light signal emitted during soft electron collisions in the dose deposition process. The signal is perhaps 1% of the total dose delivered, and in radiotherapy the typical dose delivery rate is $6\mathrm{Gy}/\min$ or $0.1\mathrm{W}/\mathrm{kg}$. A rough estimate of the Cherenkov yield might be $0.3\mathrm{mW}/\mathrm{kg}$ of tissue. Converting this to water and assuming the signal comes from about 1 cm of tissue, this yields a rough unattenuated emission estimate of $0.3\mu \mathrm{W}/{\mathrm{cm}}^2$.¹¹ Thus the signal is low, and capture of this signal from several meters away is reduced significantly by the near isotropic emission. Imaging this level of light signal in a room with lighting on is exceptionally challenging and requires care in optical filtering,¹² time-gating,¹³ and postprocessing,¹⁴ each of which are described below. The light signal is only able to be detected because the instantaneous Cherenkov light signal is above the room light’s instantaneous light signal, whereas the linear accelerator’s pulsing is off. The achievements developed for video-rate Cherenkov imaging are described here in terms of fast time-gated camera engineering, image processing, and associated optical and signal processing developments. Advances in optical engineering have been a critically important part of achieving Cherenkov imaging in practical radiotherapy use.

2.1.

Radiation Dose and Optical Cherenkov Emission

The local intensity of Cherenkov radiation can be estimated through the calculation:

However, the linearity between absorbed dose and detected Cherenkov emission is distorted in patient imaging due to various tissue optical factors.¹⁴ Using Monte Carlo simulations on flat and cylindrical skin-like phantoms, Zhang et al. determined difference ranges between Cherenkov emission and radiation dose. Entrance/exit geometry and tissue optical properties contributed the largest variations of 50% and 20% variance, respectively, with beam energy, tissue curvature, and field size also contributing uncertainty. Thus, quantitative spatial frequency domain imaging was used to gather accurate mapping of tissue optical interaction coefficients to correct subcutaneous vasculature, interstitial blood, and pigment. In clinical trials, calibrated corrections were able to reduce vasculature effects slightly from 22% to 6% in one region and from 14% to 4% in another.¹⁶

Another method of correlating observed Cherenkov to absorbed surface dose uses x-ray radiodensity for macroscopic tissue types.¹⁷ Using a linear correlation between dose-normalized Cherenkov intensities and average CT number (HU) per patient, the $R^2$ linear regression coefficient between Cherenkov intensity and absorbed surface dose increased from 0.67 to 0.85 for a 6 MV beam and from 0.91 to 0.95 for a 10 MV beam. Thus it may be possible to calibrate Cherenkov intensity to dose, although more extensive validation is yet to be completed.

2.2.

Intensity of Emission per X-Ray and Distance

The emission of Cherenkov light from both megavoltage beams and radioactive decay is known to be relatively dim.^11,18 During ${}^{18}\mathrm{F}$ decay, $<0.006\%$ of total energy released is Cherenkov emission, releasing $\sim 3\text{photons}$ per decay over the 250 to 800 nm range.¹⁸ Glaser et al. determined the Cherenkov radiation rate in biological tissues to be on the order of 0.01 to $1\mathrm{nW}{\mathrm{cm}}^{-2}\mathrm{per}\mathrm{MBq}{\mathrm{g}}^{-1}$. Additionally, the light fluence rate of Cherenkov radiation for radiotherapy beams in tissue was found to be 1 to $100\mu \mathrm{W}{\mathrm{cm}}^{-2}\mathrm{per}\mathrm{Gy}{\mathrm{s}}^{-1}$. In phototherapy applications, the total light fluence was on the order of $1\mathrm{nJ}{\mathrm{cm}}^{-2}$ for radionuclides and $1\mathrm{mJ}{\mathrm{cm}}^{-2}$ for radiotherapy beams.¹¹ Although phototherapy may not be possible due to low rates of Cherenkov light fluence, diagnostic applications are reasonable for Cherenkov excitation of molecular probes. Figure 2(a) displays typical detected Cherenkov fluence values compared to typical room lighting conditions.

Fig. 2

(a) Schematic of relative intensity of Cherenkov light, ranging with bright room light having the highest intensity of light, with Cherenkov emitted at tissue dimmer. (b) 1% intralipid and gelatin phantom with varying % blood concentrations in each column and $200\mu \mathrm{m}$ layers of varying melanin concentrations in each row. Detected Cherenkov due to irradiation by 6MV x-ray beam was imaged, and then normalized by the reflectance image in the gray channel. Displays altering intensity of Cherenkov detection due to blood and melanin concentration. Reprinted with permission from Ref. 14. (c) Increased blood concentration alters the pure $1/{\mu }^2$ Cherenkov spectrum by increasing absorption in the blue region while leaving the red region relatively stable. Reprinted with permission from Ref. 10 © Optica. (d) Increased blood oxygenation (${\mathrm{SO}}_2$) level (96%) increases the emitted spectrum in the red wavelength region Reprinted with permission from Ref. 10 © Optica. (e) Cherenkov emission spectrum shifts to be more blue with lower x-ray beam energy, and the overall signal increases due to deeper build-up region.¹⁹ (f) The spectrum is not significantly affected by beam type (electrons versus photons), but the overall yield is higher for electron beams.¹⁹

Real-time Cherenkov imaging requires optimization of the detection limits. Photon production, propagation, and detection are the key parameters.⁹ LaRochelle et al. determined that the distance from the source to the camera was the primary factor decreasing detection intensity due to the inverse square intensity dependence. The inverse squared dependence in detection dictates that the camera should be placed such that it is as proximal to the target as possible in order to maximize detected counts.⁹ Small camera to source distance, low $f$-number lens, and high quantum efficiency are required for sufficient light capture.

2.3.

Tissue Effects and Cherenkov Spectrum

Tissue light absorption strongly effects Cherenkov intensity.¹¹ In the absence of absorption, the fluence of Cherenkov light due to radionuclides is on the order of 5 to $500\mathrm{fW}{\mathrm{cm}}^{-2}\mathrm{per}\mathrm{MBq}{\mathrm{g}}^{-1}$ and 4000 to $8000\mathrm{nW}{\mathrm{cm}}^{-2}\mathrm{per}\mathrm{Gy}{\mathrm{s}}^{-1}$ for external radiotherapy beams.¹¹ Figure 2(b) visualizes the effect of tissue composition on detected Cherenkov intensity, where both hemoglobin and melanin were varied to simulate attenuation in human tissues.

In the absence of absorptive materials, optical Cherenkov radiation exhibits strong emission of blue wavelengths and the number of photons per wavelength is proportional to $\sim 1/{\lambda }^2$, as described by the Frank–Tamm formula.^9,10 The inverse squared relationship between wavelength and intensity in pure water spans from 300 to 1500 nm.¹⁶

At the location of dose deposition, the inverse square relationship holds, but as the light moves through the tissue, absorption and scatter impart spectral changes, modulating the signal.¹⁰ Specifically, the modulation arises from the optical absorption of water, lipids, and hemoglobin in tissue.^10,11 The exact shape of spectral distortion is dependent on hardware factors, such as beam profile and distance to the detecting fiber, as well as biological factors, such as hemoglobin concentration and oxygenation affecting specific wavelengths. Axelsson et al. found that the prominent change in the spectrum with varying hemoglobin concentrations occurs at wavelengths shorter than $\sim 630\mathrm{nm}$, seen in Fig. 2(c). An increase of hemoglobin oxygen saturation alters the spectral intensity resulting in a shift to the red region of the spectrum, demonstrated in Fig. 2(d).¹⁰ Similarly, Glaser et al. found that the fluence rate of emitted Cherenkov light in the presence of water, lipids, and hemoglobin was the strongest beyond 600 nm.

Changes to the spectrum of Cherenkov light emitted from tissue have additionally been investigated under varying beam type, incidence, energy, and field size using Monte Carlo simulation.¹⁹ Simulated beams of 6, 10, and 18 MV, 6 MeV, entrance and exit orientations, and field sizes of $1\times 1{\mathrm{cm}}^2$ up to $40\times 40{\mathrm{cm}}^2$ all show, effectively, the same spectral shape. When coupled to the quantum efficiency of red-optimized and green-optimized photocathodes, it was determined that there is a large increase in the detected Cherenkov with increasing field sizes, a slight increase in absorption with higher energy beams, and increased with electron beams compared to photon beams of the same energy.¹⁹ Figure 2(e) displays altering spectra according to beam energy, and Fig. 2(f) displays the varying spectra upon altering beam type.

3.1.

Time-Gating Approach

By exploiting the pulsed nature of clinical linacs with time-gated cameras for Cherenkov detection, the ambient room light signal can be decreased by 1000×.¹³ Typical linac photon and electron beams are delivered in 3 to $6\mu \mathrm{s}$ bursts with a repetition frequency of 60 to 360 Hz,²⁰ as illustrated in Fig. 3, correlating to a duty cycle of roughly 0.1%. By only exposing the Cherenkov camera sensor during the pulses, the ambient light signal is reduced to the same order of magnitude as the Cherenkov signal.¹³ Background images collected between pulses are normalized and subtracted from the pulse images to remove any remaining room light signal.²¹

Fig. 3

(a) The majority of the ambient light produced in the linac vault is continuous, and the duty cycle of the pulses is actually very low, around 0.1%, and so continuous imaging would largely just capture room light. (b) The timing scheme of a gated camera is shown, where a potential between the photocathode and the MCP input is applied during each trigger, leading to exposure only during the linac pulses.

In Ref. 13, the optical signal spectra from the Cherenkov emission was captured using an intensified charge-coupled device (CCD) camera (PI-MAX3 RB Gen II, Princeton Instruments, Acton) coupled to a 13 m optical fiber placed in the center of the linac beam. The PI-MAX camera’s acquisition was gated to the linac pulses through an analog trigger signal fed through a BNC cable from the linac. An ambient light spectrum was collected with identical acquisition timing without any linac pulses, which was then subtracted from the original beam gated spectra. Resulting data showed the spectrum of Cherenkov light, demonstrating the feasibility of imaging weak Cherenkov emission during radiation therapy while leaving the room lights on for patient comfort and safety.¹³

Early measurements utilized the Klystron voltage (KlyV) or the target current signal port (TargI) as direct electrical line triggering from the linac control unit for camera intensifier gating.²⁰ However, electronic feedback in the BNC cables along with manufacturer restrictions limit the practicality of wired triggering. Ashraf et al.²⁰ developed remote trigger gating, which leverages the scatter and leakage photon signal detected by a silicon photomultiplier (SiPM) tube coupled with a scintillator detector to gate the image intensifier, as illustrated in Fig. 4. The SiPM’s current signal was converted with a current-to-voltage amplifier to act as a trigger and was saturated, regardless of how far from the linac the camera was placed in the room.²⁰

Fig. 4

(a) The signal produced by the scintillator crystal coincidence system matches the time structure of the wired klystron signal coming from the linac. The success in linac pulse detection using the remote scintillator signal eliminates the need for a wired connection. (b) The scintillators detect leakage and scatter x-rays, emitting light detected by an SiPM.

The remote system resulted in a cumulative ROI intensity increase of roughly 1.5% over the cabled system and the average pixel intensity was found to be 0.9% greater on average for the remote triggering, increasing both the practicality of use in the clinical setting and Cherenkov signal.²⁰ The study also addressed trigger activation from other radiation sources (such as cosmic rays, neutron activated products, and scintillator radioactivity) by connecting two detectors for coincidence gating with an AND gate, effectively suppressing false positive beam detection.²⁰ Modern C-dose Cherenkov imaging cameras employ a similar remote triggering unit to distinguish the false scatter radiation events from the beam pulse.

3.2.

Spectral Filtering

Cherenkov imaging can be used in high-dose rate Brachytherapy QA²² and cyclotron-based proton treatments.²³ However, as these modalities have high repetition rates approximating continuous source treatments, it is not practical to gate the camera intensifier to the treatment duty cycle. Additionally, captured ambient room light remains a prevalent source of image quality degradation in EBRT. Thus spectral filtering light suppression might be necessary to improve Cherenkov imaging for continuous radiation treatments.

Rahman et al.¹² investigated the spectral properties of Cherenkov emission with various fluorescent and LED lights. Lights already present in the linac vault in the Dartmouth-Hitchcock Medical Center Oncology Department, including two fluorescent lights and an emitting tungsten source, with spectra shown in Fig. 5(a), were compared to various possible replacement lights.¹²

Fig. 5

(a) The camera generally used for Cherenkov imaging has a quantum efficiency weighted in the red wavelength region. This allows for the efficient collection of much of the ambient room lights. (b) Some replacement room light choices fall outside the range of the red wavelength collected by the camera. Reprinted with permission from Ref. 12.

Because the Cherenkov spectrum exiting patient tissue falls in the red wavelength range and the Cherenkov camera is optimized for this region,¹⁰ blue LED ambient room lights are optimal, as demonstrated in Fig. 5(b). However, this can be an uncomfortable lighting situation for patients. White light LED sources coupled with a 675 nm filter produces comparable results to a purely blue light with a color temperature of 6000 to 6500 K more closely matching the typical temperatures found in room lights (2500 to 6500 K).¹² LED lights can produce a flicker in real-time Cherenkov imaging due to the repetition rate of the linac and LEDs, although constant-current LED drivers can eliminate this.¹²

4.1.

Time-Gated Intensified Cameras

In vivo surface beam monitoring with Cherenkov emission requires optimized camera hardware for qualitative and quantitative images.²⁴ There are many options for cameras for low-light imaging, including sensors that are composed of (i) complementary metal-oxide-semiconductor (CMOS), (ii) CCD, and (iii) single-photon avalanche detector (SPAD) arrays. There are electron multiplying EM versions of the first two sensors, and both can be coupled with an image intensifier (II) at the front end. The choice of sensor and technique is driven by background light suppression and fast frame rate with the need for video rate acquisition. There is necessity to have amplification between the incoming light and the readout stage, accomplished using an image intensifier coupled CMOS or CCD camera.²⁴

This solution produces a functional system with many advantages. However, intensifiers have limitations in terms of moderate dynamic range and high cost. Additionally, these intensifier microchannel plates (MCPs) are limited in the maximum spatial resolution by the number of lateral channels, which is usually much less than the number of sensor pixels.²⁵ Thus, while intensified cameras are a valuable solution for low-intensity Cherenkov imaging, future innovations in optical sensors could lead to superior results or reduced cost (Fig. 6).

Fig. 6

Typical intensified camera interior. Cherenkov light is captured by the lens and focused onto the photocathode, which emits an electron image into the MCP through a low applied voltage over the photocathode and phosphor coated window. These electrons are accelerated through the MCP by the high voltage, bouncing off the dynode walls, and multiplying the electron signal within each channel. This multiplied electron image from each channel ends at a phosphor coated exit window, which emits phosphorescent optical photons. These luminescent photons are focused onto a sensor array (either CCD or CMOS) through either an imaging fiber taper or lens coupling, turning the optical signal back into an electron image with amplification in each pixel. The applied high voltage on the MCP is synced with the time-gating of the signal expected, such that when the linac beam is on, the MCP high voltage is applied, and when the linac beam is off, the MCP applied voltage is off, suppressing image capture during linac off periods.

4.2.

Single Photon Avalanche Diode and Quantum Sensors

A possible alternate single-photon-level sensor for this application is the SPAD, which have become increasingly popular for biophysics and biochemistry applications due to their single-photon sensitivity, picosecond response, and improved spatial resolution.^25,26 Unlike intensified CMOS and CCD cameras, large CMOS SPAD arrays demonstrate high timing performance with no global count limitations creating potential for multichannel single-photon counting with parallel readout and fast data processing.²⁵

Each SPAD pixel is a photodiode that is reverse biased above its breakdown voltage.²⁷ When incident on the active device area, photons produce electron–hole pairs triggering an avalanche of secondary carriers, creating a self-sustaining amplification, which is then quenched by lowering the SPAD bias voltage below the breakdown voltage.²⁷ Before the next photon can trigger an avalanche, the voltage is restored above the breakdown voltage.²⁷ Figure 7(a) demonstrates the function of an SPAD array. After sufficient exposure, each row is sequentially read into a subarray, erasing the memory of the previous row.²⁵ Thus, the internal sensor communicating the time between row readouts determines the overall readout speed.²⁵ Just as with intensified detectors, digital SPAD imagers use a time-gating technique to maximize light detection. Photons will only trigger an avalanche if they arrive during the open gate on time.²⁵

Fig. 7

Alternate potential solutions for time-gated multipixel sensor Cherenkov imaging. (a) Typical setup for a SPAD imaging sensor. The Cherenkov light is focused by the lens onto the SPAD array. In each pixel of the array, there is a reverse biased p-n junction on a photodiode. Photons incident into the active area produce electron-pairs, triggering an avalanche effect in the multiplication region of each pixel. After readout, the avalanche is quenched by lowering the reverse bias voltage below the breakdown voltage. (b) Typical QIS acquisition setup. As Cherenkov light moves through the lens, it is incident on the QIS array, where individual binary bit planes are obtained over some time. The individual bit planes are pixels and an SPAD array and represent the binary image at one point in time. As the individual bit planes are summed, a cumulative image is displayed, representing the entire temporal image that was obtained.

In order to improve image quality, specified by signal-to-noise ratio and dynamic range, there is a trade-off with the fill factor.²⁶ Factors to take into account for these cameras are the pixel isolation, pitch, fill factor, photon detection probability, photon detection efficiency, dark count rate, afterpulsing, and timing jitter performance.²⁶

More generally than the previous sensors, the concept of a quanta image sensor (QIS) was conceived in 2005 as the next major paradigm in solid-state sensors.^28,29 Through extreme miniaturization of pixel sizes in CMOS sensors, increased storage capacity, and expanded digital processing, the goal of QIS is to optimize single-photon event capture in ultralarge pixel arrays and provide images through intensive postprocessing. QIS’s comprise an array of subdiffraction limit binary sensor pixels. Each of these binary pixel values is read as a logical output of 1 or 0 corresponding to the presence or absence of a photon strike at a location in the incident field with a goal of 1000 fields read per second. Figure 7(b) displays a typical setup for a QIS acquisition. By locally processing the pixel “jot” data, postcapture at arbitrary pixel sizes, time-sequencing or integrations are all possible.²⁹ The small pixels provide increased resolution, extended low-light sensitivity, and enhanced photon-number-resolving capability over a standard CMOS camera.

Cherenkov imaging was tested with the concept of QIS using a SPAD sensor³⁰ where dose delivery was first imaged in water to validate the signal linearity. By gating the SPAD with the linac x-ray beam pulses, Cherenkov radiation was imaged from a patient during treatment, registering 1 to 2 photons per pixel above $\sim 1\text{photon}$ per pixel background.³⁰ Maximum Cherenkov radiance values were observed to be $\approx 3$ to $4\times {10}^9\mathrm{ph}·{\mathrm{sr}}^{-1}·{\mathrm{cm}}^{-2}·{\mathrm{s}}^{-1}$ for beams incident on a patient, corresponding to average radiant exitance of 2 to $3\mathrm{nW}/{\mathrm{cm}}^2$.³⁰ In theory, the SPAD could outperform the best ICMOS by a factor of 50 when comparing the noise power density for a given geometry.³⁰ However, while these SPADs can image Cherenkov with high sensitivity, the major limitation today is the low frame rate per linac pulse (4 to $5\mu \mathrm{s}$ each), which ultimately limits their use. This is because it means that relatively little of the available Cherenkov is captured in each linac pulse, ultimately lowering the detected signal because of camera deadtime. Thus the current time sequence of capture today would need to be improved for them to be viable options for Cherenkov imaging (Fig. 8).

Fig. 8

(a) Bright spots in Cherenkov images result from stray radiation coming off of the linac head, which can also lead to single-pixel readout errors manifesting as either random or permanent bright pixels salt noise. (b) Fixed pattern noise is a result of the detector architecture where all detector elements in a row share an ADC for readout. (c) Because of counting statistic properties, evenly irradiated regions of Cherenkov images produce a Gaussian spread of pixel intensities. Additionally, the read noise and dark current contribute to this Gaussian spread.

4.3.

Conclusion

The low-intensity Cherenkov emission from tissue poses a challenge for imaging techniques. ICCD cameras are beneficial to use due to their high resolution, good sensitivity, short acquisition times, and a large field of view.³¹ However, they have a low dynamic range, the MCPs are limited to a maximum achievable global count rate, and they are generally costly.²⁵ SPAD cameras, conversely, have single-photon sensitivity, picosecond response, good spatial resolution, fast data processing, high timing performance, although are limited by frame rate mismatch with the linac pulses and a low fill factor in the camera.²⁵ Ultimately some better sensor with SPAD or QIS capabilities for single-photon sensing with elimination of amplification-noise, efficient x-ray noise rejection, fast readout speeds, and decreased complexity would be ideal.^28–30 For optimal Cherenkov imaging, the fast time-gating needs to match the pulse length and duty cycle of the linac and provide high signal-to-noise ratio with mm level spatial resolution. As of now, the ideal sensor for these characteristics remains the ICMOS option; however, it is quite possible that low noise sensors with fast electronic shutters may ultimately compete with this in the future.

5.1.

Typical Cherenkov Noise and Clutter

The presentation of noise in the raw video frames of Cherenkov imaging is typical for CMOS camera systems, including noise associated with the camera system itself, the input signal, and stray radiation from the linac head. The stray and signal related noise are stochastic in nature and independent of the camera hardware, but the camera design heavily influences the camera related noise.³² These sources of noise are an ongoing challenge for Cherenkov imaging and development.

5.2.

Detective Quantum Efficiency Assessment

Several methods can be investigated to remove noise and clutter from Cherenkov imaging, including postprocessing, new acquisition techniques, and improved hardware. However, to properly compare Cherenkov images, a quantitative metric is needed. Detective quantum efficiency (DQE) is fundamentally tied to image quality in the field of radiography, describing the ability of a detector to create an image out of incident x-ray photons.⁴⁹ Alexader et al.⁵⁰ influenced the development of the Cherenkov cameras by measuring the DQE of the Gen3 and Gen2+ systems.

The DQE describes the fraction of photons incident on the imaging system required to obtain the same image quality as with a perfect camera system where there are no quanta losses⁵¹ and can be defined as a function of frequency:^52,53

Fig. 9

(a) The DQE can be described as the number of quanta used in a perfect imaging system to obtain a certain quality divided by the number of incident quanta required in a nonperfect system to acquire the same image quality. (b) Example of the DQE of various imaging systems as a function of the spatial frequency showing the superior performance of the Gen2+ system in the low-frequency domain. Reprinted with permission from Ref. 50.

5.3.

Postprocessing Techniques

Postprocessing techniques are frequently employed in CT and MRI applications to improve their diagnostic capabilities by suppressing noise artifacts. A popular approach in CT is the alpha (adaptive) trimmed mean (ATM) filter proposed by Hsieh.⁵⁴ Additionally, various postprocessing algorithms exist for digital camera images, such as nonlocal means (NLM), block matching 3D (BM3D), bilateral filter for denoising with preserved edges, and total variation with an L1 mean (TV-L1).^55–58 These algorithms were tested on Cherenkov images to determine feasible postprocessing algorithm for improved Cherenkov imaging.⁵⁹

A series of long exposure time cumulative images were used to estimate ground truth data and were compared against the same images exhibiting extensive Cherenkov noise. Based on the peak signal-to-noise ratio, the NLM and bilateral filters exhibit the best performance with the TV-L1 filter also producing comparable results. The BM3D filter performed well at low noise levels but saw a large decline in performance at a lower dose delivered (higher image noise), whereas the ATM filter performed the worst.

Although these approaches have not been implemented in the clinical setting at this point, this study demonstrates the utility of postprocessing in improving image quality. Despite the improvement in cumulative imaging, all filters produced poor results when looking at single-frame images, requiring further research.⁵⁹ Additionally, future studies are likely to combine this analysis with deep learning models to further improve Cherenkov denoising.

5.4.

Color Map

An often-neglected aspect of scientific communication is the process of displaying data in a way that is equitable and intuitive. Specifically, color maps used for the display of medical images are imperative for the ease of diagnosis and medical incident detection. Cherenkov imaging is no exception.

The incidence of green–red color deficiency in men of European descent has been estimated to be 8% while most female populations see an incidence around 0.5%.⁶⁰ Therefore, it is crucial to display Cherenkov images such that both physicians and patients experiencing some form of colorblindness can interpret the data. Color maps such as turbo and jet should be avoided as these maps contain red and green elements, which are difficult for many color-blind individuals to distinguish. More scientifically based color maps, such as viridis and thermal, maintain a constant gradient in color perception across their range and exhibit a change from light to dark when transformed to a gray scale leading to easier interpretability for someone experiencing total colorblindness.⁶¹ Clinical implementation of Cherenkov imaging uses scientific color maps to ensure the accuracy of surface dose deposition interpretation.

6.1.

Cherenkov Luminescence Imaging

Because the energy threshold for a beta particle to produce Cherenkov light in water and tissue is $\sim 0.26\mathrm{MeV}$ and $\sim 0.21$ to 0.24 MeV, respectively, which is far below the peak emission energy of electrons and positrons for most commercial radiotracers, Cherenkov light can be used to image the delivery of radiotracer agents to particular structures.⁶²

Cho et al. reported on a method of measuring cherenkov emission from ${}^{18}\mathrm{F}$-labeled substances using a CCD camera, intending to develop a method for quantitatively imaging beta particles produced in a microfluidic chip.⁶³ These chips are used for the synthesis of ${}^{18}\mathrm{F}$-labeled molecules and the proposed Cherenkov imaging allowed for the spatial detection of points of failure. Additionally, Robertson et al.⁶⁴ reported on the first in vivo detection of Cherenkov light from 2-$[{}^{18}\mathrm{F}]\text{fluoro}$-2-deoxy-D-glucose (FDG) with a CCD camera, with resulting Cherenkov images demonstrating tumor uptake of FDG inside the mouse model. Robertson et al. coined the term Cherenkov luminescence imaging (CLI) to describe this planar imaging technique. CLI is a powerful and inexpensive method for testing the uptake of new radiopharmaceuticals (Fig. 10).

Fig. 10

(a) A radioactive drug can be targeted to a tumor. The emission from the source is typically above the energy threshold for charged particles to emit Cherenkov light. If the primary radiation is uncharged, Cherenkov emission is produced when the energy is transferred to secondary electrons. (b) Several different isotopes have been imaged in vivo including gamma emitters (F-18 in the form of FDG⁶⁴), beta emitters (Y-90⁶⁵), and positron emitters (I-124⁶⁶). Reprinted with permission from Refs. 6465.–66. (c) The first Cherenkov luminescence image taken of a human cancer was of a thyroid gland targeted with I-131. Reprinted with permission from Ref. 67.

Many other radionuclides, including ${}^{90}\mathrm{Y}$, ${}^{131}\mathrm{I}$ (beta emitters), and ${}^{124}\mathrm{I}$, (positron emitter), have been studied and used for CLI procedures and in vivo studies.^65,66 Furthermore, it has been demonstrated through simulation that alpha chains used for target alpha therapy, such as ${}^{223}\mathrm{Ra}$, ${}^{212}\mathrm{Pb}$, and ${}^{149}\mathrm{Tb}$ produce significant amounts of Cherenkov light for CLI.⁶⁸ Spinelli et al. performed the first human Cherenkov luminescence image, in which the patient received orally a 550 MBq sample ${}^{131}\mathrm{I}$ 24 h before imaging. To detect the weak optical signal, a cooled electron multiplied CCD with QE optimized in the NIR spectrum was used to image the patient in the dark.⁶⁹ Resulting data demonstrated the uptake of iodine into the thyroid with the potential application of measuring the dose delivered to this superficial organ.⁶⁷ Pratt et al. published a clinical study that demonstrated similar results in patients. 96 patients injected with different cancer targeted radiopharmaceutical drugs were imaged with a fibroscope camera system in a dark patient enclosure, producing external body images of ${\mathrm{Na}}^{131}\mathrm{I}$, FDG, ${}^{68}\mathrm{Ga}$-DOTATATE, ${}^{177}\mathrm{Lu}$-DOTATATE, and ${{}^{223}\mathrm{RaCl}}_2$ uptake.⁷⁰ These studies demonstrate the feasibility of performing supplemental Chernekov imaging for radiotracer therapies.

Additionally, 3D Cherenkov images of in vivo radioactive sources are attainable, coined Cherenkov luminescence tomography (CLT). Using several 2D CLI images at different angles, the CLT image can be reconstructed inversely using the diffusion equation employed in bioluminescence tomography.^62,71 Another 3D reconstruction method proposed by Spinelli et al. removes the need for multiple view angles using spectral information obtained through taking several images at one view angle with different wavelength bandpass filters, which is sufficient for 3D reconstruction of the source.⁷²

Although difficult to apply to in vivo patient studies due to increased attenuation through human skin, clinical studies have demonstrated the use of CLI in surgical guidance and cancer screening. A clinical trial involving the use of CLI for tumor margin determination in breast conserving tumor resection surgery demonstrated promising results using $5\mathrm{MBq}/\mathrm{kg}$ ${}^{18}\mathrm{F}$-FDG. Resected tissue surrounding and including the tumor was imaged in a light tight environment with the optical margins showing excellent agreement with the histologic margin distance.^73,74 A similar clinical trial measured excised prostate tissue from patients injected with 100 MBq ${}^{68}\mathrm{Ga}$-PSMA, showing the localization of ${}^{68}\mathrm{Ga}$ uptake. In both studies, exposure to the medical staff was low. For cancer screening, CLI can act as an alternative to PET, the current clinical gold standard. A 2014 study reported on the injection of ${}^{18}\mathrm{F}$-FDG into patients who were first imaged with PET before having CLI performed in a dark room with a CCD camera. The nodal uptake signal recorded using CLI showed excellent agreement with the PET images.^75,76 Although clinical applications of CLI using radiopharmaceutical agents are still in the development stage, this field laid the groundwork for the advancements in external beam Cherenkov imaging and Cherenkov excited luminescence imaging.

6.2.

Imaging Dose with Cherenkov and Scintillation

Cherenkov light emission imaging allows for both real-time imaging of the radiation field^5,22 and real-time dosimetry for patient treatments using photon⁶ or electron beams.⁷⁷ During patient treatment, the use of Cherenkov light emission imaging allows clinicians to visualize the radiation field and MLC motion and ensure plan delivery. This system has shown that high-intensity regions within a composite Cherenkov image correspond to skin reactions on the patient⁵ and is capable of detecting stray radiation deposited in patients’ eyes during radiotherapy treatments to the brain.⁷⁸ Further clinical implementation of Cherenkov light emission imaging has also led to the detection of treatment delivery issues that were not prevented using standard QA and mitigation strategies.⁸ Cherenkov light emission imaging has also been investigated for use as a quality assurance (QA) technique in high dose rate brachytherapy where sources are imaged in water phantoms, and the Cherenkov emission is quantified to simultaneously measure dose distribution, source strength, and source position (Fig. 11).^22,80

Fig. 11

(a) CT, (b) uncorrected Cherenkov, and (c) CT corrected Cherenkov image data for breast radiotherapy treatment. (d) During these treatments, the CT corrected image data were found to have good linearity with dose. Reprinted with permission from Ref. 79.

To perform real-time dosimetry, a time-gated camera has been used in conjunction with CT scans⁷⁹ or fast-response plastic scintillators.^6,81 The Cherenkov signal produced in patients is attenuated and affected by local tissue optical properties.⁸² By imaging scintillators or correcting Cherenkov light images with CT scans, confounding factors, such as underlying patient anatomy or patient immobilization structures are eliminated as the Cherenkov light is no longer the sole signal used for dosimetry. Two areas of note where imaging of scintillators has been tested are total skin electron therapy (TSET)⁸³ and total body irradiation (TBI).⁸⁴ In both TSET and TBI, the patient is placed at a much further distance from the radiation source^85,86 and, by extension, the camera (Figs. 12 and 13).

Fig. 12

Images of a patient undergoing TSET treatment with scintillators (a) with and (b) without the background signal. (c) Scintillator signal and dose were found to be linear during TSET treatments across different patients imaged. Reprinted with permission from Ref. 77.

Fig. 13

(a) An anthropomorphic phantom with scintillators on the abdomen, chest, and forehead to monitor dose homogeneity during TBI treatments. (b) Under TBI conditions, the scintillators continued to have a linear relationship with dose. Reprinted with permission from Ref. 84.

When implemented for TSET, the dose response to scintillators was found to be linear with the other dosimeters used (TLDs and OSLDs), and signal variation between scintillators was reported as $0.3\%\pm 0.2\%$.⁸³ The standard error of the cumulative dose from the scintillators, compared to other scintillators, was 5%.⁷⁷ In TBI experiments, the imaging of scintillator dose has continued to show good agreement with other dosimeters—achieving repeated agreement with TDLs to within 2.7% and good linearity ($R^2=0.996$).⁸⁴ For contrast, other experiments using inorganic scintillators were able to achieve repeatability and maximum deviation of 0.26% and 0.5%, respectively. It is worth mentioning that these experiments used a different detection scheme with optical fibers carrying the scintillator signal to a photon counter.⁸⁷

In both cases, imaging scintillator dose provides instant dosimetric feedback to clinicians, during the treatment as a means of viewing dose as it is delivered and as a cumulative dose for the treatment fraction. There is also a necessity to correct scintillator signals with the inverse-square law as scintillators placed on the patient may vary in distance from the camera used to image them. An example is the case where a scintillator is placed on the patient’s abdomen and visible in the same image frame is a scintillator on the patient’s forehead. In this case, the camera may be physically closer to the scintillator on the patient’s abdomen, meaning that to properly compare scintillator signals between the abdomen and forehead, corrections must be made. Room lighting is also a known issue but is more prevalent in TBI cases. To correct for the presence of room lights the images are background subtracted. This improves the image quality but as TBI has lower dose rates, the scintillators produce less signal in each captured frame that requires special consideration for the TBI implementation, such as brighter, wider, or thicker scintillators, in order to produce more signal per frame.

In both the TSET and TBI implementations of Cherenkov light imaging, there are special considerations for the angles between the scintillator and the camera (scintillator-camera angle) and between the linear accelerator treatment head and the scintillator (source-scintillator angle). It was found that to maintain a radiant energy fluence precision within 1% that both the scintillator-camera angle and the source-scintillator angle need to be held below 50 deg for TSET.⁷⁷ In TBI cases, to maintain precision within 5% the source-scintillator angle is to be held below 30 deg.⁸⁴ In either case, a reason to reduce the source-scintillator angle is a result of dose-buildup effects where as the source-scintillator angle is increased the radiation beam passes through more of the scintillating material that will produce varying amounts of signal.⁷⁷

An additional effect of imaging and measuring Cherenkov in reference to scintillator dosimetry is that it can become a background signal, which then confounds the scintillator signal. When using time-gated cameras, the Cherenkov signal is often significantly less than the scintillator signal that results in a background term several orders of magnitude lower than the scintillator signal. In TBI cases, where a transparent acrylic spoiler is present consideration must be made for the Cherenkov light produced within the acrylic, which is then detected by the camera.⁸⁴ Other applications of scintillator dosimetry that utilize optical fibers to transmit the scintillator light to a photon counter also encounter a similar background signal where Cherenkov is produced within the optical fiber. In these cases, the Cherenkov light should also be removed, which can be done through a variety of means including background subtraction, simple filtration, and chromatic removal.^88,89

6.3.

Cherenkov and Surface-Guided Radiotherapy

Cherenkov imaging commercially can be considered a subset of the larger efforts in what is called surface guided radiation therapy (SGRT).^90–92 There are different terms used to describe this, driven by different vendors, but the essence is that optical imaging of the patient’s surface is utilized to guide radiotherapy deliver and decisions.⁹³ The standard systems project light onto the patient, and image the projections, recovering some surface topology and with substantial software processing calculate the locations of all relevant parts of the patient in real time. This surface mapping provides the therapy team with a recording of the location and the ability to utilize this surface to ensure that the patient position is matched with the CT simulation scan so that delivery will be accurate. Moreover, day to day positioning in fractionated radiotherapy is improved by this tool, given how much time a patient requires to find the same position in each fraction. Cherenkov imaging has been integrated into SGRT by one vendor, and the combined surface mapping and Cherenkov display of beam delivery on the surface provide an enhanced SGRT feedback to the therapy team. This innovation is still embryonic but the utilization of this for human delivery verification will likely be important tool in the suite of the therapists, as well as for investigation of variations that require further information.

6.4.

FLASH Radiotherapy with Cherenkov Imaging

FLASH radiotherapy employs an ultrahigh dose rate (UHDR) of $>40\mathrm{Gy}/\mathrm{s}$ rather than conventional radiotherapy delivery of $\sim 0.1\mathrm{Gy}/\mathrm{s}$ and has been seen to preferentially harm cancerous tissue over normal tissue in preclinical and clinical studies. Due to its $\sim 1000\times$ higher density of energy per unit time when compared to conventional radiotherapy, increased Cherenkov light will be produced in UHDR conditions, allowing Cherenkov cameras to be a potential alternative to previously used, energy-dependent dosimeters.⁹⁴ Additionally, just as with conventional therapy, there is potential for imaging Cherenkov light emitted from a patient for image guidance, as long as the camera’s temporal speed is on the order of FLASH treatments.⁹⁴

7.1.

Summary of Technology Advances

There have been several key technological advances that have been central to the realization of Cherenkov imaging in clinical radiotherapy. The first of these has been the concept and implementation of triggering the camera acquisition by remote sensing of stray radiation. This single invention allowed the cameras to be utilized for time-gated imaging without contact to the linear accelerator. The second major innovation was in the realization that time-gated imaging was necessary and that with the low duty cycle of most linacs, this allowed for substantial (1000×) removal of background light from the image. The third major innovation was in image processing, utilizing both temporal and spatial filtering (primarily median filtering) to remove spurious events in the camera that have low prevalence at any given pixel and at any given time. The fourth major event has been in wavelength filtering, recognizing that the Cherenkov coming from patients is in the 650 to 950 nm range, and therefore, allowing removal of any extra signal outside this range. The fifth innovation was in the simple process of acquiring dynamic background images, with background image acquisition when the linac is not delivering radiation, and online subtraction of this from the time-gated images. This latter process is achieved at video rate through processing (Table 1).

Table 1

Key technological innovations that have advanced the capabilities of Cherenkov imaging in radiotherapy.

7.2.

Summary of Medical Advances

The medical utility of Cherenkov is clearly the predominant motivation for the development of Cherenkov imaging, to be used as a device for visualization of the radiotherapy beam. The areas of utility are broken down in Table 2. They range from real-time visualization, which has value for the radiotherapy team to see the treatment as it happens and to be able to use this as a real-time tool for visualization. The next value is in the time-integrated Cherenkov, which can then show the total treatment and be used for 2D image comparisons, such as day to day consistency checking or for comparison to surface dose estimates from the treatment plan. The recorded delivery sequence has inherent value because the permanent record can then be used for review in the case of an investigation or an incident. The combination of Cherenkov imaging with other components will also lead to potential added value, one of these being the integration with SGRT and the other being imaging of scintillator reflectors on the tissue. These each have unique and separate value, and themselves are in varying stages of commercial deployment. Finally, the original studies of Cherenkov focused on water phantom imaging for linac quality assurance (QA) testing and as the cameras get better these may be more widely utilized. Beyond this though, there could be utility in verification of complex treatment plans through imaging of the plan delivery in a water tank or on an anthropomorphic body phantom.

Table 2

Signals that can be observed with these imaging systems have a range of values as listed.

7.3.

Summary of Science Advances

Perhaps the single most important science advance in Cherenkov imaging has been the realization that high-resolution single-photon imaging can be achieved within ambient room lighting. The technological solutions listed above in Sec. 7.1 have made this possible, but it now opens up the idea that light can be sampled from patients during radiotherapy. Additional innovations have been explored in the idea that optical sensors maybe added to the patient to sense things like tissue destruction (ref), oxygen or pH within the tissue. Engineered optical contrast agents that have long-lifetime emission properties may become important parts of sensing in radiation therapy, either for particular metabolism features, or more simply for in vivo radiation dose delivered inside the tissue (Table 3).

Table 3

There are a range of current and future possible innovations listed with their utility and the origin of the signal that can be captured.

7.4.

Remaining Challenges

The value of Cherenkov imaging is still actually emerging as adoption into clinical practice is just emerging at this point. However, there are several unexplored or undeveloped areas of Cherenkov that could have value. These can be categorized several ways, but here are outlined in terms of (1) the physical properties of Cherenkov un-utilized at this point, (2) improvements to sensors that could improve the detection and use, (3) improvements in the biological interpretation of the images, (4) improvements in the dosimetric interpretation of the images, and (5) advances in automation that might incorporate Cherenkov. Each of these is listed in Table 4, with a range of the unique features that have not been fully explored or exploited yet. These are speculative but the listed features are unique and as such could have value that is unknown at this point.

Table 4

Future potential of Cherenkov could be augmented if currently unexplored features are exploited.