2.1.

Tissue Phantoms

A previously described color Cherenkov camera that captures RGB wavelength channels separately was used in this study.¹⁰ The acquisition was time-gated to the LINAC, capturing Cherenkov emission only during radiation pulses, thereby removing background ambient light. This setup is shown in Fig. 1(a). Image acquisition and processing were based on prior work,¹⁰ using the C-Dose research software, which samples image from the camera sensors at fixed numbers of radiation pulses, with the image intensifiers pulsing on with every x-ray pulse from the LINAC and the total image data accumulating on each chip to provide a higher intensity prior to readout to the computer. Background image data was sampled as well and saved onto the FPGA of each camera, such that background subtraction could be achieved on each channel of the camera. Mean intensities for each individual channel were obtained from consistent regions of interest (ROIs) on frame-averaged image stacks.

Fig. 1

(a) Acquisition setup, using C-Dose software for on-site image processing, LINAC for beam delivery and (b) three-channel time-gated intensified cameras for RGB Cherenkov capture (arrow). Reproduced with permission from Alexander et al. (LSA 2021). (c) Detailed schematic of custom camera.¹⁰ (d) Representation of necessary time-gating and synchronization to LINAC pulses.¹⁰

To simulate tissue, synthetic epidermal layers $100\mu \mathrm{m}$ ($+/-5\mu \mathrm{m}$) thick were fabricated, with varying biological concentrations of synthetic melanin (M8631, Sigma-Aldrich, St. Louis, Missouri) at 0.0018, 0.0038, 0.0114, 0.019, 0.027, 0.045, and $0.072\mathrm{mg}/\mathrm{ml}$. Thicknesses were calculated based upon the volume of material produced and the area. The epidermal layers were placed on top of 2-cm thick bulk tissue phantoms made of silicone with flesh-colored pigment, using a process outlined in a previous publication,¹² composed of silicone embedded with flesh-colored pigments that match human soft tissue optical properties (Smooth-On, Macungie, Pennsylvania).

Melanin concentrations were selected to match the expected range of human skin color based on several factors.^13,14 Primarily, we employed the Fitzpatrick scale,¹⁵ which assessed skin color based on reaction to ultraviolet radiation. Pigmented phantoms representing the following Fitzpatrick scale types were fabricated: Fitzpatrick type 1 (score 0 through 6), type II (score 7 through 13), type III (score 14 through 20), type IV (score 21 through 27), type V (score 29 through 34), and type VI (score 35+). Additionally, we used visual inspection to confirm that the phantoms were within the apparent visual range of human skin color as expected, as we found that the limitation of just the six skin values did not fully represent the range of pigmentation levels existing in human tissue. In particular, at the higher melanin content levels, more delimitation is needed to cover the range of human values. Base concentrations for the pigmented skin phantoms were previously reported.¹⁶

Average optical properties (absorption coefficient, ${\mu }_a$, and reduced scattering coefficient, ${{\mu }_s}^{\prime }$) of the phantoms were determined to confirm tissue-like optical behavior for a range of human skin colors. These measurements were done with spatial frequency domain imaging (SFDI), explained further below in Sec. 2.6. Intralipid was used as a scatterer at a fixed level of 1%. Bovine whole blood solutions (Lampire Biological Laboratories, Pipersville, Pennsylvania) with varying biological concentrations (0.5%, 1%, 1.5%, 2%, 2.5%, 3%, and 3.5%) were prepared in optically blacked out petri dishes. These values have been used in many previous studies to match the near infrared tissue optical property range for blood absorption in soft human tissues.⁷ Whole blood is fully oxygenated in ambient aqueous solution so that the hemoglobin (Hb) can be assumed to be 100% oxygenated hemoglobin for the spectral signature.

During LINAC irradiation of the phantoms, images were captured for each color channel, and post-processing extracted average RGB Cherenkov emission intensities from the recorded images, as functions of melanin and blood concentrations, as is described below.

2.2.

Color Cherenkov Camera

The RGB color Cherenkov camera, as seen in Fig. 1(b), was composed of three independent intensified complementary metal oxide semiconductor (iCMOS) cameras (C-Dose, DoseOptics LLC, Lebanon NH) housed in a three-tube color video camera assembly (JVC, Yokohama, Japan). The red channel was outfitted with a red-sensitive intensifier and red filter, the blue and green channels were blue-green sensitive intensifiers, and appropriate color filters for each. Each camera was remotely triggered by leakage x-rays,¹⁷ allowing for synchronization with and gating to the linear accelerator pulses [Fig. 1(d)]. The camera software supported 16 bit read capability for quantitative acquisition. The beam splitter assembly of the video camera, consisting of RGB dichroic beam splitters and bandpass filters, allowed for incoming Cherenkov light to be redirected according to wavelength to the appropriate camera channel resulting in three raw image stacks for each acquisition, shown in Fig. 1(c). These RGB image stacks were multiple images acquired per imaging time, such that each RGB channel resulted in a temporal image stack for each channel. The camera was equipped with a 10 to 100 mm, $f/1.6$ zoom lens (JVC, Yokohama, Japan).¹⁰

2.3.

Imaged Color Cherenkov Emission

The absorption spectra of melanin,^18–20 Hb, and ${\mathrm{HbO}}_2$¹⁰ are shown along with the emission spectrum of Cherenkov light in Fig. 2(a). The sensitivity spectra of RGB channel filters of the RGB color Cherenkov camera are shown in Fig. 2(b). Individual camera detection spectra for the three channels (RGB) were characterized to verify the efficacy and reliability of the filters in tracking and capturing individual channel intensities from experiment tissue samples within expected bands for RGB wavelengths. A tunable light source (TLS) (Optometrics Manual TLS,²² Optometrics Manufacturing, Ayer, Massachusetts) was used to characterize the optical system response. Wavelengths from 380 to 720 nm with a step of 20 nm were used for characterization and the response is shown in Fig. 2(b). This was done using a manual TLS and the tri-color camera, shown in Fig. 2(c). The TLS maximized throughput in the visible region of the spectrum using a 20W tungsten halogen lamp, with a spectral energy between 360 and 2000 nm.²³ The light exiting through a small slit was imaged in close range with the tri-color camera using C-Dose software (DoseOptics LLC, Lebanon, New Hampshire), with images processed in MATLAB (v2022a, Natick, Massachusetts) and signal intensity graphed in Fig. 2(b).

Fig. 2

(a) Absorption spectra of melanin, Hb, and ${\mathrm{HbO}}_2$ are shown²¹ on the same graph with the emission spectra of Cherenkov, with each arbitrarily scaled. In panel (b) the RGB background subtracted (BG sub) sensitivities of the filtered color Cherenkov camera are shown. (c) A TLS was used to calibrate the camera with narrowband monochromatic light, for the data in (b).

2.4.

Blood Phantoms

Whole blood solutions at total volume 100 ml were created (Fig. 3) using stock solutions of bovine blood (Lampire Biological Laboratories, Pipersville, Pipersville), phosphate buffered saline solution (Cytiva, Marlborough, Massachusetts), and Intralipid (Sigma-Aldrich, Burlington, Massachusetts). The combined solutions were mixed with concentrations of 1% Intralipid (5 ml at 20% stock emulsion), using variations of blood at 0.5%, 1%, 1.5%, 2%, 2.5%, 3%, and 3.5%, with the remaining volume being phosphate buffered solution. The solutions were poured into 100-mm circular style cell culture dishes, with 20-mm depth [Fig. 3(b)] (Corning, Corning, New York), which had been coated with matte black paint to reduce optical edge effects in the images.

Fig. 3

(a) True color images of solutions created in their respective dishes with increasing blood concentration from left to right. (b) Example of single solution dish in blacked-out Petri container. (c) Prepared bovine whole blood and phosphate buffered saline, which were later combined with the Intralipid.

2.5.

Pigmented Melanin Layers for Phantoms

Synthetic epidermal layers of 0.1-mm thickness were fabricated based on a previously described method.¹⁶ This was done by dissolving 1 g of gelatin powder of Type A porcine powder, 300 g Bloom (Sigma-Aldrich, St. Louis, Missouri) and 0.5-g glycerol ($\ge 98\%$ purity, Sigma-Aldric, St. Louis, Missouri) in 10 ml of distilled water. Then 0.01% to 0.1% glutaraldehyde ($\ge 99.5\%$ purity, Sigma-Aldrich, St. Louis, Missouri) was added, along with varying concentrations of synthetic melanin (0.0018, 0.0038, 0.0076, 0.019, 0.027, 0.045, and $0.072\mathrm{mg}/\mathrm{ml}$), which came in the form of small crystals and was manually crushed to a fine powder (M0418-1G Lot# BCCB7179, Sigma-Aldrich, St. Louis, Missouri). The resulting solution was heated evenly with a 1200-W microwave for 5 s to 45°C temperature (Etekcity Lasergrip 1080 IR Thermometer, Anaheim, California) allowing for an even mixture. The solution was poured onto plastic molds (large Fischer weight boats) and then placed into a vacuum chamber to remove air bubbles and distortions. The volume of each melanin mixture was measured such that when it was poured onto the phantom that the layer thickness could be estimated to be 0.1 mm, and was based upon previous trials where after drying the thickness was measured with a micrometer. These layers were then dried for 48 h at 21°C in a fume hood to acquire thin, permanent pliable layers.

2.6.

Verification of Color Values with Spatial Frequency Domain Imaging

The impact of melanin concentration on the optical properties (absorption coefficient, ${\mu }_a$, and reduced scattering coefficient, ${{\mu }_s}^{\prime }$) of the epidermal layers was quantified using a validated, reflectance geometry SFDI and software (Reflect RS, Modulim, Irvine, California). SFDI²⁴ separates the effects of scattering and absorption and can be used to estimate the concentrations of chromophores in the tissue. The technique works by illuminating different patterns light on the tissue, imaging the reemitted light, and demodulating the imaged reflectance.^25,26 Optical properties were quantified at each of eight wavelengths (471, 526, 591, 621, 659, 691, 731, and 851 nm) using five spatial projection frequencies (0.00, 0.05, 0.10, 0.15, and $0.20{\mathrm{mm}}^{-1}$), and all experimental measurements were calibrated to the supplied tissue phantom from Modulim. The fit to the optical properties of the phantoms used all five spatial frequencies in the inversion, with fit to a lookup table of values within the supplied software. The epidermal layers were placed on top of 2-cm thick bulk tissue phantom made of silicone with flesh-colored pigment during this procedure.

3.1.

Average Optical Property Characterization by SFDI

Results from the phantom validation measurements by SFDI are summarized in Fig. 5. Property averages were computed based on the distribution of pixels within each wide-field, circular 5 cm ROI for each phantom. Visual inspection of these optical phantoms, combined with the SFDI property measurements, were used to confirm that the melanin layers exhibited optical properties that corresponded to the expected array of human skin colors.^26,27

3.2.

Cherenkov Images

RGB resolved Cherenkov images for varying melanin and blood concentration variations are shown in Fig. 6 and are compared to white light images. The attenuation in Cherenkov emission from tissue phantoms due to melanin absorption is shown in Fig. 6(a). Emission was observed to decrease as melanin concentration increased, without one particular color vastly differentiating itself in intensity from another. This is in line with melanin spectral properties showing no anomalous preference to a particular wavelength and decreasing in absorption units with increasing wavelength as noted in literature^28–30 as well as being confirmed via SFDI optical property validation in Figs. 4 and 5. For purposes of visualization, the blue channel needed individualized windowing and leveling, and thus, a separate color bar is shown. The zero concentration of melanin was a gelatin layer with no melanin present in it.

Fig. 4

True color view of epidermal layers created in their respective dishes with increasing melanin concentration from left to right.

Fig. 5

(a) Calibrated reflectance average values; (b) reduced scattering coefficient average values; and (c) absorption coefficient averages, all listed by concentration values in units of mg/ml in the legend.

Fig. 6

(a) White light images of melanin tissue phantoms with varying absorption. Images of the Cherenkov emission images in each of RGB wavelength bands are shown. Concentrations of melanin were 0.0018, 0.0038, 0.0076, 0.0114, 0.019, 0.027, 0.045, and $0.072\mathrm{mg}/\mathrm{ml}$, respectively, left to right, with $600\mathrm{MU}/\min$, 6 MV and 300 MU (b) White-light images of blood phantoms with varying absorption. Images of the Cherenkov emission images in each of RGB wavelength bands are shown. Concentrations of blood were 0.5%, 1%, 1.5%, 2%, 2.5%, 3%, and 3.5%, respectively, left to right, with $600\mathrm{MU}/\min$, 6 MV, and 300 MU.

The attenuation in Cherenkov emission from the blood samples is shown in Fig. 6(b). The observed attenuation increased with increasing blood concentration with the red channel showing the greatest in mean signal intensity change compared to the green and blue channels. This is in line with blood spectral properties showing preference to a particular wavelength—Hb bound to oxygen absorbs blue-green light, reflecting red-orange light³¹—appearing red and varying non uniformly in absorption units with increasing wavelength as noted in literature.^28–30 For purposes of visualization, as before, the blue channel needed individualized windowing and leveling and thus a separate color bar is shown.

3.3.

Differential Signal Response

Cherenkov images from each of the individual channels—gathered using C-Dose software and processed in MATLAB—resulted in the output of signal intensities across various concentrations in melanin and whole bovine blood, as shown in Fig. 7. Data gathered show an attenuation response for both melanin and blood. For melanin, as per visual aid of Figs. 6(a) and 6(b), the quantitative response graphed in Fig. 6(a) shows all channels following lower emission intensity with increasing pigment concentration. Comparing this response with the reference chart in Fig. 2(a), the trend seen in our results follows what is expected for Cherenkov emission and melanin concentration. Conversely, the whole bovine blood quantification [Fig. 7(b)], with visual aid from Figs. 6(a) and 6(b), shows a differential response from the three channels. The red channel Cherenkov signal intensity attenuates to a much lesser extent, while the green and blue channels follow a tight diminution together to levels lower than red as a whole. Repeated measures of the phantoms confirmed these observed trends, with the error bars being smaller than the data points themselves. The increased absorption of blue and green by Hb³¹ results in increased signal in the red channel. Furthermore, Fig. 2(a) corroborates this observation; oxygenated Hb exhibits two peaks where blue and green light wavelengths are represented.

Fig. 7

Normalized Cherenkov RGB signal intensities versus (a) melanin concentration and (b) blood concentration.

This phenomenon of different amounts of exiting light in the RGB bands in color Cherenkov imaging provides quantitative and qualitative confirmation of detection of signals that varied differently with changes in blood or melanin content.

Further color Cherenkov imaging of skin pigmentation phantoms is seen in Fig. 8. About 3 Gy of dose was given via 6-MV photon and 6-MeV electron beams to the seven phantoms to show the limits of visual analysis of Cherenkov color imaging. With both beam energies, pigmentation past $0.0076\mathrm{mg}/\mathrm{ml}$ is seen to be extremely difficult to distinguish from the background—effectively no observable Cherenkov emission. This observation also further illustrates the need for correcting Cherenkov signal attenuation due to biological tissue factors so that the emission can be seen as visually independent from patient-specific attenuating influences. It is noteworthy that the zero concentration values were removed because of reflection effects in the samples that placed their values outside of the trend with expected human tissues. Given that these concentrations are unrealistic of human tissues, their removal was deemed appropriate and provided a clearer interpretation of the trends seen here.

Fig. 8

Visual display of seven phantoms chosen to show slight variation in melanin content at the lowest values. The standard color photograph of the tissue phantoms is shown in Fig. 4 with ambient room lighting, and the Cherenkov color image is shown here in (a) and (b) taken in a darkened room with (a) MeV beams and (b) MV beams. The values depicted above the images are in units of mg/ml to represent biological melanin concentration in each phantom.