2.1.

Patient/Experimental Imaging Setup

An example set up for experimental imaging of Cherenkov luminescence from a phantom is shown in Fig. 1(a). A tripod mounted intensified CMOS camera (CDose, DoseOptics LLC, Lebanon, New Hampshire) imaged the phantom with the optical axis aligned and focused to the isocenter. A 50-mm f/1.8 lens (Nikon Inc, Tokyo, Japan) was used with a SM2 to SM1 adapter fitted with a 1-in. lens tube to attach 25-mm LP optical filters as needed (Thorlabs, Newton, New Jersey). The camera was triggered off the linear accelerator (LINAC) pulses to image during irradiation (6MV x-ray with 100 source-to-surface distance). Figure 1(a) shows an anthropomorphic tissue equivalent phantom²⁶ to represent a breast treatment. A flat tissue phantom composed of the same material was also imaged at isocenter to quantify SNR and Cherenkov to room light background ratio ($I_{\mathrm{CH}}/I_{\mathrm{BKG}}$). The phantoms were representative of human tissue in that it emitted predominantly in the red and infrared wavelength due to absorption of blue and ultraviolet Cherenkov photons. Nonetheless, the phantom is homogeneous while human tissue is composed of lipids, hemoglobin, water, and melanin, each with layers and different optical properties.³⁰ The optical properties (absorption and reduced scattering coefficient) of the tissue phantom are included in Fig. 1(b). The anthropomorphic phantom was used to represent and compare to typical patient geometry and imaging discussed further in Sec. 4. Cherenkov luminescence from patients was imaged with the ceiling mounted intensified CMOS camera directly above the tripod mounted camera, though both cameras have the same imaging parameters and specifications. The ambient background light sources are indicated in Fig. 1(a). The prospective light sources considered for the study were placed near the phantom covered by a diffuser to produce a homogeneous distribution of light on the phantom and held constant at 10 lux at the isocenter. An example of the Cherenkov emission and background images with a light source of the flat phantom are shown in Fig. 2.

Fig. 1

(a) Current patient imaging set up with an intensified time gated CMOS camera with an anthropomorphic breast tissue phantom as an example at isocenter and arrows indicating the current treatment room light sources. (b) Absorption coefficient (${\mu }_a$) and reduced scattering coefficient (${\mu }_s^{\prime }$) spectrum for the tissue phantom.²⁶

Fig. 2

Quantifying Cherenkov emission image quality using optical filters. (a) Example Cherenkov emission image with Cherenkov (solid) and background (dashed) ROI used for calculation of SNR. (b) Example background image with (dashed) ROI used for $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio.

2.2.

Light Source and Filter Characterization

A spectrometer (Ocean Optics Inc., Dunedin, Florida) coupled to an optical fiber (Thorlabs Inc., Newton, New Jersey; pure silica core, $600\text{-}\mu \mathrm{m}$ diameter, and hard polymer cladding; $0.37\pm 0.02$ numerical aperture) measured the emission spectrum of all ambient room light sources (including ones currently mounted, as shown in Fig. 3(a), and prospective ones for replacement). An intensified CCD camera (PI-MAX3, Teledyne Princeton Instruments, Trenton, New Jersey) coupled to a spectrograph (SpectraPro 2300i, Acton Research Corporation) was time gated to the LINAC pulse (at 360 Hz and 5-ms exposure time) to measure the Cherenkov emission spectrum from the tissue phantom when irradiated, as shown in Fig. 3(b). The transmission spectrum of each LP filter [Fig. 3(b)] was measured using the Varian Cary 50-Bio spectrophotometer (Agilent Technologies, Santa Clara, California). A digital lux meter (LX1330B, Dr. Meters) measured the illuminance of the light sources at the treatment room isocenter for currently mounted sources (Table 2) and prospective replacement ones (to ensure 10 lux at the isocenter).

Fig. 3

(a) Quantum efficiency spectrum of camera⁴ and ambient room light source spectrum. (b) Optical filter transmission spectrum and tissue phantom Cherenkov emission spectrum considered for improving patient imaging quality.

2.3.

Prospective Light Sources and Camera Photocathode

The types of light source, power level, color temperature, and cost were provided by each of the manufactures for the prospective replacements (included in Table 1). The lights considered were halogen lights, various LED light sources, and representative compact fluorescent and tungsten light bulbs. All the lights included in the study were commercially available and within $25 per bulb. These light sources were considered due to power levels that are typical of treatment room lights and represented a range of the color temperatures available commercially. An alternative blue sensitive photocathode for the camera was also considered to change the sensitivity spectrum of the camera and for further preferential imaging of Cherenkov emission. The Cherenkov emission was imaged with different light sources on and with or without the 675-nm LP filter. The light source was held constant at 10 lux at the isocenter and surface of the flat tissue phantom.

Table 1

Alternative and representative light sources considered to replace current treatment room lights (manufacturers are in Table S1 in the Supplementary Material) with specifications including power level, color temperature, and cost of each source. CFL stands for compact fluorescent light bulbs and LED stands for light emitting diodes.

Table 2

Illuminance for current ambient room light sources at the treatment room isocenter.

2.4.

Image Processing

The impact of ambient light sources on Cherenkov emission image quality was quantified from both ambient room light (background) and Cherenkov images. There were 100 images per acquisition, and all had the dark signal subtracted. Each image was spatiotemporal median filtered ($5\times 5\times 5\text{pixel}$ window) to remove stray radiation noise.¹⁸ Example acquired images of the Cherenkov emission and background are shown in Figs. 2(a) and 2(b). The SNR was evaluated for each frame based on delivery of a $10\times 10{\mathrm{cm}}^2$ field from the LINAC with the ROI of the signal (rectangle) and background (triangle) shown in Fig. 2(a). The background was chosen to be on the tissue phantom but in a region with minimal Cherenkov emission from the tissue. It was defined as $\mathrm{SNR}=\text{mean}(\frac{\mathrm{CH}(i,j)-{\mu }_{\mathrm{BKG}}}{{\sigma }_{\mathrm{BKG}}})$, where $i$, $j$ are the pixel coordinates, CH is the Cherenkov emission ROI, and ${\mu }_{\mathrm{BKG}}$, ${\sigma }_{\mathrm{BKG}}$ are the mean and standard deviation of the background ROI. The ratio of the intensity in Cherenkov and background light source ($I_{\mathrm{CH}}/I_{\mathrm{BKG}}$) was evaluated based on the rectangular ROI’s at the isocenter shown in Figs. 2(a) and 2(b). The current ambient sources’ SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ are included in Fig. 4. The emission spectrum of each prospective light source is included in Fig. 5, and SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ are included in Fig. 6. The absolute mean Cherenkov emission signal CH for the tested LP filters and optical density (OD) of the background signal from a 675-nm LP filter with RGB LED light source are included in Fig. 7.

Fig. 4

(a) SNR and (b) $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio for current light sources with tested filters. (SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio values are included in Table S2 in the Supplementary Material.)

Fig. 5

Emission spectrum of each prospective light source and quantum efficiency of considered camera photocathodes.^4,11

Fig. 6

(a) SNR and (b) $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio for proposed light sources considered with and without a 675-nm LP filter. (SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio values are included in Table S3 in the Supplementary Material.)

Fig. 7

(a) Absolute Cherenkov emission count per frame with room lights off and different optical filters. (b) OD for each LED channel and RGB composite comparing with and without the 675-nm LP filter. (Method for measuring OD is included in Fig. S1 in the Supplementary Material.)

2.5.

Potential Utility of Spectral Filtering for Improved Patient Imaging

The anthropomorphic phantom was imaged with an optical filter and ambient room light source combination that produced the best conditions for patient imaging (considering room light temperature, SNR, and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio) to illustrate a potential utility of spectral filtering (Fig. 8). Representative of a patient treated with an optically clear bolus [Figs. 8(a) and 8(b)], a 1-cm bolus was placed on top of the right breast of the phantom and the breast was irradiated with a $20\times 20{\mathrm{cm}}^2$ field. The camera imaged the background and Cherenkov emission of the phantom with or without the optical filter [Figs. 8(c)–8(f)]. The ratio of the Cherenkov images was determined dividing the emission profile with and without the optical filter [$I_{\text{NoFilt}}/I_{\text{Filt}}$, Fig. 8(g)]. For comparison, the phantom was also imaged with the current room light source (tungsten) and with or without the optical filter, which is included in Fig. S2 in the Supplementary Material.

Fig. 8

Utility of spectral background subtraction for removing image artifacts [e.g., specular reflection from ambient room light (arrows) and total internal reflection of light within transparent bolus (dotted ovals)]. (a), (b) Example Cherenkov emission acquisition from a patient treated with bolus. (c)–(g) Effects of 675-nm LP optical filter on image artifacts with RGB LED light source.

3.1.

Current Patient Imaging Set up

The current patient imaging set up captures a non-negligible superposition of both ambient room light and Cherenkov emission from the patient surface, as shown in Figs. 3(a) and 3(b). The tissue equivalent phantom, representative of patients, emitted predominantly above 600 nm, which was appropriately imaged with a photocathode sensitive to photons above 560 nm (quantum efficiency above 20%). Nonetheless, the ambient light sources exhibited a broad emission spectrum as well. The fluorescent light sources have similar emission spectrum with peaks ranging from 400 to 710 nm, with the largest peak at 607 and 614 nm for fluorescent sources 1 and 2, respectively. The two fluorescent sources had a much higher light level of 204 and 51.3 lux in comparison to the tungsten source of 0.6 lux. Though dim, the tungsten light source emitted predominantly above 600 nm and in the near-infrared (NIR), which is comparable to the Cherenkov emission spectrum of the tissue phantom.

3.2.

Spectral Filtering

LP spectral filters attached to the camera were used to evaluate their ability to suppress ambient light while maintaining image quality of Cherenkov emission surface profiles. All filters suppressed light with an $\mathrm{OD}>3$ below the cut on wavelength specified in Fig. 3(b). While an ideal 600-nm LP filter would allow the most photons to be detected (in comparison to the other filters), the spectrum shows the 600-nm LP filter transmits $\sim 90\%$ of the light above its cut on wavelength, but the other filters (650, 675, and 720 nm) transmit $\sim 95\%$ or more of light above its cut on wavelength. Nonetheless, the SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio for each of the filters and with each light source on are shown in Figs. 4(a) and 4(b). As expected, when no room lights were on, spectral filtering simply reduced Cherenkov emission signal and the SNR with the greatest reduction was the 720-nm filter (from SNR of $67\pm 6$ to $25\pm 2$ per frame). The 600-, 650-, and 675-nm LP filters had comparable SNR of $\sim 40$ ($43\pm 4$, $38\pm 4$, $41\pm 4$, respectively).

When room lights were on, the spectral filters had contrasting results for the tungsten light sources and the fluorescent sources. The SNR for fluorescent 1 light source was less than fluorescent 2 while having similar spectrums, which can be attributed to the greater illuminance (fluorescent 1 was brighter). However, spectral filtering up to 720 nm for both fluorescent sources improved the SNR from $1.95\pm 0.02$ to $4.8\pm 0.2$ and $2.14\pm 0.05$ to $7.1\pm 0.4$, respectively. The $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio for the two fluorescent sources improved by a factor of $˜4$. However, the SNR was comparable and even reduced when using the filters with the tungsten light source. There was a reduction of the $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio by about 70%. Although SNR did improve with the fluorescent sources and the 720-nm filter, it is worth noting that the Cherenkov emission signal was reduced by a factor of $\sim 2.3$ (shown in Fig. S1 in the Supplementary Material).

3.3.

Light Source Replacements

From the spectra shown in Fig. 5, most of the light sources considered have very broad spectra, including the LED lights. The compact fluorescent light (CFL) and tungsten light sources had very similar spectra to the ones that are currently installed. The halogen bulbs emitted in a broad spectrum above 500 nm that extended to the NIR. The amber, soft white, and white LED though emitted in a broad spectrum, predominately emitted below 700 nm. The RGB(W) LED light source exhibited the greatest potential for replacing the current light sources due to narrow peaks for the red, green, and blue source (though the white source in the RGB(W) was similar to the soft white). A blue sensitive photocathode was also considered, but it was least sensitive at the wavelengths of Cherenkov radiation emitted from tissue.

As shown in Fig. 6, the highest SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio were achieved by the blue channel on the RGB(W) LED light source, which can be attributed to the emission peak (465 nm) being in the spectrum where the red photocathode is the least sensitive. The tungsten and halogen bulbs resulted in the least SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio, which can be attributed to the broad emission spectrum weighted toward red and NIR wavelengths. Nonetheless, the 675-nm LP filters imaged Cherenkov emission showed a great improvement in the SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio, particularly for the red (R), green (G), and blue (B) light sources due to their narrow peaks. Imaging with any of these sources and their composite with the 675-nm LP filters showed comparable SNR and $I_{\mathrm{CH}}/I_{\mathrm{BKG}}$ ratio. This can be attributed to the LP filter’s ability to suppress the RGB LED light source with an OD of $3.13\pm 0.05$ [shown in Fig. 7(b)]. The optical filter also improved the image quality for other LED light sources (white, soft white, and amber), halogen, and CFL because the filter suppressed a large fraction of the light emitted from the sources below 675 nm while maintaining $\sim 70\%$ of the Cherenkov emission detected by the camera [Fig. 7(a)].

3.4.

Potential Utility of Spectral Filtering for Improved Patient Imaging

Figure 8 shows a potential utility of the spectral filtering in a current image setup with optically clear bolus. Imaging patients with room lights on presented specular reflection [indicated with arrows in Figs. 8(a) and 8(b)], which can make it difficult to correlate Cherenkov emission to surface dose through the bolus.⁵ Furthermore, the light piping of optically clear Cherenkov emitters such as the bolus would make it difficult to quantify dose at the edge of the bolus or treatment fields [indicated with dashed ovals in Fig. 8(b)]. Figures 8(c) and 8(e) show results with narrow band RGB LED light sources and a 675-nm LP filter; the specular reflection was negligible as the ambient light detected was reduced by an $\mathrm{OD}>3$. Figure S2 in the Supplementary Material shows the specular reflections were prominent in the Cherenkov emission profile with the tungsten light source, both with and without the LP filter. Figure 8(g) shows that Cherenkov emission from optically clear bolus was disproportionately suppressed in comparison to the tissue phantom, as the edges of the bolus with the light piping effect showed the $I_{\text{NoFilt}}/I_{\text{Filt}}\sim 2.5$, and the surface most orthogonal to the camera’s optical axis and surfaces without bolus had $I_{\text{NoFilt}}/I_{\text{Filt}}\sim 1.7$.