2.1.

Animal Welfare

All animal experimental procedures were carried out in accordance with the Johns Hopkins University Animal Care and Use Committee under strict accordance with institutional guidelines and regulations for the use of vertebrate animals.

2.2.

Orthotopic GBM Mouse Model

To establish the orthotopic GBM mouse model, a C57BL/6J mouse (female, 8-weeks-old; Jackson Laboratory, Bar Harbor, Maine, United Stated) was immobilized on a stereotaxic instrument (Catalog No. 51730; Stoelting, Wood Dale, Illinois, United States) at prone position. A 5- to 8-mm sagittal incision on the scalp was made. A 0.5-mm-diameter parietal burr hole at 20-mm anterior to the lambdoid suture and 2-mm left to the sagittal suture of the skull was made with an electric bone drill (Catalog No. 51449; Stoelting, Wood Dale, Illinois, United States). $1.2\times {10}^5$ GL261-Luc2 cells in $2\mu \mathrm{L}$ of phosphate-buffered saline (PBS) were injected 3-mm deep into the mice brain by utilizing a $10\text{-}\mu \mathrm{L}$ Hamilton gas-tight syringe with 32-gauge blunt-tip needle (Catalog No. 53485-1; Stoelting, Wood Dale, Illinois, United States). $1\mu \mathrm{L}$ of cell suspension was injected and after waiting for 1 min these two steps were repeated until all suspension was injected. After the injection was complete, the syringe was raised 1 mm and after 1 min wait these two steps were repeated until the needle was completely moved out of mouse head. The skull opening and incision were sealed with tissue adhesive (No. 1469SB; 3M Vetbond^TM/MC, St. Paul, Minnesota, United States).

2.3.

Implantation of Self-Luminous Light Source into Mouse Pancreas

A self-luminous light source (cylinder, 0.9 mm in diameter, 2 mm in length) was surgically implanted into the pancreas of albino C57BL/6J mouse (female, 8-weeks-old; Jackson Laboratory, Bar Harbor, Maine, United States). The surgical site was located at the skin on the mouse left flank, around spleen position, posterior to rib. An 8- to 12-mm transverse incision was made on the skin at the surgical site, and another 8- to 12-mm transverse incision was made on the peritoneum right under the skin incision. The spleen and pancreas were gently exteriorized, and the pancreas was spread out on the skin. A small pocket was created in the pancreatic parenchyma, and the light source was inserted into the tissue pocket. A 7-0 suture (CC1107N-45; AROSurgical™, Newport Beach, California, United States) was used to close the tissue pocket of the pancreatic parenchyma. After placing the pancreas and spleen back into the abdominal cavity, the incisions on the peritoneum and skin were closed using a 4-0 suture (E15A04L-45; AROSurgical™, Newport Beach, California, United States).

2.4.

BLT System

Two BLT systems in a similar configuration and equal performance were utilized in this study. The first one is an in-house BLT system¹⁸ consisting of an optical assembly, a thermostatic system, a transportable mouse bed, and a mobile cart. The optical assembly is driven by a motorized linear stage to dock onto the mouse bed for imaging. The assembly contains a rotatable three-mirror system (98% reflective, protected silver coating) with four light-emitting diodes mounted at its corners for photo imaging, a filter wheel (Edmund Optics Inc., Barrington, New Jersey, United States), a charge-coupled device (CCD, iKon-L936; Andor Technology, Belfast, United Kingdom) mounted with a 50-mm $f/1.2$ lens (Nikkor, Nikon Inc., Melville, New York, United States), and a light-tight enclosure. The optical signal emitted from an imaged object was directed by the three-mirror system, passing through the filter wheel and captured by the CCD camera. The image taken at top of the mouse bed is labeled as 0-deg projection imaging. The three-mirror system can rotate 180 deg (from $-90\mathrm{deg}$ to 90 deg) around the imaged object for multiprojection imaging. The focal plane was set at $\sim 11\mathrm{mm}$ above the mouse bed with the three-mirror system placed at 0 deg. The pixel scale is the corresponding physical size of CCD pixel at focal plane, which is 0.12 mm per CCD pixel. The optical path from the focal plane to the front surface of the camera lens is 45 cm. Four 20-nm full-width-half-maximum (FWHM) bandpass filters (Chroma Technology Corp., Bellows Falls, Vermont, United States) at 590, 610, 630, and 650 nm were mounted in the filter wheel for multispectral imaging. The thermostatic system, which is built in the light-tight enclosure (except for the heat gun) and consists of a resistor, a thermocouple with monitor, seven fans, and a heat gun with heat transport pipeline linked to the enclosure which was used to boost and maintain the temperature around the imaged object at 37°C. The transportable mouse bed allows the imaged object transferred from the optical system to CBCT system for CBCT imaging. Eight ball bearings (BBs; PTFE balls, 2.4-mm diameter; McMaster-Carr, Santa Fe Springs, California, United States), which can be identified both in optical photo image and in CBCT image, were attached on the bed as the fiducial markers for registering the coordinates of optical and CBCT system. The in vivo data of GBM-bearing mouse were acquired by the in-house BLT system.

Our team and our industrial partner Xstrahl Inc. developed the second BLT system, a commercial configuration MuriGlo (Xstrahl Inc., Suwanee, Georgia, United States), built from the same concept of the in-house system with improved capability of acquiring full 360-deg optical projection. The 360-deg imaging acquisition is especially useful for abdominal imaging, such as pancreas site, to obtain all the essential projections for BLT reconstruction. Optical signal emitted from the animal is reflected from a three-mirror system to a fixed 45-deg mirror, passing through filter and captured by a CCD camera (iKon-M 934; Andor Technology, Belfast, United Kingdom). The three-mirror system supports the 360 deg multiple projection imaging. The fixed mirror and iKon-M934 CCD allow for a compact configuration. The focal plane was set at $\sim 16\mathrm{mm}$ above the mouse bed with the three-mirror system placed at 0 deg, and the pixel scale is 0.10 mm per CCD pixel. The optical path from the focal plane to the front surface of the lens is 40 cm. The in-vivo data of the mice implanted with light source in pancreas was acquired by the MuriGlo system. Details of both systems utilized in this work, together with system characterization and evaluation have been outlined previously.^18,21,23

2.5.

Quantification of System-Specific Cell Spectrum

Because of the multispectral BLT approach, it is important to quantify the optical system spectral response and the emission spectrum of bioluminescent tumor cells. The measurement via our optical systems includes spectral responses of the system and cell, and the resulted spectrum is referred to as the system-specific cell spectrum. Therefore, the wavelength dependent BLIs can be normalized to the measured spectrum weighting, used as the input data for the BLT reconstruction.

The system-specific spectral weights of GL261-Luc2 cells at 590, 610, 630, and 650 nm in petri dishes with cells $>80\%$ confluency at concentration of 0.75 mg of D-Luciferin per 1 mL of PBS at 37°C as set by the in-house optical system were measured. To eliminate the change of the in vitro spectral signal as function of luciferin incubation time, open field images without filters were taken before and after each spectral BLI to quantify the in vitro signal variation over time. The time point for each image was recorded and the open field images were used to generate an in vitro time-resolved signal curve. The intensity of multispectral BLIs was corrected based on time-variant data from the time-resolved curve. The measured spectrum of the GL261-Luc2 at 590, 610, 630, and 650 nm are 1, $0.916\pm 0.014$, $0.674\pm 0.019$, $0.389\pm 0.012$ ($n=20$), respectively. The same procedure was applied to measure the spectrum of the self-luminous light sources. At 590, 610, 630, and 650 nm, the spectrum weights for the light sources in the mice [as shown in Figs. 7(a) and 7(b)] are 0.942, 1, 0.893, 0.633, and 0.949, 1, 0.886, 0.626, respectively. The spectral response of the in-house and MuriGlo system was also investigated and as the difference is insignificant, the measured spectrum was applied to both systems.

2.6.

In Vivo Bioluminescence Imaging

The GBM-bearing mouse was subject to multiprojection and multispectral bioluminescence imaging with the in-house optical system two weeks after the cell implantation. In preparation for bioluminescence imaging session for GBM-bearing mouse, mouse hair was shaved with a clipper, followed by hair depilation. D-Luciferin ($125\mu \mathrm{L}$, $30\mathrm{mg}/\mathrm{mL}$ for 25 g mouse to reach $150\mathrm{mg}/\mathrm{kg}$, XenoLight D-Luciferin ${\mathrm{K}}^{+}$ Salt, PerkinElmer Inc., Waltham, Massachusetts, United States) was administrated via intraperitoneal injection. BL imaging was conducted 10 min after the D-Luciferin injection. Mouse was anesthetized with 1% to 2% isoflurane (Fluriso, MWI Veterinary Supply Co. Boise, Idaho, United States) in oxygen while imaging. Multiprojection (0 deg, 90 deg, and $-90\mathrm{deg}$) and multispectral (590, 610, 630, and 650 nm) BLIs were acquired at $8\times 8$ binning ($0.96\mathrm{mm}/\text{pixel}$ at imaging plane), $4\times$ preamplifier gain, and 1 MHz readout rate. Because the in vivo signal at 590 nm was weak compared with those at other wavelengths, which would affect the stability of the BLT reconstruction, the images at 610, 630, and 650 nm were only used for this study. Open field images were acquired before and after each spectral image to build time-resolved curve for in vivo bioluminescence signal among different projections,¹⁸ which was used to quantify the bioluminescence signal change during the bioluminescence imaging course.

The mouse bearing self-luminous light source in pancreas was subject to multiprojection and multispectral BLI ($8\times 8$ binning, $0.8\mathrm{mm}/\text{pixel}$ at imaging plane) in MuriGlo one week after the light source was implanted. The procedures of BLI acquisition for the pancreatic light source study were similar to those used for the GBM study. Since the implanted light source is closer to the surface of abdomen than that of dorsum, the supine position was chosen for the BLI acquisition. Due to the location of light source, no or minimum signal was detected at 90-deg and 180-deg projections, only the 0 deg and $-90\mathrm{deg}$ projections were chosen as the surface input data for the BLT reconstruction. Images at multiple projections were taken to retrieve the BB positions on mouse bed for mapping the BLI onto 3D mesh surface generated from the CBCT imaging as the input data for BLT reconstruction.

2.7.

CBCT Imaging

After the bioluminescence imaging, the mouse bed with an animal was transferred from the optical system to the small animal radiation research platform (SARRP; Xstrahl Inc., Suwanee, Georgia, United States) designed for CBCT-guided irradiation study.¹⁸ CBCT image was acquired by rotating the animal with a 4D robotic-controlled bed (3-axis translation and 360-deg rotation) between an amorphous silicon flat-panel detector and an x-ray source at 0.4-mm focal spot, 65 kVp and 0.7 mA with 1-mm-thick aluminum filter. Studied animal was anesthetized with 1% to 2% isoflurane in oxygen during the animal transportation and imaging course. The acquired SARRP CBCT image was used to provide anatomical structure of animals to generate a computational model (tetrahedral mesh) for the BLT reconstruction. For the pancreatic mouse model, the mesh consisted of 43,580 nodes corresponding to 237,110 linear tetrahedral elements, whereas for the GBM model it consisted of 6294 corresponding to 32,704 linear tetrahedral elements.

Contrast CBCT was used to assess the actual volume of the in vivo GBM,¹⁸ and the center of mass of the contrast-labeled tumor volume was calculated for BLT localization assessment. An in-house high-resolution CBCT system,²⁴ providing superior image contrast than that of SARRP CBCT, was employed to identify the contrast-labelled GBM. The mouse was imaged 1 min after the contrast injection at dose of $2\mathrm{gI}/\mathrm{kg}$ (Iodixanol, retro-orbital injection at 320 mgI/mL, Visipaque, GE Health Care, Chicago, Illinois, United States).

2.8.

Data Mapping for Multiprojection BLIs

Because SARRP CBCT imaging defines the coordinate used for BLT reconstruction, our geometry calibration method was published in Refs. 18 and 25 were used to map the 2D BLIs acquired at different projections onto the discretized mesh surface of the animal CBCT image. The mapped BLIs were used as the input data for BLT reconstruction. The method has two steps: (1) mapping the CBCT coordinate to the 3D optical coordinate with rigid transformation and then (2) projecting the 3D optical coordinate to the 2D optical (CCD) imaging plane. After the 3D CBCT and 2D optical coordinates are registered, for a given projection, the surface BLI can then be mapped to the CBCT image. The data mapping process requires knowledge of the geometrical parameters of optical system. The BBs on the mouse bed can be located in both CBCT and 2D optical images. An optimization routine with the constrained multivariable optimization function (fmincon, MATLAB, The MathWork Inc., Natick, Massachusetts, United States) was developed to retrieve the geometrical parameters by minimizing the difference between the calculated and measured BB positions in the 2D optical coordinate; the BB positions in 2D images were used as the measured positions, and the corresponding BB positions retrieved from the optimization routine based on the optimized geometrical parameters and the BB positions in 3D CBCT were used as the 2D calculated positions. The geometric calibration was performed for each animal imaging session to ensure accurate data mapping for BLT reconstruction.

The BLIs were then mapped onto the 3D mesh surface of the imaged mouse generated from the CBCT image. At the overlapped region on the mesh surface, for a given node between two mapped images from different projections, the maximum value of the two images was chosen as the value on that surface node. The mapped surface data larger than 10% of the maximum value among all the surface points were used as input data for BLT reconstruction.

2.9.

Algorithm

The final algorithm needed to complete the entire framework is one that calculates both the optical properties, given the surface fluence rate data at multiple wavelengths, as well as the location of the source within the subject. To resolve unknown optical properties, DOT is a commonly used imaging technique, known as named due to the fact that the transport of light through tissue at wavelengths in the visible and near-infrared (NIR) bands become nearly isotropic, therefore is well defined by photon diffusion. The software package, NIR fluorescence and spectral tomography (NIRFAST), was utilized to allow for the simulation of light propagation within biological tissue using a finite element method (FEM), documented elsewhere.² Within the package are a number of forward and inverse models, and a spectrally constrained case is utilized in this work.²⁶ The algorithm has been modified to take the estimated internal bioluminescence source location and surface fluence data with the goal of directly estimating the concentrations of total hemoglobin (cTHb) as well as scattering power and amplitude,²⁶ where cTHb is the sum of oxy-hemoglobin and deoxy-hemoglobin. The assigned chromophore concentrations can be used to calculate the underlying wavelength dependent absorption coefficient of tissue through the use of extinction coefficients of individual chromophores. This can similarly be done using the scattering power and amplitude to obtain a corresponding reduced scattering coefficient at each wavelength using Mie scattering theory. Using a continuous wave model, the Jacobian (function to allow mapping of measured data to internal optical properties) is calculated and then inverted using the Moore–Penrose generalized inverse, which is typically more suitable toward underdetermined problems.²⁷ The Moore–Penrose generalized inverse finds the “best fit” or minimum norm solution to a system of linear equations, the implementation of which has been extensively detailed elsewhere.² From this an update in optical properties is calculated and the whole process is repeated, new boundary data is simulated and compared with the original data to calculate a projection error. This error is then used as a stopping mechanism for when it is considered that convergence has occurred, typically within 2% change. Regularization is used alongside the Moore–Penrose generalized inverse with a starting value of 0.1, which is the standard value used as detailed elsewhere.² All of the reconstruction parameters used for all the data presented are consistent across all model, with the only variable being the initial estimate of the optical properties.

Algorithm 1 outlines the steps for image reconstruction and is briefly summarized: First, multispectral data are collected as described already, which are used with an initial “estimate” of underlying optical properties to perform tomographic reconstruction of the spatial bioluminescence light distribution as achieved using a compressive sensing conjugate gradient (CSCGNW) algorithm described elsewhere.²⁹ The center of mass of the reconstructed bioluminescence source is then defined as the “source position” and using the DOT algorithm and the same dataset, the optical properties of the medium are calculated. From this, if the error conditions have not yet been met, the optical properties are updated by the average value across the entire model or of a particular predefined volume within. Using these new updated optical properties, the CSCGNW algorithm is then used again to reconstruct the new spatial bioluminescence light distribution and whole process is repeated iteratively. As with the inverse problem outlined already, a stopping condition is put in place to stop the algorithm when it is considered to have converged. The stopping criteria is the same as that used in the DOT algorithm whereby a projection error is calculated between the original and modelled data using the updated optical properties. The iterative process is continued until the change in projection error is below a tolerance which is typically set to 2%. When modeling the new source location for the update of optical parameters using the DOT algorithm, first the reconstructed bioluminescence source size and location is represented as its FWHM and the distributed source (i.e., uniform intensity) at the recovered FWHM is then set as the new update.

Algorithm 1

Outline of image reconstruction