Here we present a protocol to perform preclinical positron emission tomography-based radiotherapy in a rat glioblastoma model using algorithms developed in-house to optimize the accuracy and efficiency.
A rat glioblastoma model to mimic chemo-radiation treatment of human glioblastoma in the clinic was previously established. Similar to the clinical treatment, computed tomography (CT) and magnetic resonance imaging (MRI) were combined during the treatment-planning process. Positron emission tomography (PET) imaging was subsequently added to implement sub-volume boosting using a micro-irradiation system. However, combining three imaging modalities (CT, MRI, and PET) using a micro-irradiation system proved to be labor-intensive because multimodal imaging, treatment planning, and dose delivery have to be completed sequentially in the preclinical setting. This also results in a workflow that is more prone to human error. Therefore, a user-friendly algorithm to further optimize preclinical multimodal imaging-based radiation treatment planning was implemented. This software tool was used to evaluate the accuracy and efficiency of dose painting radiation therapy with micro-irradiation by using an in silico study design. The new methodology for dose painting radiation therapy is superior to the previously described method in terms of accuracy, time efficiency, and intra- and inter-user variability. It is also an important step towards the implementation of inverse treatment planning on micro-irradiators, where forward planning is still commonly used, in contrast to clinical systems.
Glioblastoma (GB) is a malignant and very aggressive primary brain tumor. GB is a solid heterogeneous tumor typically characterized by infiltrative boundaries, nuclear atypia, and necrosis1. The presence of the blood-brain-barrier and the brain's status as an immune-privileged site makes the discovery of novel targets for chemo- and immunotherapy a daunting task2,3,4. It is noteworthy that the treatment of GB patients has barely changed since the introduction, in 2005, of the Stupp protocol that combines external beam radiation therapy (RT) with concomitant temozolomide, usually followed by adjuvant temozolomide5. Typically, the Stupp protocol is preceded by maximal surgical resection. Therefore, alternative treatment approaches are of pivotal importance.
Current radiation therapy for glioblastoma patients delivers a uniform radiation dose to the defined tumor volume. In radiation oncology, there is an important dose-response correlation for glioblastoma with increasing dose, which seems to cap around 60 Gy, due to increased toxicity to the normal brain6,7. However, tumors can be very (radiobiologically) heterogeneous, with gradients of oxygen level and/or large differences in cellular density. Metabolic imaging techniques, such as PET, can visualize these biological features and can be utilized to customize the dose prescription. This approach is known as dose painting RT. This term was introduced by Ling et al. in 2000. The authors defined dose painting RT as producing "exquisitely conformal dose distributions within the constraints of radiation propagation and scatter"8.
There are two types of dose painting RT, dose painting by contours (DPBC), by which a dose is prescribed to a set of nested sub-volumes, and dose painting by numbers (DPBN), whereby a dose is prescribed at the voxel level. The dose distribution for DPBN RT can be extracted from functional images. The dose in each voxel is determined by the intensity I of the corresponding voxel in the image, with a lower and upper limit, to make sure that, on the one hand, a sufficient dose is delivered to every part of the tumor. On the other hand, doses do not exceed an upper limit to protect organs at risk and avoid toxicity. The most straightforward method is a linear interpolation (see Eq. 1) between minimum dose Dmin and maximum dose Dmax, proportionally varying between minimum intensity Imax and maximum intensity within the target volume9,10
Eq. 1
Because there is some skepticism about the quality assurance of DPBN RT, the deposition of the dose should be verified through preclinical and clinical research10. However, only limited data can be acquired from clinical trials, and it has been hypothesized that more insights can be obtained by downscaling to laboratory animals11,12. Hence, preclinical studies utilizing precision image-guided radiation research platforms that allow coupling with some very specific techniques, such as autoradiography, are suited for examining open issues and paving the way towards personalized medicine and novel treatment strategies, such as dose painting RT13,14. However, the interpretation of preclinical data must be performed with caution, and drawbacks of these preclinical setups have to be considered14.
Micro-irradiation systems, such as the Small Animal Radiation Research Platform (SARRP), are equipped with similar technologies as their clinical counterpart. They include on-board cone-beam CT (CBCT) imaging, a preclinical treatment-planning system (PCTPS), and provide sub-millimeter precision. Clinical dose calculations are performed by inverse treatment planning, whereby one initiates from the desired dose distribution to determine the beams via an iterative algorithm. Preclinical irradiators often use forward planning. In forward planning, the required amount and angle of the beams are selected, and the PCTPS then calculates the dose distribution. The optimization of the plans is performed by manual iteration, which is labor-intensive15.
After 2009, novel developments have made the implementation of inverse planning on these research platforms possible16,17,18. To increase the similarity with the clinical method, a motorized variable rectangular collimator (MVC) was developed as a preclinical counterpart of the multi-leaf collimator. A two-dimensional dose painting method utilizing a variable collimator was published by Cho et al.19. This research group implemented a three-dimensional (3D) inverse treatment-planning protocol on a micro-irradiator and determined minimum and maximum doses for the target volume and a maximum dose for the organs at risk. These techniques have mainly been evaluated in silico, and their preclinical applications need to be explored.
This paper presents an in silico study to compare two methodologies for [18F]-fluoro-ethyl-L-tyrosine ([18F]FET) PET-based dose painting in a GB rat model20,21,22 using a small animal radiation research platform. These two methodologies are (1) sub-volume boosting using predefined beam sizes and (2) dose painting using a motorized variable collimator where jaw dimensions are modified based on the PET tracer uptake in the tumor volume. [18F]FET is a PET tracer often used in neuro-oncology because of its ability to detect brain tumors23. [18F]FET is an artificial amino acid that is internalized into tumoral cells but not incorporated into cell proteins. [18F]FET uptake corresponds with cell proliferation rate, tumor cell density, and angiogenesis24. As this is the most commonly used oncologic brain PET tracer in these authors' institute, this radiotracer was chosen to evaluate the new workflow.
The study was approved by the local ethics committee for animal experiments (ECD 18/21). Anesthesia monitoring is performed by acquiring the respiratory rate of the animals using a sensor.
1. F98 GB rat cell model
2. Confirmation of tumor growth
3. Multimodality imaging of target volume selection
NOTE: PET/MRI-guided irradiation requires the sequential acquisition of a multimodal dataset. After intravenous administration of the radiotracer, PET imaging is started, followed by contrast-enhanced T1-weighted MRI and finally a treatment-planning CT.
4. Image co-registration
NOTE: The co-registration is performed with a semi-automatic MATLAB code developed in-house. The code can be found on Github at https://github.com/sdonche/DosePainting. The different steps are described below.
5. Radiation treatment planning
NOTE: A MATLAB app and multiple MATLAB scripts were written for the radiation treatment planning. The code can be found on Github at https://github.com/sdonche/DosePainting. The different steps are explained below.
6. Plan evaluation
NOTE: To compare the two methods, calculate the dose-volume histograms (DVH) and Q-volume histogram (QVH) in the V50 PET volume. Here, a MATLAB script, developed in-house, was used. The code can be found on Github at https://github.com/sdonche/DosePainting.
The feasibility of PET- and MRI-guided irradiation in a glioblastoma rat model using the SARRP to mimic the human treatment strategy has been previously described20,21,22. While the animal was fixed on a multimodality bed made in-house, it was possible to create an acceptable radiation treatment plan combining three imaging modalities: PET, MRI, and CT. In these methods, an external software package (see the Table of Materials) was used to co-register the images using rigid-body transformations manually. The contrast-enhanced T1-weighted MRI and PET images were visually assessed from which the isocenters were manually selected. However, this methodology proved to be labor-intensive and certainly has an impact on the animals as they have to stay under general anesthesia during the multimodality imaging and the creation of a treatment plan. Therefore, the new methodology aims to automate specific steps in this process to reduce the overall variance and time required to create a radiation treatment plan.
In this paper, two methodologies are compared. Method 1 is very similar to the previously published methodology20,21,22 with a few adjustments (Table 1). However, in contrast to the previously published methodology, most of the process is automated using a MATLAB code developed in-house. Method 2 is a more sophisticated method in which a series of isocenters and jaw dimensions for the MVC will be determined based on the [18F]FET PET uptake (Figure 5). The isocontours for V50, V60, V70, V80, and V90 are shown in Figure 6.
Both methods were applied to three different cases (Figure 7). These cases can be divided into two different types: [18F]FET PET uptake in the infiltrative tumor front and the presence of tumor necrosis and [18F]FET PET uptake indicating no tumor necrosis. Case 1 can be described as a spherical homogeneous PET uptake, while Cases 2 and 3 have a ring-shaped uptake where the reduced PET-uptake is most likely necrotic tissue. Case 3 also shows an additional region growing out towards the dorsal region.
After calculating the setup parameters for both methods, the dose distributions for each case (Figure 8) were determined using the SARRP's PCTPS. The DVHs (Figure 9) can be obtained from the dose distributions in the volumes defined by the pixels with signal intensity higher than 0.50 × maximal PET signal intensity (in the bounding box). One can observe that the DVHs for Method 2 are systematically closer to the ideal dose distribution than those for Method 1. A substantial tumor volume receives insufficient irradiation in Cases 2 and 3 when treated with Method 1. Table 3 confirms these conclusions: the D90 and D50 values are considerably lower for Method 1 than for Method 2. The QVHs (Figure 10) can also be obtained from these dose distributions. Ideally, these curves make a sharp drop at a Q-value equal to one. Method 2 always results in dose distributions that are closer to the dose objective. Table 4 also demonstrates superior overall Q-factors for Method 2. The minimal dose (D90) of 2000 cGy has been achieved for all cases with Method 2, while it was not achieved with Method 1 in 2 cases. This means that the tumor volume received insufficient irradiation using Method 1.
Figure 1: Bounding box placement. The T1-weighted contrast enhancement is visible in the F98 GB rat model, and a generous bounding box is placed around the tumor using the MATLAB code developed in-house. Please click here to view a larger version of this figure.
Figure 2: T1-weighted contrast-enhancing tumor delineation: step 1. The tumor volume is delineated on the contrast-enhanced T1-weighted MRI using thresholding. Abbreviation: MRI = magnetic resonance imaging. Please click here to view a larger version of this figure.
Figure 3: T1-weighted contrast-enhancing tumor delineation: step 2. If multiple volumes are detected during the thresholding step, the largest volume is retained for further processing. Please click here to view a larger version of this figure.
Figure 4: Isocenter calculation for Method 1. Contrast-enhanced T1-weighted MRI, CT, and PET images are depicted. The blue and red circles represent the MRI- and PET-based isocenters, respectively. Abbreviations: MRI = magnetic resonance imaging; CT = computed tomography; PET = positron emission tomography. Please click here to view a larger version of this figure.
Figure 5: Explanation of jaw setup calculation. Step 1: the tumor volume is determined (blue dots, top image). Step 2: a plane (black grid) is created perpendicular to the incident beam (magenta line, top image) at specific couch and gantry positions. Step 3: the tumor voxels (blue dots, top image) are perpendicularly projected onto the aforementioned plane, resulting in a set of projected voxels (red dots). Step 4: determine the isocenter and jaw dimensions (green lines, bottom image) so that all the projected voxels are included within the rectangular beam defined by the two symmetrical jaws of the variable collimator (bottom image). These figures were generated in MATLAB. Please click here to view a larger version of this figure.
Figure 6: Tumor isocontours. Transaxial, coronal, and sagittal slices through the brain tumor with tumor volumes V50, V60, V70, V80, and V90 determined by the isocontours corresponding to 50%, 60%, 70%, 80%, and 90% of the maximum tumor uptake in the PET images. Abbreviations: TV = transaxial; COR = coronal; SAG = sagittal; PET = positron emission tomography. Please click here to view a larger version of this figure.
Figure 7: [18F]FET PET imaging for the three cases. The sagittal, transverse, and frontal views are displayed for all three cases. Please click here to view a larger version of this figure.
Figure 8: Dose distributions for both methods. Sagittal, transverse, and frontal views for all three cases are displayed for both Method 1 and Method 2. The dose distribution is shown together with the cone-beam CT imaging from the SARRP. Abbreviations: CT = computed tomography; SARRP = small animal radiation research platform. Please click here to view a larger version of this figure.
Figure 9: DVH curves for all cases. DVH curves (in cGy) are shown for Method 1, Method 2, and the Ideal Dose Map. Abbreviation: DVH = dose-volume histogram. Please click here to view a larger version of this figure.
Figure 10: Q-volume histogram for all cases. QVH curves are shown for Method 1, Method 2, and the Ideal Dose Map. Ideally, the calculated QVH must have a sharp drop at Q-value = 1 (Ideal dose map, blue line). Abbreviation: QVH = Q-volume histogram. Please click here to view a larger version of this figure.
Previous Method | Method 1 | Method 2 | ||
Tumour | Diameter | 5 mm | 7-8 mm | 7-8 mm |
PET | Resolution (mm) | 1.2 | 0.85 | 0.85 |
Base irradiation | Dose (cGy) | 2000 | 2000 | 2000 |
Target | CE T1 tumour | CE T1 tumour | V50 | |
Collimator (mm²) | 5×5 | 10×10 | MVC | |
Delivery | 3 non-coplanar arcs | 3 non-coplanar arcs | 16 beams | |
Couch positions | -45°, 0°, 45° | 0°, -45°, -90° | 0°, -45°, -90° | |
Irradiation boost or dose painting | Dose (cGy) | 500 | 800 | 800 |
Target | Max PET uptake | Max PET uptake | V60-V90 | |
Collimator (mm²) | 1×1 | 3×3 | MVC | |
Delivery | 3 non-coplanar arcs | 3 non-coplanar arcs | 40 beams | |
Couch positions | -45°, 0°, 45° | 0°, -45°, -90° | 0°, -45°, -90° |
Table 1: Method comparison. This table further clarifies Method 1, Method 2, and the Previous Method (referring to the method that has already been published)20,21,22. Methods 1 and 2 utilize a preclinical PET scanner27 with sub-millimeter spatial resolution, making it possible to visualize the tumor heterogeneity more clearly. At couch position -90°, it is only possible to use 60° out of 120° to avoid collision with the animal. Despite this drawback, this couch position has easier access to the tumor because it is situated in the right hemisphere. The other couch positions can make the full 120° rotations. Abbreviations: CE T1 = contrast-enhanced T1-weighted; MVC = motorized variable collimator; PET = positron emission tomography.
Couch Position | Gantry position | ||||||
0° | – | 20° | 40° | 60° | 80° | 100° | 120° |
-45° | – | 20° | 40° | 60° | 80° | 100° | 120° |
-90° | 0° | 20° | 40° | 60° | – | – | – |
Table 2: Beam setup for Method 2. The gantry and couch positions of all the different beams are displayed. V50 uses all configurations, whereas V60-V90 only use the configurations shown in bold.
D90 | D50 | D10 | ||
Case 1 | Ideal Dose Map | 2336.94 | 2461.21 | 2745.63 |
Method 1 | 2024.47 | 2389.75 | 2796.82 | |
Method 2 | 2164.21 | 2490.18 | 2747.64 | |
Case 2 | Ideal Dose Map | 2391.76 | 2540.55 | 2752.56 |
Method 1 | 1894.93 | 2127.86 | 2606.48 | |
Method 2 | 2322.11 | 2597.31 | 2848.03 | |
Case 3 | Ideal Dose Map | 2377.47 | 2556.7 | 2761.38 |
Method 1 | 1874.58 | 2103.78 | 2691.69 | |
Method 2 | 2354.03 | 2602.64 | 2907.41 |
Table 3: DVH values. D10, D50, and D90 were calculated as substitutes for maximum, mean, and minimal doses, respectively. Dx stands for the dose received by x% of the volume. Abbreviation: DVH = dose-volume histogram.
Q-factor | Case 1 | Case 2 | Case 3 |
Method 1 | 0.0898 | 0.1573 | 0.1773 |
Method 2 | 0.0572 | 0.057 | 0.0778 |
Table 4: Q-factors. The table displays the overall Q-factors for Method 1 and Method 2 for each case. The Q-factor will be zero if the delivered dose and prescribed dose are equal.
A rat GB model to mimic the chemo-radiation treatment in the clinic for glioblastoma patients was previously described20. Similar to the clinical method, CT and MRI were combined during the treatment-planning process to obtain more precise irradiation. A multimodality bed to minimize (head) movement was used when the animal was moved from one imaging system to another. Subsequently, PET imaging was added to the treatment-planning process, and PET-based sub-volume boosting could be successfully implemented21,22. The inclusion of a functional image modality, such as PET, in the treatment-planning process allows the visualization of the (biological) tumor heterogeneity. This facilitates the targeting of aggressive and/or radiation-resistant tumor regions. Although this method is feasible, it proved to be very labor-intensive because multimodal imaging, treatment planning, and dose delivery must be completed sequentially in a preclinical setting. Moreover, during this process, the animals have to stay under general anesthesia22. Therefore, it is essential to improve the efficiency of the preclinical treatment-planning process.
This paper presents a user-friendly semi-automatic algorithm to further optimize preclinical multimodal imaging-based radiation treatment planning. Co-registration between planning CT, MRI, and PET were automated, in combination with the detection of the target isocenters. Of note, the software tool should not be considered as a black box, and it is crucial to perform proper quality checks. The most critical step in this entire process is to evaluate the results of the automatic co-registration of planning CT, MRI, and PET that should be as accurate as possible. The output of the algorithm consists of the positions of the target isocenters and the jaw dimensions of the MVC for the different radiation beams. These values can be imported into the most recent version of the PCTPS.
This software tool was used to evaluate the accuracy and efficiency of PET-based dose painting on the micro-irradiator by using an in silico study design. The optimized treatment-planning process was superior to the previously described method21,22 in terms of time efficiency, intra- and inter-user variability, and accuracy. While conventional preclinical treatment planning, including multimodal imaging, can require up to 180 min22, this time could be reduced to ~80 min with both the semi-automatic methods presented in this manuscript. Moreover, human errors are more likely in the conventional treatment-planning process during manual co-registration and visual determination of the isocenters, resulting in larger intra- and inter-user variability. The automatic co-registration and detection of the target isocenters by the algorithm will reduce these intra- and inter-user variabilities. In addition, the optimized and automated workflow provides more accurate irradiation of the tumor volume. This is illustrated by the lower Q-factors (Table 4), which assesses the difference between the dose calculated/delivered by the PCTPS and the prescribed dose.
It is also noteworthy that the use of an MVC results in a reduced dose to the surrounding normal brain tissue, compared to collimators with a fixed beam size. This is illustrated in Figure 7 and is important to narrow the gap between clinical trials evaluating DPBN RT strategy (where multi-leaf collimators are used) and laboratory animal radiation research. However, we assume that dose delivery might be slightly slower when using an MVC to switch between beam positions and adjust the jaw dimensions for each individual beam. Finally, preclinical treatment planning is most often done by forward planning. The methodology described in this paper is a crucial step towards inverse planning, which is generally used in the clinic, and further narrows the gap between preclinical radiation research and the clinic.
This study also has some limitations. For the experiments described in this manuscript, the most commonly used amino acid PET tracer [18F]FET was used. When using other PET tracers to guide radiation treatment, the semi-automatic workflow should be properly examined because co-registration might be less accurate. Further, the impact of using a different voxel size for PET and/or MRI on treatment planning and dose delivery should be further investigated. In conclusion, the methodology described here to optimize the preclinical treatment-planning process has many advantages compared to the previously described method21,22. Using an in silico study design, it was proven that the novel workflow for preclinical multimodal treatment planning is more accurate in terms of dose delivery, more time-efficient, and shows less intra- and inter-user variability. These improvements are essential to narrow the gap between clinical and preclinical radiation research and for the development of new therapeutics and/or radiation therapy procedures for glioblastoma.
The authors have nothing to disclose.
The authors would like to thank Lux Luka Foundation for supporting this work.
Cell culture | |||
F98 Glioblastoma Cell Line | ATCC | CRL-2397 | https://www.lgcstandards-atcc.org/products/all/CRL-2397 |
Dulbeco's Modified Eagle Medium | Thermo Fisher Scientific | 22320-030 | |
Cell culture flasks | Thermo Fisher Scientific | 178883 | 75 cm² |
FBS | Thermo Fisher Scientific | 10270106 | |
L-Glutamine | Thermo Fisher Scientific | 25030-032 | 200 mM |
Penicilline-Streptomycin | Thermo Fisher Scientific | 15140-148 | 10,000 U/mL |
Phosphate-Buffered Saline (PBS) | Thermo Fisher Scientific | 14040-224 | |
Trypsin-EDTA | Thermo Fisher Scientific | 25300-062 | 0.05% |
GB Rat Model | |||
Ball-shaped burr | Foredom | A-228 | 1.8 mm |
Bone Wax | Aesculap | 1029754 | https://www.aesculapusa.com/en/healthcare-professionals/or-solutions/or-solutions-cranial-closure/hemostatic-bone-wax.html |
Ethilon | Ethicon | 662G/662H | FS-2, 4-0, 3/8, 19 mm |
Fischer F344/Ico crl Rats | Charles River | – | |
Insulin Syringe Microfine | Beckton-Dickinson | 320924 | 1 mL, 29 G |
IR Lamp | Philips | HP3616/01 | |
Meloxicam (Metacam) | Boehringer Ingelheim | – | 2 mg/mL |
Micromotor rotary tool | Foredom | K.1090-22 | |
Micropump system | Stoelting Co. | 53312 | Stoelting Stereotaxic Injector |
Stereotactic frame | Stoelting Co. | 51600 | |
Xylocaine (1%, with adrenaline 1:200,000) | Aspen | – | 1%, with adrenaline 1:200,000 |
Xylocaine gel (2%) | Aspen | – | 2% |
Animal Irradiation | |||
Micro-irradiator | X-Strahl | SARRP | Version 4.2.0 |
Software | X-Strahl | Muriplan | Preclinical treatment planning system (PCTPC), version 2.2.2 |
Small Animal PET | |||
[18F]FET | Inhouse made | – | PET tracer; along with Prohance: MRI/PET agent |
Micro-PET | Molecubes | Beta-Cube | https://www.molecubes.com/b-cube/ |
Small Animal MRI | |||
Micro-MRI | Bruker Biospin | Pharmascan 70/16 | https://www.bruker.com/products/mr/preclinical-mri/pharmascan.html |
30 G Needle for IV injection | Beckton-Dickinson | 305128 | |
PE 10 Tubing | Instech Laboratories Inc | BTPE-10 | BTPE-10, polyethylene tubing 0.011 x 0.024 in (0.28 x 60 mm), non sterile, 30 m (98 ft) spool, Instech laboratories, Inc Plymouth meeting PA USA- (800) 443-4227- http://www.instechlabs.com |
Prohance contrast agent | Bracco Imaging | – | 279.3 mg/mL, gadolinium-contrast agent (along with [18F]FET: MRI/PET agent) |
Tx/Rx Rat Brain – Mouse Whole Body Volumecoil | Bruker Biospin | – | 40 mm diameter |
Water-based Heating Unit | Bruker Biospin | MT0125 | |
Consumables | |||
Isoflurane | Zoetis | B506 | Anesthesia |
Insulin Syringe Microfine | Beckton-Dickinson | 320924 | 1 mL, 29 G |
Image Analysis | |||
MATLAB | Mathworks | – | Version R2019b |
PMOD | PMOD technologies LLC | Preclinical and molecular imaging software |