ZHANG TIEZHI (US)
BIEGALSKI STEVEN (US)
OANCEA CRISTINA (US)
KANG MINGLEI (US)
LIU WEI (US)
SIMONE II (US)
WAHLS NIKLAS (US)
LIN LIYONG (US)
LIU RUIRUI (US)
HARRISON NATHAN (US)
BRADLEY JEFFREY (US)
HIGGINS KRISTIN (US)
DYNAN WILLIAM (US)
ZHOU JUN (US)
YANG XIAOFENG (US)
CHARYYEV SERDAR (US)
STANFORTH ALEXANDER (US)
ZHOU SHUANG (US)
ZHANG TIEZHI (US)
BIEGALSKI STEVEN (US)
OANCEA CRISTINA (US)
KANG MINGLEI (US)
LIU WEI (US)
SIMONE II CHARLES (US)
WAHLS NIKLAS (US)
CLAIMS Therefore, the following is claimed: 1. A radiation therapy method, comprising: receiving a beam of particles; directing the beam of particles to a patient specific sparse passive filter to form an adjusted beam of particles, wherein the patient specific sparse passive filter is configured to modulate the beam of particles, wherein the patient specific sparse passive filter is formed based on a simultaneous optimization of a dose of particles from the beam of particles, a dose-averaged dose rate (DADR) of particles from the beam of particles, and dose-averaged linear energy transfer (LETd) of the particles from the beam of particles to target a target area of a patient and substantially spare organs at risk (OARs); and administering the adjusted beam of particles to the target area of the patient. 2. The radiation therapy method of claim 1, wherein the particles are high energy charged particles, optionally wherein the high energy charged particles are electrons, protons, or heavy ions. 3. The radiation therapy method of claim 1, wherein the particles are protons. 4. The radiation therapy method of claim 1, wherein the radiation therapy method is proton FLASH radiotherapy. 5. A method for treating cancer in a patient, the method comprising administering to the patient at least one fraction of proton ultra-high dose rate radiotherapy (FLASH), wherein the fraction of the proton beam pass through a patient specific sparse passive filter prior to being administered to the patient, wherein the patient specific sparse passive filter, is formed based on a simultaneous optimization of a dose of protons from the beam of protons, a dose-averaged dose rate (DADR) of protons from the beam of protons, and dose-averaged linear energy transfer (LETd) of the protons from the beam of protons to target a target area of a patient and substantially spare organs at risk (OARs). 6. A system for radiation therapy, comprising: a particle source for a beam of particles; and a patient specific sparse passive filter, wherein the patient specific sparse passive filter is configured in the system to receive the beam of particles, wherein the patient specific sparse passive filter is configured to modify the beam of particles to form an adjusted beam of particles, wherein the patient specific sparse passive filter is formed based on a simultaneous optimization of a dose of particles from the beam of particles, a dose-averaged dose rate (DADR) of particles from the beam of particles, and dose-averaged linear energy transfer (LETd) of the particles from the beam of particles to target a target area of a patient and substantially spare organs at risk (OARs). 7. The system for radiation therapy of claim 6, wherein the particles are high energy charged particles, optionally wherein the high energy charged particles are electrons, protons, or heavier ions than protons. 8. The system for radiation therapy of claim 6, wherein the particles are protons. 9. The system for radiation therapy of claim 6, wherein the radiation therapy method is proton FLASH radiotherapy. 10. The system for radiation therapy of claim 6, wherein components of the patient specific sparse passive filter are partially or entirely recessed within a nozzle adjacent a patient. 11. The system for radiation therapy of claim 10, wherein positioning the patient specific sparse passive filter recessed within the nozzle increase the dose rate about 40% or more as compared to the patient specific sparse passive filter positioned outside of the nozzle. 12. A method of optimizing an administration plan in particle FLASH radiotherapy, comprising: simultaneously optimizing a dose of particles from the beam of particles, a dose- averaged dose rate (DADR) of particles from the beam of particles, and dose-averaged linear energy transfer (LETd) of the particles from a beam of particles to a clinical target volume (CTV), beam-specific planning target volumes (BSPTVs), and organs at risk (OARs), wherein the optimization includes iteratively adjusting a geometry of patient-specific sets of geometric modulating and compensating components for a patient specific sparse passive filter, and the weight of a particle beam, optionally the weight of a proton pencil beam spot map, wherein simultaneously optimizing the dose of particles from the beam of particles, the DADR of particles from the beam of particles, and the LETd of the particles from the beam of particles, wherein the simultaneously optimizing is designed to reduce the dose of particles from the beam of particles, the DADR of particles from the beam of particles, and the LETd of the particles from a beam of particles in the OARs as compared to intensity modulated particles therapy; selecting an optimized patient specific sparse passive filter and an optimized weight of a particle beam optionally a weight of a proton pencil beam spot map; and implementing particle FLASH radiotherapy using the optimized patient specific sparse passive filter and the optimized weight particle beam, optionally the weight of a proton pencil beam spot map. 13. The method of claim 12, wherein the particles are high energy charged particles, optionally wherein the high energy charged particles are electrons, protons, or heavier ions than protons. 14. The method of claim 12, wherein the particles are protons. 15. The method of claim 12, wherein the particle FLASH radiotherapy is proton FLASH radiotherapy. 16. A method of designing a patient specific sparse passive filter, comprising: receiving a scan of a patient; determining an initial geometry of a sparse passive filter based at least in part on the scan; determining a dose influence matrix and an LET influence matrix; in parallel with determining the dose influence matrix and the LET influence matrix, simulating a plurality of geometry variations using a particle simulation; and optimizing output data from the particle simulation to determine an optimized geometry, the optimization being based at least in part on the dose influence matrix and the LET influence matrix. 17. The method of claim 16, further comprising forming the patient specific sparse passive filter. 18. The method of claim 16, wherein the initial geometry is determined by applying a ray tracing algorithm to the scan. 19. The method of claim 16, wherein optimizing the output data from the particle simulation further comprises optimizing a dose of protons from the beam of protons, a dose-averaged dose rate (DADR) of protons from the beam of protons, and dose- averaged linear energy transfer (LETd) of the protons from a beam of protons. 20. A patient-specific sparse passive filter for simultaneous intensity and energy modulation in proton therapy, the patient-specific sparse passive filter designed by the process of: determining an initial geometry of a sparse passive filter based at least in part on a scan of a patient; determining a dose influence matrix and an LET influence matrix; simulating a plurality of geometry variations using a particle simulation; and optimizing output data from the particle simulation to determine an optimized geometry; the optimization being based at least in part on the dose influence matrix and the LET influence matrix. 21. The patient-specific sparse passive filter of claim 20, wherein the process further comprising receiving the scan of a patient. 22. The patient-specific sparse passive filter of claim 20, wherein the process further comprising applying a ray tracing algorithm to the scan of the patient, and determining the initial geometry based at least in part on a result of the ray tracing algorithm. 23. The patient-specific sparse passive filter of claim 20, wherein the process further comprising optimizing a dose of protons from the beam of protons, a dose-averaged dose rate (DADR) of protons from the beam of protons, and dose-averaged linear energy transfer (LETd) of the protons from a beam of protons from the output data from the particle simulation. 24. The patient-specific sparse passive filter of claim 20, wherein the process for determining a dose influence matrix and an LET influence matrix and simulating a plurality of geometry variations using a particle simulation are accomplished in parallel. 25. The patient-specific sparse passive filter of claim 20, wherein the process further comprising fabricating the patient-specific sparse passive filter based at least in part on the optimized geometry. 26. A system for designing a patient-specific sparse passive filter, comprising: at least one computing device comprising a processor and a memory; and machine-readable instructions stored in the memory that, when executed by the processor, cause the computing device to at least: receive a scan of a patient; determine an initial geometry of a sparse passive filter based at least in part on the scan; determine a dose influence matrix and an LET influence matrix; in parallel with determining the dose influence matrix and the LET influence matrix, simulate a plurality of geometry variations using a particle simulation; and optimize output data from the particle simulation to determine an optimized geometry; the optimization being based at least in part on the dose influence matrix and the LET influence matrix. 27. The system of claim 26, wherein the dose influence matrix and the LET influence matrix are determined using a Monte Carlo particle simulation or an analytical dose engine. 28. The system of claim 26, wherein the machine-readable instructions which cause the at least one computing device to optimize output data from the particle simulation further cause the at least one computing device to optimize a dose of protons from the beam of protons, a dose-averaged dose rate (DADR) of protons from the beam of protons, and dose-averaged linear energy transfer (LETd) of the protons from a beam of protons. 29. The system of claim 26, wherein the machine-readable instructions which cause the at least one computing device to determine an initial geometry, further cause the at least one computing device to apply a ray tracing algorithm to the scan to determine the initial geometry. 30. The system of claim 26, wherein the machine-readable instructions, when executed, further cause the at least one computing device to at least send the optimized geometry to a fabrication system. 31. A radiation therapy device, comprising: a particle source for a beam of particles; and a nozzle that receives the beam of particles, wherein the nozzle includes a filter recessed within the nozzle. 32. The radiation therapy device of claim 31, wherein the radiation therapy device is a particle FLASH radiotherapy device. 32. The radiation therapy device of claim 32, wherein a particle FLASH radiotherapy device is a proton FLASH radiotherapy device. 33. The radiation therapy device of claim 31, wherein the filter is a patient specific sparse passive filter, wherein the patient specific sparse passive filter is configured in the system to receive the beam of particles, wherein the patient specific sparse passive filter is configured to modify the beam of particles to form an adjusted beam of particles. 34. The radiation therapy device of claim 31, wherein the particles are high energy charged particles other than protons. 35. The radiation therapy device of claim 34, wherein high energy charged particles other than protons are electrons or atomic nuclei that are heavier than protons. 36. The radiation therapy device of claim 35, wherein atomic nuclei that are heavier than protons are helium, lithium, carbon, or neon. 37. The radiation therapy device of claim 34, wherein positioning the patient specific sparse passive filter within the nozzle increases the dose rate up to 40% or more as compared to the patient specific sparse passive filter positioned outside of the nozzle. |
Here, η t & η o are the reference doses, N t & N o are the number of voxels, and α t & α o are the penalty factors for target and OAR, respectively. Values for d i and DADR i are given by equations (2.4) and (2.5), respectively. D ij is the influence matrix of dose, L ij is the influence matrix of LET d , w is the spot weight in MU, i is the voxel index, and j is the spot index. I nozzle , T min & N MU are nozzle current, minimum spot duration and number of protons per MU, respectively. Sparse Ridge Filters (also referred to as “sparse passive filter”) [0099] Regular ridge filters, designed using the IPO-IMPT framework provide increased DADR for some OARs while maintaining tumor coverage. However, the optimization does not take depth modulation into account. Sparse ridge filters, from which some pins are omitted, provide a means to further increase the DADR for optimal FLASH sparing. Removing filter pins at specific locations preserves a higher proton flux, while the remaining filter pins still provide adequate SOBP dose coverage to the BSPTV. [0100] To generate the sparse ridge filters, the dose influence matrices are calculated for a regular ridge filter and for a range compensator alone with no pins. The filter pin location map is used as the proton spot map, so that the dose of each beamlet reflects the contribution of a specific ridge filter pin. Using these two dose influence matrices, it is possible to obtain an optimized IPO-IMPT plan. The spot weighting factors can then be derived as well. If the pin at location j results in where ƒ j is a user-defined threshold, is the weighting factor for filter pin location ^^ of the regular ridge filter and is the weighting factor for pin location j of the filter compensator, the pin is kept; otherwise, the pin is removed. The sparse ridge filter is generated from this process. After the pin locations are selected, the sparse ridge filter design is generated. The sparse filter design allows higher DADRs for OARs, including lung and heart. Example Filters Design and Treatment Plans [0101] To demonstrate the IPO-IMPT framework, we designed ridge filters and developed treatment plans for three example lung cancer patients. Patient-specific ridge filter and range shifter assemblies were designed to achieve conformal target dose coverage using a 250 MeV proton beam. The BSPTV was created with 5% range uncertainty and 5 mm setup uncertainty. For our scanning beam proton therapy system, using a minimum duration of 1 millisecond and a constant current 300 nA, a value of 300 was taken as the minimum MU. The clinical target volume (CTV) received a prescribed dose of 50 Gy (10 Gy × 5 fractions) with a maximum allowable dose for hotspots corresponding to 125% of the prescription dose (62.5 Gy). For all three patients, lung and heart were considered as OARs. For Patients 2 and 3, esophagus was also considered. Three beam angles were considered for each patient. [0102] IPO-IMPT plans were generated for regular and sparse ridge filters at each beam angle and compared with conventional IMPT plans, as detailed in the Results section. A preliminary dose verification with a patient-specific ridge filter was also conducted through the experiment as detailed in the Result section. To generate the evaluating structures, Heart_eva and Lung_eva, we first created a uniform 5 mm expansion of the BSPTV. The 5 mm BSPTV expansion was chosen to include the gradual dose fall off beyond the BSPTV, recognizing that the dose within this margin region may exceed the lower threshold for a FLASH effect. Next, the CTV was removed from the expanded BSPTV and Lung_eva was defined as the overlap between this and the lung. The Heart_eva and Esophagus_eva structures were generated using a similar approach. The rationale for using only the defined Heart_eva, Lung_eva, and Esophagus_eva volumes, rather than the whole heart and lung, was that evaluation of a very large structure might mask the significance of a high dose or high dose rate due to a large volume with a low dose and low dose rate. For multiple beam plans, the overall evaluating structure is the Boolean union of the evaluation structures for each beam. [0103] For each plan, the distribution of dose, DADR, and LETd were calculated and corresponding volume histograms were generated. The FLASH effect has been reported to have a dose threshold between 4 Gy to 10 Gy 39–42. Here, 4 Gy per fraction per field was used as a conservative estimate. The FLASH dose rate threshold has been reported to be between 40 and 100 Gy/s. Here, 40 Gy/s was used. Each field independently meets the dose and dose rate contraints for the FLASH effect. These thresholds can be modified as knowledge of the FLASH effect improves. [0104] For generating the dose rate volume histograms, the DADR were assigned as zero for the voxels that do not meet the dose threshold. Thus, the fraction of volume achieving FLASH can be directly observed by inspection of the DADR volume histogram, as only the voxels that meet the dose and dose rate thresholds contribute to the histogram. IPO-IMPT with Regular Ridge Filters [0105] To demonstrate the functionality of the IPO-IMPT framework, regular ridge filters were designed and treatment plans for three sample lung cancer patients were developed. Patient 1 had a central lung tumor, very close to the heart. Heart and uninvolved lung were OARs. Patient 2 had a metastatic tumor in the right lower lobe and Patient 3 had a tumor in the subcarinal lymph node. The esophagus was an additional OAR in both these patients. [0106] A single-beam IPO-IMPT plan was generated for Patient 1, with a primary goal of reducing LETd to heart while maintaining target coverage. The target coverages for the IPO-IMPT and IMPT plans are similar. However, the IPO-IMPT framework resulted in a marked reduction of LETd in the heart. For comparison, a multi-beam plan was constructed for the same patient, where the primary goal was to optimize DADR, while maintaining adequate dose and LETd optimization. Together, the results demonstrate that adoption of the IPO-IMPT framework, in combination with regular ridge filters, results in at least modest improvements to DADR and LETd for OARs, while maintaining tumor coverage and meeting other constraints. [0107] Patients 2 and 3 were chosen to illustrate the potential to spare the esophagus. Using the regular ridge filter approach, almost 100% of the Esophagus_eva structure meets the 40 Gy/s FLASH threshold. This very high coverage was seen with both IPO-IMPT or IMPT. For Patient 2, IPO-IMPT modestly decreased LETd for heart and esophagus and increased in DADR for heart (97% FLASH coverage with IBO-IMPT versus 90% for IMPT). [0108] Dose distributions of FLASH plans and conventional IMPT plans were also compared for each patient. Dose distributions to the CTV and OARs are generally comparable. For completeness, the FLASH plans were also compared to the original VMAT plans for Patients 2 and 3. Again, dose distributions are comparable. IPO-IMPT with Sparse Ridge Filters [0109] Regular ridge filters, originally designed for dose optimization only, cannot fully realize the benefits of the IPO-IMPT framework, because spot-specific dose-depth modulation is not optimized. To address this, sparse ridge filter designs were explored, in which some pins are removed using the heuristic decision process described above. [0110] For Patient 1, a fully optimized IPO-IMPT plan was generated with sparse ridge filters and multiple beams. An IMPT-optimized plan using regular ridge filters was used for comparison. The OAR optimization constraints for the two plans were the same. Tumor coverage was maintained and hotspots were well controlled with both plans. However, the IPO-IMPT plan based on sparse ridge filters results in a marked improvement to DADR in the OARs. The volume that received a dose rate of ≥40 Gy/s increased by 31% for Heart_eva and by 50% for Lung_eva 50%. The LET d for the two plans was substantially the same. Together, results show that the use of sparse ridge filters and multiple beams helps realize the full potential of the IPO-IMPT framework. [0111] A separate set of optimized single-beam plans using sparse ridge filters was demonstrated. The increased DADR for lungs using sparse ridge filters versus regular ridge filters is evident. The individual plans have some hotspots within BSPTV (which slightly exceed the 125% prescription dose), but sequential delivery as SBRT fractions reduces these and improves target coverage. Together, dose coverage is similar to the multi-field plan, but with better FLASH sparing due to increased volumes that meet the 40 Gy/s dose threshold in each field and fraction. [0112] Sparse ridge filter-based plans were also developed for Patients 2 and 3. A comparison of IPO-IMPT optimized plans based on regular and sparse ridge filters was conducted. The FLASH plan with a sparse ridge filter further increases the DADR to the esophagus while maintaining similar tumor coverage and meeting other constraints. Preliminary dose verification with a patient-specific ridge filter [0113] To verify the ability of the ridge filter assembly to deliver the predicted dose, proton dose measurements were performed. The ridge filter assembly, which includes filter pins and a compensator, was placed on the T0 beam axis. A range shifter, solid water, and an ionization chamber array were placed downstream. A treatment plan optimized for Patient 1 and designed to provide a uniform dose to the CTV was delivered. The calculated dose distribution was 25 mm depth from the solid water surface. The total gamma passing rate was 92.9% (3mm/3%, 10% threshold) for the absolute doses, which exceeds the standard patient QA passing criteria of 90%. Results provide a preliminary demonstration that the ridge filter assembly can facilitate the delivery of a clinically acceptable dose distribution. Measurement of dose rate and LET, using a novel time- resolved and spatially resolved detector, is in progress. Discussion [0114] In this feasibility study, the use of the IPO-IMPT framework and optimized ridge filters were shown capable of improving DADR and LETd, for lung and heart, relative to a plan generated using a standard IMPT approach. The IPO-IMPT framework, which explicitly incorporates objective functions of dose, DADR, and LETd, provides degenerate solutions for patient-specific ridge filter and spot maps while providing an ability to study the contribution of each term. [0115] Optimization of DADR and LETd, while maintaining a similar dose distribution, is crucial for disentangling the biological contributions of DADR and LET from that of dose per se. For OARs such as great vessels, which have a maximum tolerated dose close to the prescription dose, increasing the DADR above the FLASH threshold (≥ 40 Gy/sec) may be the best selection. Alternatively, lower dose or LET might be a better option for OARs, such as spinal cord, which have a maximum tolerated dose that is smaller than the prescription dose. Such options can be explored using the IPO-IMPT optimization framework. [0116] Ridge filters have been used previously in particle therapy to avoid the need to switch energy layers, reducing treatment time. They have gained new popularity in the era of FLASH therapy. The work described herein, embodies several further advances, including a flexible scheme for simultaneous optimization of competing objectives of dose, DADR, and LETd, providing multiple solutions. The work also introduces sparsity, that is, ridge filters from which some pins are omitted, to further optimize dose rate and thus FLASH coverage. Preliminary experimental validation is also presented. [0117] The sparse ridge filters are more efficient than regular filters, providing more flexibility to improve the DADR. Use of the sparse filters can lead to some hotspots within the CTV, although this can be mitigated by alternating the beam orientation over SBRT fractions. Different user-defined thresholds for pin removal can lead to different filter designs. A threshold of 50%, provides reasonably good results for large tumors (such as Patient 1), whereas a threshold of 30% was a good starting point for smaller targets (such as Patient 2 and 3). The sparse ridge filter design process is currently based on a heuristic method, where several trial-and-error iterations are generally required to achieve an acceptable result. In some embodiments, a faster dose calculation engine may be used for patient-specific ridge filters, which would allow a combination of the ridge filter and plan optimization processes through a stepwise optimization scheme or using mixed- integer programming. This would allow simultaneous optimization of the proton spot map and the filter pin location map. [0118] It is important to note that the biological mechanism of FLASH sparing remains a subject of active investigation. IPO-IMPT optimization can assist this work by enabling biologists to separate the contribution of LET from dose rate effects. With the IPO-IMPT framework, different beam designs can be examined in parallel to determine the contribution of each term. When better biological models of the FLASH effect are available, the IPO-IMPT can be extended to incorporate them directly, rather than indirectly via DADR and LETd terms. Other examples include replacing the DADR with other dose rate approaches in IPO-IMPT. In some embodiments, a constant beam current is assumed, which allows a simplified optimization model for DADR, keeping spot MUs as the sole decision variables. In some embodiments, solutions for adding current as a decision variable are integrated into the IPO-IMPT framework. EXAMPLE 2: SIEMAC approach to solving the IPO-IMPT problem [0119] Extending the Traditional IMPT Optimization Problem to Solve IPO- IMPTPreviously, ziggurat-shaped pins have been used to create SOBPs. However, the present disclosure provides a simpler square pyramid-shaped pin to create the SOBPs to reduce computational effort in design. The objective function is expanded to include dose rate and LET objectives. Thus, the new problem becomes: where The dose rate and LET objectives, and , can be easily defined in a way directly analogous to Equation 16, and the arguments are again typically constrained by upper and lower bounds. More specifically, the objective function used in this analysis is
subject to upper and lower bounds on each optimization variable: where the generic variable v has been introduced for simplicity to represent the concatenation of w , l b , and l p ; and d, DR , and LET are the prescription dose, target dose rate, and target LET, respectively; and Θ is the Heaviside function. D 0 is a dose cutoff, where voxels with a dose below this value are not considered in the objective; typical values are 5% - 10% of the prescribed dose. ROB refers to the rest-of-body which is everything in the body besides the CTV and BSPTVs. Since dose rate and LET have contributions from each spot, a dose averaged dose rate and LET are used, i.e., and where DR i is the DADR in voxel i, I j is the nozzle current of spot j (i.e., 300 nA in some embodiments), LET i is the dose averaged LET in voxel i, and LET i j (the LET influence matrix) is the dose averaged LET in voxel i due to spot j. [0120] A restricted influence grid (RIG) is introduced to limit the extent of dose and LET influence matrices by inclusion of the spots that are within FLASH millisecond timing proximity of the location of the highest instantaneous dose. A RIG exists for each voxel i, and includes voxel i plus the neighboring voxels surrounding it. A time value for each RIG can then be defined as where ƒ ij is the fraction of spot j that impinges on RIG i and is the actual time duration of spot j. Alternatively, ƒ ij could also be defined as a Boolean value equal to 1 when the threshold of 0.5 is met, and 0 otherwise. In other words, is the hypothetical irradiation time on RIG i from spot j assuming the entire spot impinges on RIG i rather than just a fraction of it, and is a sum over these values without accounting for scan time + delivery time of other spots. Thus, when the spot and RIG mostly overlap, and when the spot and RIG overlap very little. [0121] For simplicity, a very rudimentary version of a RIG is used as well as a restricted dose (or LET) influence matrix: where is the dose to voxel i due to spot j considering the entirety of the CT grid. The restricted D ij , illustrated in FIGS.1B and 7, significantly trims down the CT grid for the sake of computational performance by assuming the dose is negligible in voxels far away from the spot. The dot-dash line with double-ended arrows in FIGS.1A and 1C show the spots interjoining with sparse pins/bars subject to RIG. [0122] These changes to the optimization problem present several challenges. First, the added arguments and objectives make the problem more complex and make solving the problem more CPU intensive. Furthermore, L ij must also be calculated in addition to D ij . Second, the added arguments and objectives can make the problem non- convex. Third, the variability of the geometry parameters means that D ij and L ij need to be re-calculated many times as the geometry changes, and it also makes the gradient calculation much more difficult since D ij and L ij are not constant. This leads to a further and very significant increase in necessary computing power. Implementation of the SIEMAC approach [0123] To address these challenges, a parallel computing framework is used. The first step is to define the initial (i.e., zeroth order) geometry. In some embodiments, this can be done using ray tracing to design a filter meant to produce a conformal dose distribution. In some embodiments, the initial geometry is defined using a forward heuristic, such as the sparse modulation technique. In other embodiments, the initial geometry is defined using a global search algorithm such as differential evolution, dual annealing, or other global search algorithm as can be appreciated. [0124] Next, a quasi-Newton method (the Limited-memory Broyden-Fletcher- Goldfarb-Shanno B (L-BFGS-B) algorithm) is used to better optimize the initial geometry, along with the spot weights. The gradient of the objective function is
The partial derivatives are straightforward to calculate analytically. The remaining partial derivatives are estimated using the finite difference approximation
[0125] Since f depends on D ij and L ij , and since D ij and L ij depend on the pin and bar lengths, it can be seen in Equations 9 and 10 that the number of geometries, and therefore the number of D ij ’s and L ij ’s that need to be calculated with MC, is N b + N p + 1 for each field and for each iteration of the optimization.
[0126] In order to complete these calculations in a reasonable amount of time, they were broken down into parallelizable pieces and submitted to a computing cluster or supercomputer. The exact parallelization scheme is illustrated in FIG. 1D. A red X in the bottom row of FIG. 1D represents a simulation that can be skipped due to the spot being far away from the modified geometry component (represented by the dot-dash line with double-ended arrows in FIG.s 1Aand C), which therefore saves time.
[0127] The overall workflow of the optimization can be seen in FIG. 2. The process begins by using a ray tracing algorithm with a patient’s computed tomography (CT) scan to define the initial geometry of the pins and bars. A Monte Carlo tool (such as TOPAS MC), analytical engine, artificial intelligence, or other approach is then used to calculate D tj and L tj . In parallel, many geometry variations are also simulated that are needed to calculate the gradient of the objective function. The simulation output data are then fed into an optimization algorithm and the process is repeated until an acceptable solution is reached.
SIEMAC for preclinical applications
[0128] Unlike clinical treatment plans, preclinical studies typically aim to deliberately irradiate an OAR and therefore require different optimization objectives. Preclinical objectives should include minimizing the spreads of the dose, dose rate, and LET distributions in the OAR target, thereby minimizing uncertainty when separating the contributions from each of these quantities on extra biological dose (XBD). The SEMAC algorithm was tested to see if it is feasible to indirectly optimize XBD via the physical quantities of dose, dose rate, and LET. In an example, the objective function, Equation 11 , was first set to deliver a uniform dose of 20 Gy to the target, which represents the threshold for short-term pneumonitis and long-term fibrosis. In other words, the last two lines of Equation 11 were not used initially. [0129] Then, a second round of optimization was done that included the last two lines of Equation 11 to attempt to narrow the dose, DADR, and LETd distributions, and therefore reduce the uncertainty in these quantities, while maintaining similar target dose coverage. The magnitude of spreads of dose, dose rate and LET distributions and their XBD i (DADR) and XBD i (LET) on a 36-mm spherical irradiation target of a minipig lung were compared before and after IPO-IMPT. XBD can be XBD i (DADR) and XBD i (LET) (where i is the voxel number), defined as and which represent adjustments to the physical dose that take into account biological responses to radiation. Advantages to healthy tissue are represented by larger values of XBD i (DADR) and smaller values of XBD i (LET). Here, a, k, DR t , and c are parameters that depend on biological mechanisms. Optimizing the pin and bar lengths can improve sparse compensation and sparse modulation, along with improved spot maps. The design of the minipig simulations is shown in FIG.3, which includes an anterior 250 MeV proton pencil beam and sets of variable length pins and bars that can be optimized to irradiate the spherical target. [0130] Traditional IMPT optimizes the weights (w) of a pencil beam spot map in order to produce a conformal dose distribution. A brief summary of traditional IMPT optimization is described below where important quantities such as the dose influence matrix (D ij ), objective function (f), prescribed dose (d), and penalty factor (p) are also defined. [0131] Traditional IMPT optimization consists of solving the problem where w are the spot weights, N s is the number of spots, and where f is the overall objective function, and ^ are the individual dose objectives with relative weights (or “penalties”) p n . The solution to the optimization problem is usually bounded by upper and lower limits on the spot weights (e.g., positivity or minimum MU constraints). [0132] Many different dose objectives can be defined. For example, one common one is the squared deviation objective where S is the set of voxels within a given structure (e.g., tumor, heart, lungs, etc.), N v is the number of voxels in S, d i is the dose to voxel i, and d is the prescribed dose. This objective penalizes the overall objective function every time a voxel’s dose deviates from the prescription, with larger deviations leading to larger penalties. [ 0133] The dose to a given voxel, d i , requires the dose influence matrix, D ij , which gives the dose per particle to voxel i due to spot j, to be known, i.e. where w j is the weight of, or number of particles in, spot j. D ij is typically calculated using a MC simulation or an analytical dose engine, with MC being preferable. While this calculation can be CPU intensive, it is not in general problematic given modern computing power, and it only needs to be performed once, since D ij is a constant in this context. This sort of optimization problem usually represents a convex problem and, once D ij is known, it can be solved fairly easily using standard optimization techniques. [0134] The arguments of the objective function are expanded to include geometry parameters. Specifically, b are the lengths of the range compensating bars (bars for short) with N b representing the number of bars, and are the lengths of the range modulating pins (pins for short) with N p representing the number of pins, as shown in FIG.1A. Usually, but this is not strictly necessary, so they are kept as two separate variables. A summary of these geometry components can be found in FIG.4. FIG.1A also shows the interjoining of spots and pins, i.e., the red spots impinge on pin peaks and the blue spots impinge on the valleys. Furthermore, since IMPT is typically delivered using multiple fields, superscripts are used on the variables to identify to which field that variable belongs (e.g., is the number of bars for field number 2) and use N f to represent the total number of fields. SIEMAC lung cancer patient plan [0135] To demonstrate SIEMAC, a three-field treatment plan was created for a representative lung cancer patient. The dose prescription to the centrally located CTV was 50 Gy, with nearby OARs including the heart and left lung. FIG. Error! Reference source not found. summarizes the result. Panel A shows the spot map, bar lengths (e.g., 1 mm to 100 mm), and pin lengths (e.g., 1 mm to 100 mm) for one of the three fields used (field A = gantry 40, field B = gantry 0, field C = gantry 320) in this study for iteration 0. Iteration 0 is defined to be the result after spot-weight-only optimization (i.e., the geometry parameters are held fixed) has been done using traditional IMPT techniques. Panel B is the same as panel A except after 9 iterations. Similarly, panels C and D show before and after distributions of dose, dose rate, and LET for an axial slice of the patient. Panel E shows the different components of the objective function vs the optimization iteration number. Finally, panel F shows dose, dose rate, and LET volume histograms for the CTV, left lung, and heart. [0136] The plots in FIG.4F show sizeable improvements to the dose rate and LET distributions in the lung and heart, with a negligible sacrifice to the dose distributions, when comparing traditional IMPT to IPO-IMPT with SIEMAC. For the OARs, we use an evaluation volume, which refers to the overlap between the OAR and BSPTV, excluding the CTV and any voxels with dose below 4 Gy. For the heart and lung, the percentage of the evaluation volume receiving above the FLASH threshold of 100 Gy/s rose from 93% to 100% and from 57% to 96%, respectively. Additionally, LET d coverage above 4 keV/μm dropped from 68% to 9% in the lung and from 26% to <1% in the heart. These improvements can be attributed to the shortening of up to 100 mm bar length and 60 mm pin length (FIG.4A-B), demonstrating improvements to sparse compensation and sparse modulation, along with improved spot maps that were optimized simultaneously with the pin and bar lengths. The sparse compensation in particular might explain the improvements in the LET distribution over our initial forward heuristic solution. Preclinical Optimization [0137] To demonstrate this functionality of SIEMAC, a single field plan has been generated for an animal irradiation with narrower dose, dose rate, LET, and XBD distributions in the irradiated OAR compared to an unoptimized plan, therefore reducing the uncertainty in these variables when deriving XBD models from preclinical studies. Values of c=0.04 μm/keV, DR t =40 Gy/s, k=0.5, and a=4/DR t , were used in equations 12 and 13. [0138] FIG. 5 summarizes the results of the SIEMAC-generated plan, which shows dose, dose rate, LET, XBD(DADR), and XBD(LET) distributions for the target in the left lung of the minipig before and after optimization. The improvements to the optimized plan are quantified by reporting both the full-width-half-maximum (FWHM) of each distribution, which is large for undesirable widely spread distributions and approaches zero for ideal distributions, as well as the area under the histograms after normalizing to a maximum value of 1, which, similarly, is large for undesirable distributions and smaller for ideal distributions. The results show that SIEMAC can be used to reduce the unoptimized wide spread in dose, DADR, and LET d (red vs blue lines in FIG.5 panels f-j) distributions in animal studies. [0139] For dose, SIEMAC decreased the FWHM by 30% (10 Gy to 7 Gy) and the area of the normalized histogram by 15% (4.8 to 4.1 a.u.). For DADR, the FWHM decreased by 1.2% (122 Gy/s to 120 Gy/s) and area decreased by 21% (4.8 to 3.8 a.u.). And for LET, FWHM decreased by 57% (7.1 keV/μm to 4.0 keV/μm) and area decreased by 44% (7.1 to 4.0 a.u.). To associate extra toxicity (i.e. biological effect) due to dose rate and LET distributions, XBD(DADR) and XBD(LET) are calculated using the proposed XBD model described in equations 12 and 13. The inverse solution of IPO-IMPT demonstrated a modest reduction of XBD(DADR) because the optimization algorithm considers the unoptimized DADR is well above the full UHDR benefit of 100 Gy/s (FIG. Error! Reference source not found.g), at 300 nA nozzle current. Such inverse solution of IBO-IMPT can improve much more XBD(DADR) for other organs and other beam conditions when needed as demonstrated for XBD(LET) (FIG. Error! Reference source not found.h). In summary, the results show a sizable XBD(LET) with a wide FWHM and area without optimization therefore XBD(LET) must be considered and optimized when studying UHDR sparing of lung toxicity. [0140] These results demonstrate a proof-of-concept that SIEMAC can be used to produce proton FLASH treatment plans that provide considerable improvements over existing planning algorithms (FIG.4 for clinical results and FIG.5 for preclinical results). The inverse SIEMAC solution improves upon the initial forward heuristic solution by iteratively optimizing range modulation, range compensation, and spot intensity map. The solution provides an opportunity to modulate sub-spot proton energy and proton intensity, which are vital for microscale radiation transport and thus FLASH optimization for simultaneous improvements in dose rate and LET of OARs. The technique has been shown to also be useful in animal studies for narrowing the dose, dose rate, and LET distributions in the target, therefore making derivation of XBD models more efficient and less uncertain. [0141] MC simulation of radiation transport and biochemical processes in microscale timing and spatial dimensions for each of the incoming protons (on the order of 10 9 ) is too time consuming, even for supercomputers, given the complexity of quantum physics equations embedded and implemented by MC. However, the full, LET dependent, quantum physics processes can be simplified in complex MC at microscale radiation transport to simulate the FLASH biological effects. Here, an inverse solution to IPO-IMPT that can be implemented on a modest cluster is provided. Such an inverse solution to IPO-IMPT can potentially improve cancer patient outcomes because microscale radiation transport underlies biochemical processes responsible for FLASH sparing of OARs. [0142] The optimization technique described herein is flexible enough that additional optimization parameters and objectives may be easily added. For example, the downstream distance from the nozzle of the patient can have a significant impact on dose rate. In some embodiments, this distance is fixed, but in other embodiments, this value is made variable and included in the optimization. Similarly, other quantities such as material density, beam current, dose threshold for the FLASH effect, dose rate threshold for the FLASH effect, etc., are fixed in some embodiments, and optimized in other embodiments using this technique. [0143] In the limited preclinical feasibility study, the focus was on the reduction of the unoptimized wide spread of dose, XBD(DADR) and XBD (LETd) distributions using IPO-IMPT for a minipig lung to help with quick convergence of XBD model derivations. The preliminary case was chosen to show the capability of IPO-IMPT to inversely solve the most relevant issues for preclinical application, because the lung is considered to have largest benefit from FLASH sparing with the biggest impact for often-fatal ultra-central lung cancer. Inverse solutions to IBO-IMPT for sparing of other organs, such as the esophagus, central airways, and heart, can be derived. FIG.5 shows that it is feasible for our preliminary SIEMAC to reduce the spreads of dose, dose rate, and LET distribution. Giving researchers control over the average values and spreads of dose, dose rate, and LET distributions can minimize the overlaps of dose, dose rate and LET among irradiations, therefore improving the efficiency with which XBD models can be derived and reducing the number of needed animal irradiations. In some embodiments, alternative methods of quantifying the distribution spreads, besides FWHM and integrated area, may be used. Minimization of overlaps of dose, dose rate and LET among irradiations are vital for biologists to separate their XBD(DADR) and XBD(LET d ) from physical dose contribution, which can be badly entangled among these three terms and multiple irradiations without careful optimizations, in observed OAR toxicities. [0144] In addition to XBD, alternative definitions of dose rate besides DADR may prove to be more useful, and could allow more elegant XBD(DADR) and XBD(LET d ) models. The definition of (equation 7) can also be varied. Associated with spot peak dose rate, RIG can potentially provide solutions more relevant to FLASH biology. Although SIEMAC’s current tapping of the underlying quantum physics radiation transport is rudimentary, further accessing the three microscopic dimensions and micro-timing is possible with better computing power and better implementations of inverse optimization. In addition, the objective function can be fine-tuned to account for the varying radiosensitivities of different OARS. For example, different OARS have suggested FLASH dose rate thresholds that vary by more than a factor of 2. EXAMPLE 3: Quality assurance to validate that actual dose rates and LETs correspond to planned or predicted values Introduction [0145] Over the past decade, there has been a burgeoning interest in FLASH radiotherapy, where FLASH refers to ultra high dose rates (UHDR), typically above 40 Gy/s. This interest is due to a number of studies that demonstrated that these high dose rates can significantly reduce damage to the healthy tissue of organs at risk (OARs) during treatment when compared to more conventional techniques, without sacrificing tumor killing efficacy. Proton pencil beam scanning (PBS) systems are an especially promising candidate for delivering FLASH treatments since many existing proton therapy centers can be made capable of delivering such beams with minimal overhead. Besides FLASH dose rates, which typically deliver a dose within a few milliseconds, protons offer biological effectiveness related to linear energy transfer (LET) according to their spatial and timing distributions, i.e. quantum physics processes besides the traditional classical physics processes that are described by dose distributions. [0146] One approach to treating patients with a FLASH proton (or other charged particle) beam is to use a patient-specific ridge filter to modulate the beam and therefore deliver a conformal dose distribution within a given beam specific target volume. Advancements in 3D printing technology make fabricating these patient-specific ridge filters accessible and affordable, and 3D printing has already been shown to be a useful tool in radiotherapy applications. However, FLASH treatment planning requires optimization of dose rate and LET in addition to dose, as well as the consideration of the distributions of these three quantities in the OARs, besides the target. Simultaneous optimization of dose and dose rate has been achieved, and simultaneous optimization of dose and LET has also been achieved. Recently, work has been done to show that dose, dose rate, and LET distributions can all simultaneously be optimized using a ridge filter. [0147] All of these new developments have generated an increased demand for innovation in detectors and techniques for measuring dose, dose rate, and LET for UHDR treatment plans. Detectors for measuring dose distributions, such as IBA Dosimetry's DigiPhant+MatriXX PT, have been around for years and continue to be widely used. Dose rate measurements for UHDR irradiations have been demonstrated using a scintillator- based detector in preclinical studies. Hybrid semiconductor pixel detectors based on the Timepix3 chips developed at CERN can be used to measure LET distributions and timing. These detectors have a silicon sensor with 14.08 mm x 14.08 mm sensitive regions corresponding to 256 x 256 pixels (pixel pitch of 55 μm) and nanosecond scale timing resolution. Novel strip ionization chambers have also recently been developed. [0148] The goal of this work is to develop techniques for performing QA measurements for UHDR treatment plans. For example, physical and biological optimization frameworks can be used to generate such plans. The MLSIC and Timepix3- based pixelated silicon detectors, which have never been validated under UHDR conditions without undesirable beam modifications, are selected for measuring dose, dose rate, and LET for a UHDR plan with a ridge filter. Solutions to challenges likely to be encountered with such measurements are also presented. Materials and Methods [0149] In this work, dose, dose rate, and LET are measured for a simple mock UHDR treatment plan generated using in-house software. The plan included of 250-MeV protons delivered by the Varian ProBeam PBS system according to a circular spot map with 5 mm spacing between each of the 149 spots with uniform intensities. Range shifters, along with a 3D printed ridge filter designed using an algorithm, were used to uniformly irradiate a 70 mm diameter spherical region inside of a water phantom with the target center at a depth of 60 mm, as shown in FIG.8A. Multiple measurements were performed using different novel detectors and detection techniques in order to get a complete set of dose, dose rate, and LET data in the desired target and target margin locations. The results were validated using a popular commercial detector, IBA Dosimetry's MatriXX PT, as well as by TOPAS Monte Carlo simulations. [0150] Although the treatment plan used was a UHDR plan, we chose to perform some of the dose and LET measurements at low beam currents that did not achieve UHDR. This does not diminish the validity of the measurements, since total dose and LET do not depend on the beam current or the dose rate used. This was done to solve issues with detector saturation and overheating. [0151] To properly design the experiment, perform realistic simulations, and carry out measurements, some preliminary calibration needed to be performed. First, material characterization for the 3D printed ridge filter was conducted using standard techniques. Subsequently, to ensure reliable LET measurements for single particle events, we developed an undersample-and-recover technique which is described below in the Principles of LET Measurements section. Experimental Setup [0152] FIG. 8A illustrates the experimental design used in this experiment. A proton beam of 250 MeV energy first impinges upon the ridge filter, followed by traversing an additional 30 mm + 80 mm of lucite in order to modulate the protons to achieve the desired depth. Subsequently, the protons deposit their remaining energy in the water phantom. Additionally, the figure contains photographs of the experimental setups for the MLSIC (FIG.8B) used to simultaneously measure dose and dose rate, and two Timepix3 detectors (FIG.1C). The Timepix3 detectors were employed together to simultaneously measure LET using the upstream Advapix detector, and timing via prompt gamma rays using the downstream Minipix detector. [0153] The nozzle of the machine is equipped with a laser grid running parallel to the downstream face of the range shifter for safety purposes, such that the beam will be shut off if any of the lasers are blocked. The mounting mechanism of the 80 mm lucite block had to be carefully designed to avoid blocking these lasers. This was done by mounting the block via four narrow bolts that could fit in-between adjacent lasers. The bolts allowed the block to be mounted such that there is a 15 mm air gap between the 30 mm range shifter and 80 mm block, thus avoiding blocking the lasers. Principles of LET Measurements [0154] LET is defined as the ratio of the energy a particle deposits along its trajectory (E) to its path length (L) and normalized by the density of the transport medium (ϱ), i.e. Since we use a silicon detector, LET is reported in units of (keV/μm)/(g/cm 3 ) in silicon. [0155] To ensure accurate measurements of LET for single particle events, there are two main parameters that should be accounted for: energy deposition and path length of individual particles through the sensor. Limitations can rise from the imprecision in the calculation of path length when the detector orientation angle is smaller than 30°, where 0° is defined as when the beam is perpendicular to the sensor surface. Moreover, as the Timepix3 chips are measuring the time over threshold (ToT) it is necessary that the acquisition time of each frame to be set high enough so the shaping time would allow registration of all charge produced by a particle. Previously, the lowest frame acquisition time reported was set at 500 μs. However, using the recommended conditions in our setup, a pileup of particles (i.e., saturation) was still noticed. Consequently, we had to employ shorter acquisition times and position the detector perpendicular to the beam, resulting in charge sharing over a smaller area of individual tracks (i.e., fewer pixels) for each detected proton. The shortened acquisition time resulted in an underestimation of the deposited energy due to incomplete charge collection by the detector electronics. Moreover, when the detector was oriented perpendicularly to the beam without making any additional adjustments to the formula used, it led to an underestimate of the LET. [0156] To address these issues, systematic studies of the detector response as a function of both acquisition time and detector angle were conducted. The aim was to derive LET correction factors that could be applied to the collected experimental data. For this purpose, LET distributions were measured with varying frame acquisition times (or detector angles), and then each fitted with a Gaussian to find the peak location. Subsequently, the peak positions were plotted as a function of acquisition time or detector angle. This comprehensive analysis enabled the establishment of correction factors that account for the aforementioned effects and improve the accuracy of our measurements. [0157] Measurements of the LET distributions require some filtering of the data to remove noise and background. A Savitzky-Golay filter was applied to the data to smooth out regions of low statistics. Statistics were limited by the detector overheating when using too high beam current or dose, and by available beam time. Measurement Details and Workflow [0158] The proton PBS system used was the Varian ProBeam, which can deliver energies up to 250 MeV at nozzle currents beyond 300 nA with the latest monitor unit (MU) chamber. Each of the 149 spots received 250 MU, where a MU is proportional to the number of protons ( N p ). The MU to N p conversion factor is energy dependent; at 250 MeV, there are 5.343×10 6 protons per MU. [0159] Three groups of measurements were performed, one for each of the detector configurations described in FIG.8, along with a validation run using the MatriXX PT: [0160] The first group of measurements was done using the MLSIC detector. Since the MLSIC is a 4D detector, time-dependent dose and IDR values in 3D space can be collected by running the beam one single time. Two datasets were collected, one with 7 nA nozzle current for a low dose rate measurement, and one with 50 nA nozzle current for a UHDR measurement. The integration duration of the MLSIC of 272 μs was used to calculate the average dose rate for each voxel within each time window, i.e. the dose rate for voxel i and time window k is where d ik is the dose to voxel i during time window k, and Δ t k is the duration of time window k and is always a constant value of 272 μs. Dose uncertainty for the MLSIC was estimated to be 2.5%. [0161] The MLSIC detector is comprised of x and y strips at different depths and principally reconstructs the 3D dose and dose rate distributions with certain assumptions about the dose profiles of the pencil beams. The presence of the ridge filter causes irregular dose profiles and makes dose reconstruction from MLSIC data challenging, therefore the results for relative dose for just one selected spot are demonstrated, specifically the first spot of the spot map, which lies on the lateral margin. [0162] The second group of measurements was done with two Timepix3 detectors; the primary (upstream Advapix) detector was used to measure LET while the secondary (downstream Minipix) detector was used to measure timing by detecting prompt gamma rays. Several datasets were collected for different depths and lateral offsets of the primary detector. Our minimum nozzle current of 0.8 nA was used to avoid detector saturation. In these measurements, we also lowered the plan MU from 250 MU per spot to 5 MU per spot to avoid overheating the detector. [0163] The third group of measurements was done with a 2D DigiPhant+MatriXX PT detector moving along the depth direction for validation. Dose in the xy-plane was measured at 17 different depths between 28 mm and 95 mm. Dose uncertainty for the MatriXX PT was estimated to be 1%. A nozzle current of 10 nA was used. Monte Carlo Simulations [0164] For further validation, TOPAS Monte Carlo simulations were performed of each of the previously described measurements. Following the guidance of AAPM Task Group 268, we have provided Table 1 which summarizes relevant details of the simulation. t P Table 1: Monte Carlo simulation details with items recommended by AAPM Task Group 268. [0165] The experimental measurements were compared to the simulations using a gamma analysis that accounts for differences in pixel size, statistical uncertainty, and uncertainty in detector positioning. A 3D gamma analysis was performed for each measurement using in-house software following the algorithm. For LET distributions, experimental data was compared to simulations by calculating the Bhattacharyya distance. Results [0166] The data for the first spot can be seen in FIG.9. A 3D gamma analysis was performed using in-house software. Using a standard 3%/3mm criteria on points with at least 5% of the maximum dose, all 32 measurements had a passing rate above 90% and 29 of the 32 measurements had a passing rate above 95%. [0167] The measurements done with the MLSIC detector used 7 nA and 50 nA beams each with 149 spots and 250 MU per spot. For the 7 nA beam, the irradiation time (IRT) for the first spot measured by MLSIC was 26.928 ms. This should be compared to the value from Varian log files, which recorded an IRT of 27.265 ms, a 1.2% difference. For the 50 nA beam, the IRT for the first spot measured by MLSIC was 11.832 ms and the log files recorded 11.691 ms for a difference of 1.2%. [0168] The time-dependent instantaneous dose rate curve for the first spot was also measured. The integration duration of the MLSIC is 272 μs, which corresponds to 99 samples within the time window of the first spot for the 7 nA beam and 44 samples for the 50 nA beam. The relative dose per sample, which remained fairly constant but for fluctuations on the order of 10% or less, was scaled to absolute dose using simulation data. Results for the first spot are plotted in FIG.10 at a depth of 50 mm along the central axis of the spot. The small variations in dose rate are most likely due to fluctuations in the anode, cathode, and RF of the cyclotron. [0169] For the LET correction factors, each LET distribution was fit with a Gaussian in order to find the location of the peak, as shown in FIGS.11A and 11B. One standard deviation of the fit parameter estimates is used as error bars. The peak positions were then plotted as a function of acquisition time (FIG. 11C) and detector angle (FIG. 11D) which can then be used to calculate the necessary LET correction factors. [0170] FIG.12 shows the corrected experimental and simulated LET distributions for five locations, represented by small circles in panel A: 30 mm depth on axis (panel B), 60 mm depth at lateral margin (panel C), 85 mm depth on axis (panel D), 90 mm depth on axis (panel E), and 95 mm depth on axis (panel F). The similarity between each experimental and simulated result was quantified by calculating the Bhattacharyya distance, which was 3.3e-3, 2.5e-3, 1.3e-2, 4.4e-4, 6.7e-4 for FIG. 12 panels B-F, respectively. The data show more and more high LET (greater than 4 keV/μm) components with increasing depth. [0171] FIG.13 shows the absolute dose measured with the MatriXX PT at 10 nA along with simulations for comparison. In FIG.13C, which shows data at depth 90 mm, which is within the distal falloff region, an additional simulation result at depth 91 mm is shown to demonstrate that the 10% disagreement between data and simulation represents a less than 1 mm difference. A gamma analysis using in-house software was performed between the dose measured with the MatriXX PT and TOPAS simulation that accounts for differences in pixel size (1 mm for simulation, 7.619 mm for experiment), statistical uncertainty, and uncertainty in detector positioning. The gamma analysis was limited to points greater than 5% the maximum dose of each measurement. Using a standard 3%/3mm criteria, 16 of the 17 measurements had a passing rate greater than 90% (representing depths from 28 mm to 94 mm). The measurement at a depth of 95 mm was the only measurement with a gamma passing rate less than 90%, though the maximum dose in this measurement was 0.751 Gy, representing a dose less than 10% the maximum dose of the plan. Of the 16 measurements which had at least a 90% passing rate, 15 passed with at least 95% of points meeting the 3%/3mm criteria (representing depths from 28 mm to 94 mm). Discussion [0172] There have been several optimization methods to include FLASH dose rate and/or LET. Here time-dependent instant dose rate curves and LET spectra are used to describe the quantum physics processes underlying such integrated optimizations. This example represents the first validation of quantum physics key parameters under unmodified primary FLASH beams. This is possible due to the novel under-sample and recover method using microsecond acquisition time to avoid saturation and recover the original LET values by calibration of under response of such short acquisitions. [0173] Overall, the measurements agreed very well with simulations. Small offsets of ∼1 mm caused by imperfect detector positioning can be seen in FIGS.9C, 13D, and 13E. FIG. 13C shows another <1 mm effect caused by imperfect modeling of the distal falloff region, which is known to be challenging. These small effects were accounted for in the gamma analysis. For the LET results, the data and simulations typically agreed within statistical error bars. In a small number of bins, the disagreement is slightly larger, and can be explained by incomplete filtering of noise and background. [0174] As FLASH radiotherapy continues to grow in popularity, advancements in detector technology will be important. The measurements with the MLSIC detector demonstrate a proof of concept for measuring the 4D dose distribution at UHDR. This work also represents a proof of concept for conducting preclinical FLASH trials, as the experiments were run with modifications to the nozzle in a non-invasive way without tripping any safety mechanisms. [0175] In FIG. 9, shown is a classical physics representation of a spot dose distribution. This concept of a spot is invalid under FLASH, which in reality is represented by quantum physics uncertainty of positioning, which can be seen over time by FIG.10 with uncertainty of proton energy represented by LET spectra in FIG.12. [0176] Although this work was done with a ridge filter and proton beam, the methods described have broad applicability to non-ridge and non-proton therapy modalities as well. For example, these methods could apply to other energetic particle beams. Furthermore, with the feasibility of simultaneously optimizing dose, dose rate, and LET, these techniques will be important for validating solutions to that optimization problem, as it is necessary to validate the underlying quantum physics processes related to FLASH dose rate and LET. [0177] The quantum processes of how incoming protons deposit dose over milliseconds (FLASH IDR) and how proton energies are distributed over microns (LET) both contribute to radiobiological effectiveness (RBE). In OAR, RBE sparing provided by FLASH can be negated by RBE enhancement as a result of high LET. Therefore, explicitly tapping into the quantum physics parameters (LET and dose rate) besides classical physics parameter (dose) can provide multiple and better solutions of the integrated optimization of dose, dose rate and LET. Supplemental information for Example 3 [0178] The ridge filter was 3D printed using a proprietary resin from Formlabs called Rigid 4000. Since the exact chemical formula of the resin is not public information, we had to develop a technique for realistically characterizing and modeling the material for simulations. To do this, an 80 mm long and 20 mm diameter cylinder was printed with the same resin. We then used a proton beam and commercial Zebra detector from IBA to measure the R80 of the protons with and without the cylinder in the beam. The water equivalent thickness (WET) of the cylinder is then From here, the relative stopping power (RSP) can be calculated by [0179] To realistically model the stopping power of this material in the simulations, we assumed a density To realistically model the scattering power, we assumed a chemical composition of , and then simulated many values of x between 0 and 1 and compared the dose profile to data to choose the best value for was chosen because it is known to be one of the major ingredients in resins besides PMMA. [0180] The characterization of the resin yielded an RSP value of 1.265 and an x- value of 0.95. [0181] Dose rate is calculated by simply dividing the total dose by the IRT. Since the dose measurements have already been completed as described above, dose rate measurements can be simplified to just measuring the IRT. [0182] In this work, we tested the feasibility of simultaneously measuring LET and IRT with a two-detector setup. Due to challenges with measuring LET described above, these measurements were done at low current (i.e. not UHDR). [0183] FIG.17 shows timing measured with Minipix Timepix3 at 250 MeV, 10 nA, 250 MU per spot, and 149 spots. Four repetitions were done to demonstrate reproducibility. The plots show clearly when the beam first turns on and when it stops, making extraction of the timing information from the data very straightforward, in this case between 2.598 and 3.081 seconds. Varian log files recorded IRTs ranging from 2.58553 to 3.06717 seconds, with an average of 2.86393 seconds, showing good agreement with the detector within 0.3%. [0184] Commissioning data from Raystation reports that at 250 MeV, there are 5.3×10 6 protons per MU. Therefore, the expected IRT for these measurements is [0185] (3.5) [0186] or 3.159 seconds. The measured time values are 2-18% lower than this calculated value, this is most likely due to the cyclotron to nozzle conversion efficiency on the day of the measurement being lower than the nominal value. [0187] The four timing results from FIG. 17 show fluctuations of 7-9% from the average. This is not unexpected due to fluctuations in the cyclotron to nozzle conversion efficiency. [0188] [0189] The terminology used herein is for purposes of describing particular embodiments only and is not intended to be limiting. The defined terms are in addition to the technical, scientific, or ordinary meanings of the defined terms as commonly understood and accepted in the relevant context. [0190] The terms “a,” “an” and “the” include both singular and plural referents, unless the context clearly dictates otherwise. Thus, for example, “a device” includes one device and plural devices. The terms “substantial” or “substantially” mean to within acceptable limits or degrees acceptable to those of skill in the art. For example, the term “substantially parallel to” means that a structure or device may not be made perfectly parallel to some other structure or device due to tolerances or imperfections in the process by which the structures or devices are made. The term “approximately” means to within an acceptable limit or amount to one of ordinary skill in the art. Relative terms, such as “over,” “above,” “below,” “top,” “bottom,” “upper” and “lower” may be used to describe the various elements’ relationships to one another, as illustrated in the accompanying drawings. These relative terms are intended to encompass different orientations of the device and/or elements in addition to the orientation depicted in the drawings. For example, if the device were inverted with respect to the view in the drawings, an element described as “above” another element, for example, would now be below that element. [0191] Relative terms may be used to describe the various elements’ relationships to one another, as illustrated in the accompanying drawings. These relative terms are intended to encompass different orientations of the device and/or elements in addition to the orientation depicted in the drawings. [0192] It should be noted that ratios, concentrations, amounts, and other numerical data may be expressed herein in a range format. It is to be understood that such a range format is used for convenience and brevity, and thus, should be interpreted in a flexible manner to include not only the numerical values explicitly recited as the limits of the range, but also to include all the individual numerical values or sub-ranges encompassed within that range as if each numerical value and sub-range is explicitly recited. To illustrate, a concentration range of “about 0.1% to about 5%” should be interpreted to include not only the explicitly recited concentration of about 0.1 wt% to about 5 wt%, but also include individual concentrations (e.g., 1%, 2%, 3%, and 4%) and the sub-ranges (e.g., 0.5%, 1.1%, 2.2%, 3.3%, and 4.4%) within the indicated range. In an embodiment, the term “about” can include traditional rounding according to significant figures of the numerical value. In addition, the phrase “about ‘x’ to ‘y’” includes “about ‘x’ to about ‘y’”. [0193] It should be emphasized that the above-described embodiments of the present disclosure are merely possible examples of implementations, and are set forth only for a clear understanding of the principles of the disclosure. Many variations and modifications may be made to the above-described embodiments of the disclosure without departing substantially from the spirit and principles of the disclosure. All such modifications and variations are intended to be included herein within the scope of this disclosure.