Exploration of differentiability in a proton computed tomography simulation framework.

Objective.Gradient-based optimization using algorithmic derivatives can be a useful technique to improve engineering designs with respect to a computer-implemented objective function. Likewise, uncertainty quantification through computer simulations can be carried out by means of derivatives of the computer simulation. However, the effectiveness of these techniques depends on how 'well-linearizable' the software is. In this study, we assess how promising derivative information of a typical proton computed tomography (pCT) scan computer simulation is for the aforementioned applications.Approach.This study is mainly based on numerical experiments, in which we repeatedly evaluate three representative computational steps with perturbed input values. We support our observations with a review of the algorithmic steps and arithmetic operations performed by the software, using debugging techniques.Main results.The model-based iterative reconstruction (MBIR) subprocedure (at the end of the software pipeline) and the Monte Carlo (MC) simulation (at the beginning) were piecewise differentiable. However, the observed high density and magnitude of jumps was likely to preclude most meaningful uses of the derivatives. Jumps in the MBIR function arose from the discrete computation of the set of voxels intersected by a proton path, and could be reduced in magnitude by a 'fuzzy voxels' approach. The investigated jumps in the MC function arose from local changes in the control flow that affected the amount of consumed random numbers. The tracking algorithm solves an inherently non-differentiable problem.Significance.Besides the technical challenges of merely applying AD to existing software projects, the MC and MBIR codes must be adapted to compute smoother functions. For the MBIR code, we presented one possible approach for this while for the MC code, this will be subject to further research. For the tracking subprocedure, further research on surrogate models is necessary.

Improving knowledge-based treatment planning for lung cancer radiotherapy with automatic multi-criteria optimized training plans.

BACKGROUND: Knowledge-based planning (KBP) is a method for automated radiotherapy treatment planning where appropriate optimization objectives for new patients are predicted based on a library of training plans. KBP can save time and improve organ at-risk sparing and inter-patient consistency compared to manual planning, but its performance depends on the quality of the training plans. We used another system for automated planning, which generates multi-criteria optimized (MCO) plans based on a wish list, to create training plans for the KBP model, to allow seamless integration of knowledge from a new system into clinical routine. Model performance was compared for KBP models trained with manually created and automatic MCO treatment plans. MATERIAL AND METHODS: Two RapidPlan models with the same 30 locally advanced non-small cell lung cancer patients included were created, one containing manually created clinical plans (RP_CLIN) and one containing fully automatic multi-criteria optimized plans (RP_MCO). For 15 validation patients, model performance was compared in terms of dose-volume parameters and normal tissue complication probabilities, and an oncologist performed a blind comparison of the clinical (CLIN), RP_CLIN, and RP_MCO plans. RESULTS: The heart and esophagus doses were lower for RP_MCO compared to RP_CLIN, resulting in an average reduction in the risk of 2-year mortality by 0.9 percentage points and the risk of acute esophageal toxicity by 1.6 percentage points with RP_MCO. The oncologist preferred the RP_MCO plan for 8 patients and the CLIN plan for 7 patients, while the RP_CLIN plan was not preferred for any patients. CONCLUSION: RP_MCO improved OAR sparing compared to RP_CLIN and was selected for implementation in the clinic. Training a KBP model with clinical plans may lead to suboptimal output plans, and making an extra effort to optimize the library plans in the KBP model creation phase can improve the plan quality for many future patients.

Uncertainty-aware spot rejection rate as quality metric for proton therapy using a digital tracking calorimeter.

Objective.Proton therapy is highly sensitive to range uncertainties due to the nature of the dose deposition of charged particles. To ensure treatment quality, range verification methods can be used to verify that the individual spots in a pencil beam scanning treatment fraction match the treatment plan. This study introduces a novel metric for proton therapy quality control based on uncertainties in range verification of individual spots.Approach.We employ uncertainty-aware deep neural networks to predict the Bragg peak depth in an anthropomorphic phantom based on secondary charged particle detection in a silicon pixel telescope designed for proton computed tomography. The subsequently predicted Bragg peak positions, along with their uncertainties, are compared to the treatment plan, rejecting spots which are predicted to be outside the 95% confidence interval. The such-produced spot rejection rate presents a metric for the quality of the treatment fraction.Main results.The introduced spot rejection rate metric is shown to be well-defined for range predictors with well-calibrated uncertainties. Using this method, treatment errors in the form of lateral shifts can be detected down to 1 mm after around 1400 treated spots with spot intensities of 1 × 107protons. The range verification model used in this metric predicts the Bragg peak depth to a mean absolute error of 1.107 ± 0.015 mm.Significance.Uncertainty-aware machine learning has potential applications in proton therapy quality control. This work presents the foundation for future developments in this area.

Patient selection, inter-fraction plan robustness and reduction of toxicity risk with deep inspiration breath hold in intensity-modulated radiotherapy of locally advanced non-small cell lung cancer.

Background: State-of-the-art radiotherapy of locally advanced non-small cell lung cancer (LA-NSCLC) is performed with intensity-modulation during free breathing (FB). Previous studies have found encouraging geometric reproducibility and patient compliance of deep inspiration breath hold (DIBH) radiotherapy for LA-NSCLC patients. However, dosimetric comparisons of DIBH with FB are sparse, and DIBH is not routinely used for this patient group. The objective of this simulation study was therefore to compare DIBH and FB in a prospective cohort of LA-NSCLC patients treated with intensity-modulated radiotherapy (IMRT). Methods: For 38 LA-NSCLC patients, 4DCTs and DIBH CTs were acquired for treatment planning and during the first and third week of radiotherapy treatment. Using automated planning, one FB and one DIBH IMRT plan were generated for each patient. FB and DIBH was compared in terms of dosimetric parameters and NTCP. The treatment plans were recalculated on the repeat CTs to evaluate robustness. Correlations between ΔNTCPs and patient characteristics that could potentially predict the benefit of DIBH were explored. Results: DIBH reduced the median Dmean to the lungs and heart by 1.4 Gy and 1.1 Gy, respectively. This translated into reductions in NTCP for radiation pneumonitis grade ≥2 from 20.3% to 18.3%, and for 2-year mortality from 51.4% to 50.3%. The organ at risk sparing with DIBH remained significant in week 1 and week 3 of treatment, and the robustness of the target coverage was similar for FB and DIBH. While the risk of radiation pneumonitis was consistently reduced with DIBH regardless of patient characteristics, the ability to reduce the risk of 2-year mortality was evident among patients with upper and left lower lobe tumors but not right lower lobe tumors. Conclusion: Compared to FB, DIBH allowed for smaller target volumes and similar target coverage. DIBH reduced the lung and heart dose, as well as the risk of radiation pneumonitis and 2-year mortality, for 92% and 74% of LA-NSCLC patients, respectively. However, the advantages varied considerably between patients, and the ability to reduce the risk of 2-year mortality was dependent on tumor location. Evaluation of repeat CTs showed similar robustness of the dose distributions with each technique.

Impact of RBE variations on risk estimates of temporal lobe necrosis in patients treated with intensity-modulated proton therapy for head and neck cancer.

BACKGROUND: Temporal lobe necrosis (TLN) is a potential late effect after radiotherapy for skull base head and neck cancer (HNC). Several photon-derived dose constraints and normal tissue complication probability (NTCP) models have been proposed, however variation in relative biological effectiveness (RBE) may challenge the applicability of these dose constraints and models in proton therapy. The purpose of this study was therefore to investigate the influence of RBE variations on risk estimates of TLN after Intensity-Modulated Proton Therapy for HNC. MATERIAL AND METHODS: Seventy-five temporal lobes from 45 previously treated patients were included in the analysis. Sixteen temporal lobes had radiation associated Magnetic Resonance image changes (TLIC) suspected to be early signs of TLN. Fixed (RWDFix) and variable RBE-weighed doses (RWDVar) were calculated using RBE = 1.1 and two RBE models, respectively. RWDFix and RWDVar for temporal lobes were compared using Friedman's test. Based on RWDFix, six NTCP models were fitted and internally validated through bootstrapping. Estimated probabilities from RWDFix and RWDVar were compared using paired Wilcoxon test. Seven dose constraints were evaluated separately for RWDFix and RWDVar by calculating the observed proportion of TLIC in temporal lobes meeting the specific dose constraints. RESULTS: RWDVar were significantly higher than RWDFix (p < 0.01). NTCP model performance was good (AUC:0.79-0.84). The median difference in estimated probability between RWDFix and RWDVar ranged between 5.3% and 20.0% points (p < 0.01), with V60GyRBE and DMax at the smallest and largest differences, respectively. The proportion of TLIC was higher for RWDFix (4.0%-13.1%) versus RWDVar (1.3%-5.3%). For V65GyRBE ≤ 0.03 cc the proportion of TLIC was less than 5% for both RWDFix and RWDVar. CONCLUSION: NTCP estimates were significantly influenced by RBE variations. Dmax as model predictor resulted in the largest deviations in risk estimates between RWDFix and RWDVar. V65GyRBE ≤ 0.03 cc was the most consistent dose constraint for RWDFix and RWDVar.

Enhancing Radiotherapy for Locally Advanced Non-Small Cell Lung Cancer Patients with iCE, a Novel System for Automated Multi-Criterial Treatment Planning Including Beam Angle Optimization.

In this study, the novel iCE radiotherapy treatment planning system (TPS) for automated multi-criterial planning with integrated beam angle optimization (BAO) was developed, and applied to optimize organ at risk (OAR) sparing and systematically investigate the impact of beam angles on radiotherapy dose in locally advanced non-small cell lung cancer (LA-NSCLC). iCE consists of an in-house, sophisticated multi-criterial optimizer with integrated BAO, coupled to a broadly used commercial TPS. The in-house optimizer performs fluence map optimization to automatically generate an intensity-modulated radiotherapy (IMRT) plan with optimal beam angles for each patient. The obtained angles and dose-volume histograms are then used to automatically generate the final deliverable plan with the commercial TPS. For the majority of 26 LA-NSCLC patients, iCE achieved improved heart and esophagus sparing compared to the manually created clinical plans, with significant reductions in the median heart Dmean (8.1 vs. 9.0 Gy, p = 0.02) and esophagus Dmean (18.5 vs. 20.3 Gy, p = 0.02), and reductions of up to 6.7 Gy and 5.8 Gy for individual patients. iCE was superior to automated planning using manually selected beam angles. Differences in the OAR doses of iCE plans with 6 beams compared to 4 and 8 beams were statistically significant overall, but highly patient-specific. In conclusion, automated planning with integrated BAO can further enhance and individualize radiotherapy for LA-NSCLC.

Investigating particle track topology for range telescopes in particle radiography using convolutional neural networks.

BACKGROUND: Proton computed tomography (pCT) and radiography (pRad) are proposed modalities for improved treatment plan accuracy and in situ treatment validation in proton therapy. The pCT system of the Bergen pCT collaboration is able to handle very high particle intensities by means of track reconstruction. However, incorrectly reconstructed and secondary tracks degrade the image quality. We have investigated whether a convolutional neural network (CNN)-based filter is able to improve the image quality. MATERIAL AND METHODS: The CNN was trained by simulation and reconstruction of tens of millions of proton and helium tracks. The CNN filter was then compared to simple energy loss threshold methods using the Area Under the Receiver Operating Characteristics curve (AUROC), and by comparing the image quality and Water Equivalent Path Length (WEPL) error of proton and helium radiographs filtered with the same methods. RESULTS: The CNN method led to a considerable improvement of the AUROC, from 74.3% to 97.5% with protons and from 94.2% to 99.5% with helium. The CNN filtering reduced the WEPL error in the helium radiograph from 1.03 mm to 0.93 mm while no improvement was seen in the CNN filtered pRads. CONCLUSION: The CNN improved the filtering of proton and helium tracks. Only in the helium radiograph did this lead to improved image quality.

Mixed Effect Modeling of Dose and Linear Energy Transfer Correlations With Brain Image Changes After Intensity Modulated Proton Therapy for Skull Base Head and Neck Cancer.

PURPOSE: Intensity modulated proton therapy (IMPT) could yield high linear energy transfer (LET) in critical structures and increased biological effect. For head and neck cancers at the skull base this could potentially result in radiation-associated brain image change (RAIC). The purpose of the current study was to investigate voxel-wise dose and LET correlations with RAIC after IMPT. METHODS AND MATERIALS: For 15 patients with RAIC after IMPT, contrast enhancement observed on T1-weighted magnetic resonance imaging was contoured and coregistered to the planning computed tomography. Monte Carlo calculated dose and dose-averaged LET (LETd) distributions were extracted at voxel level and associations with RAIC were modelled using uni- and multivariate mixed effect logistic regression. Model performance was evaluated using the area under the receiver operating characteristic curve and precision-recall curve. RESULTS: An overall statistically significant RAIC association with dose and LETd was found in both the uni- and multivariate analysis. Patient heterogeneity was considerable, with standard deviation of the random effects of 1.81 (1.30-2.72) for dose and 2.68 (1.93-4.93) for LETd, respectively. Area under the receiver operating characteristic curve was 0.93 and 0.95 for the univariate dose-response model and multivariate model, respectively. Analysis of the LETd effect demonstrated increased risk of RAIC with increasing LETd for the majority of patients. Estimated probability of RAIC with LETd = 1 keV/µm was 4% (95% confidence interval, 0%, 0.44%) and 29% (95% confidence interval, 0.01%, 0.92%) for 60 and 70 Gy, respectively. The TD15 were estimated to be 63.6 and 50.1 Gy with LETd equal to 2 and 5 keV/µm, respectively. CONCLUSIONS: Our results suggest that the LETd effect could be of clinical significance for some patients; LETd assessment in clinical treatment plans should therefore be taken into consideration.

Clinical iterative model development improves knowledge-based plan quality for high-risk prostate cancer with four integrated dose levels.

BACKGROUND: Manual volumetric modulated arc therapy (VMAT) treatment planning for high-risk prostate cancer receiving whole pelvic radiotherapy (WPRT) with four integrated dose levels is complex and time consuming. We have investigated if the radiotherapy planning process and plan quality can be improved using a well-tuned model developed through a commercial system for knowledge-based planning (KBP). MATERIAL AND METHODS: Treatment plans from 69 patients treated for high-risk prostate cancer with manually planned VMAT were used to develop an initial KBP model (RapidPlan, RP). Prescribed doses were 50, 60, 67.5, and 72.5 Gy in 25 fractions to the pelvic lymph nodes, prostate and seminal vesicles, prostate gland, and prostate tumour(s), respectively. This RP model was in clinical use from July 2019 to February 2020, producing another set of 69 clinically delivered treatment plans for a new patient group, which were used to develop a second RP model. Both models were validated on an independent group of 40 patients. Plan quality was compared by D 98% and the Paddick conformity index for targets, mean dose (D mean) and generalised equivalent uniform dose (gEUD) for bladder, bowel bag and rectum, and number of monitor units (MU). RESULTS: Target coverage and conformity was similar between manually created and RP treatment plans. Compared to the manually created treatment plans, the final RP model reduced average D mean and gEUD with 2.7 Gy and 1.3 Gy for bladder, 1.2 Gy and 0.9 Gy for bowel bag, and 2.7 Gy and 0.8 Gy for rectum, respectively (p < .05). For rectum, the interpatient variation (i.e., 95% confidence interval) of DVHs was reduced by 23%. CONCLUSION: KBP improved plan quality and consistency among treatment plans for high-risk prostate cancer. Model tuning using KBP-based clinical plans further improved model outcome.

Helium radiography with a digital tracking calorimeter-a Monte Carlo study for secondary track rejection.

Radiation therapy using protons and heavier ions is a fast-growing therapeutic option for cancer patients. A clinical system for particle imaging in particle therapy would enable online patient position verification, estimation of the dose deposition through range monitoring and a reduction of uncertainties in the calculation of the relative stopping power of the patient. Several prototype imaging modalities offer radiography and computed tomography using protons and heavy ions. A Digital Tracking Calorimeter (DTC), currently under development, has been proposed as one such detector. In the DTC 43 longitudinal layers of laterally stacked ALPIDE CMOS monolithic active pixel sensor chips are able to reconstruct a large number of simultaneously recorded proton tracks. In this study, we explored the capability of the DTC for helium imaging which offers favorable spatial resolution over proton imaging. Helium ions exhibit a larger cross section for inelastic nuclear interactions, increasing the number of produced secondaries in the imaged object and in the detector itself. To that end, a filtering process able to remove a large fraction of the secondaries was identified, and the track reconstruction process was adapted for helium ions. By filtering on the energy loss along the tracks, on the incoming angle and on the particle ranges, 97.5% of the secondaries were removed. After passing through 16 cm water, 50.0% of the primary helium ions survived; after the proposed filtering 42.4% of the primaries remained; finally after subsequent image reconstruction 31% of the primaries remained. Helium track reconstruction leads to more track matching errors compared to protons due to the increased available focus strength of the helium beam. In a head phantom radiograph, the Water Equivalent Path Length error envelope was 1.0 mm for helium and 1.1 mm for protons. This accuracy is expected to be sufficient for helium imaging for pre-treatment verification purposes.

Optic nerve constraints for carbon ion RT at CNAO - Reporting and relating outcome to European and Japanese RBE.

BACKGROUND AND PURPOSE: Until now, carbon ion RT (CIRT) dose constraints for the optic nerve (ON) have only been validated and reported in the NIRS RBE-weighted dose (DNIRS). The aim of this work is to improve CNAO's RBE-weighted dose (DLEM) constraints by analyzing institutional toxicity data and by relating it to DNIRS. MATERIAL AND METHODS: A total of 65 ONs from 38 patients treated with CIRT to the head and neck region in the period 2013-14 were analyzed. The absorbed dose (DAbs) of the treatment plans was reproduced and subsequently both DLEM and DNIRS were applied, thus relating CNAO clinical toxicity to DNIRS. RESULTS: Median FU was 47 (26-67) months. Visual acuity was preserved for the 56 ONs in which the old constraints were respected. Three ONs developed visual decline at DLEM|1% ≥71 Gy(RBE)/DLEM|20% ≥68 Gy(RBE), corresponding to DNIRS|1% ≥68 Gy(RBE)/DNIRS|20% ≥62 Gy(RBE). Dose recalculation revealed that NIRS constraints of DNIRS|1% ≤40 Gy(RBE)/DNIRS|20% ≤28 Gy(RBE) corresponded to DLEM|1% ≤50 Gy(RBE)/DLEM|20% ≤40 Gy(RBE). Reoptimization of treatment plans with these new DLEM constraints showed that the dose distribution still complied with NIRS constraints when evaluated in DNIRS. However, due to uncertainties in the method, and to comply with the EQD2-based constraints used at GSI/HIT, a more moderate constraint relaxation to DLEM|1% ≤45 Gy(RBE)/DLEM|20% ≤37 Gy(RBE) has been implemented in CNAO clinical routine since October 2018. CONCLUSION: New DLEM constraints for the ON were derived by analyzing CNAO toxicity data and by linking our results to the experience of NIRS and GSI/HIT. This work demonstrates the value of recalculating and reporting results in both DLEM and DNIRS.

Design optimization of a pixel-based range telescope for proton computed tomography.

PURPOSE: A pixel-based range telescope for tracking particles during proton imaging is described. The detector applies a 3D matrix of stacked Monolithic Active Pixel Sensors with fast readout speeds. This study evaluates different design alternatives of the range telescope on basis of the protons' range accuracy and the track reconstruction efficiency. METHOD: Detector designs with different thicknesses of the energy-absorbing plates between each sensor layer are simulated using the GATE/Geant4 Monte Carlo software. Proton tracks traversing the detector layers are individually reconstructed, and a Bragg curve fitting procedure is applied for the calculation of each proton's range. RESULTS: Simulations show that the setups with 4 mm and thinner absorber layers of aluminum have a low range uncertainty compared to the physical range straggling, systematic errors below 0.3 mm water equivalent thickness and a track reconstruction capability exceeding ten million protons per second. CONCLUSIONS: In order to restrict the total number of layers and to yield the required tracking and range resolution properties, a design recommendation is reached where the proposed range telescope applies 3.5 mm thick aluminum absorber slabs between each sensor layer.