Feng, Hongying

Abstract Background Accurate and efficient dose calculation is essential for online adaptive planning in proton therapy. Deep learning (DL) has shown promising dose prediction results for pencil beam scanning proton therapy (PBSPT) in recent years, but existing DL‐based dose prediction methods still suffer from limited generalizability and an inability to effectively handle outlier clinical cases. This may lead to inaccurate dose delivery to targets or excessive irradiation to organs at risk (OARs), thereby compromising the safety and efficacy of online adaptive proton therapy. Purpose To design a physics‐aware and generalizable AI‐based PBSPT dose prediction method that incorporates underlying physics to enhance generalizability, particularly in handling outlier clinical cases. Methods This study analyzed PBSPT plans of 103 prostate (93 for training and 10 for testing) and 78 lung cancer patients (68 for training and 10 for testing) from our institution, with each case comprising CT images and structure sets. Using the doses generated by our Monte Carlo‐based dose engine as the reference standard, we compared three methods: the region of interest (ROI)‐based method, the beam mask and sliding window method, and the proposed noisy probing dose method, which rapidly generates a low‐statistics dose via uniformly weighted spots on an expanded spot‐placement target volume without optimization. To evaluate the generalizability of these methods to rare treatment planning scenarios, 12 cases with uncommon beam angles or prescription doses were used to assess their performance, which was evaluated using dose‐volume histogram (DVH) indices, 3D Gamma passing rates (3%/2 mm/10%), and Dice coefficients for dose agreement, while prediction times were measured to gauge model efficiency. Results The proposed noisy probing dose method consistently outperformed the ROI‐based and beam mask baselines across all evaluated metrics, with more accurate dose agreement and superior generalizability. For DVH indices, the noisy probing dose method achieved the smallest deviation in clinical target volume (CTV) dose coverage: in prostate cancer, CTV D98 deviation was reduced by 45% (from 0.53 ± 0.22 Gy [RBE] for ROI‐based) and 29% (from 0.41 ± 0.28 Gy [RBE] for beam mask) to 0.29 ± 0.06 Gy [RBE]; in lung cancer, similar improvements were observed, with CTV D98 deviations reduced to 0.34 ± 0.12 Gy [RBE]. The 3D Gamma passing rates improved to 99.65% ± 1.15% for prostate targets and 97.04% ± 1.17% for lung targets. The dice coefficients of the 90% iso‐dose lines were also the highest with the noisy probing dose method (prostate: 0.983 ± 0.005; lung: 0.967 ± 0.01). For the 12 outlier cases, the noisy probing dose method maintained superior generalizability, yielding higher 3D Gamma passing rates (prostate targets: 96.79% ± 0.83%, OARs: 94.29% ± 1.01%; lung targets: 93.38% ± 1.34%, OARs: 93.95% ± 1.32%), demonstrating robust generalizability to rare clinical scenarios. The dose predictions for all testing cases were completed within 0.3 seconds. Conclusions A novel noisy probing dose method was proposed for PBSPT dose prediction in prostate and lung cancer patients. By embedding more proton‐specific physics, this method demonstrated an improvement in the generalizability of dose prediction.

Abstract Background Efficient and accurate delineation of organs at risk (OARs) is a critical procedure for treatment planning and dose evaluation. Deep learning‐based auto‐segmentation of OARs has shown promising results and is increasingly being used in radiation therapy. However, existing deep learning‐based auto‐segmentation approaches face two challenges in clinical practice: generalizability and human‐AI interaction. A generalizable and promptable auto‐segmentation model, which segments OARs of multiple disease sites simultaneously and supports on‐the‐fly human‐AI interaction, can significantly enhance the efficiency of radiation therapy treatment planning. Purpose Meta's segment anything model (SAM) was proposed as a generalizable and promptable model for next‐generation natural image segmentation. We further evaluated the performance of SAM in radiotherapy segmentation. Methods Computed tomography (CT) images of clinical cases from four disease sites at our institute were collected: prostate, lung, gastrointestinal, and head & neck. For each case, we selected the OARs important in radiotherapy treatment planning. We then compared both the Dice coefficients and Jaccard indices derived from three distinct methods: manual delineation (ground truth), automatic segmentation using SAM's ’segment anything’ mode, and automatic segmentation using SAM's ‘box prompt’ mode that implements manual interaction via live prompts during segmentation. Results Our results indicate that SAM's segment anything mode can achieve clinically acceptable segmentation results in most OARs with Dice scores higher than 0.7. SAM's box prompt mode further improves Dice scores by 0.1∼0.5. Similar results were observed for Jaccard indices. The results show that SAM performs better for prostate and lung, but worse for gastrointestinal and head & neck. When considering the size of organs and the distinctiveness of their boundaries, SAM shows better performance for large organs with distinct boundaries, such as lung and liver, and worse for smaller organs with less distinct boundaries, like parotid and cochlea. Conclusions Our results demonstrate SAM's robust generalizability with consistent accuracy in automatic segmentation for radiotherapy. Furthermore, the advanced box‐prompt method enables the users to augment auto‐segmentation interactively and dynamically, leading to patient‐specific auto‐segmentation in radiation therapy. SAM's generalizability across different disease sites and different modalities makes it feasible to develop a generic auto‐segmentation model in radiotherapy.

Abstract Purpose . To enhance an in-house graphic-processing-unit accelerated virtual particle (VP)-based Monte Carlo (MC) proton dose engine (VPMC) to model aperture blocks in both dose calculation and optimization for pencil beam scanning proton therapy (PBSPT)-based stereotactic radiosurgery (SRS). Methods and materials . A module to simulate VPs passing through patient-specific aperture blocks was developed and integrated in VPMC based on simulation results of realistic particles (primary protons and their secondaries). To validate the aperture block module, VPMC was first validated by an opensource MC code, MCsquare, in eight water phantom simulations with 3 cm thick brass apertures: four were with aperture openings of 1, 2, 3, and 4 cm without a range shifter, while the other four were with same aperture opening configurations with a range shifter of 45 mm water equivalent thickness. Then, VPMC was benchmarked with MCsquare and RayStation MC for 10 patients with small targets (average volume 8.4 c.c. with range of 0.4–43.3 c.c.). Finally, 3 typical patients were selected for robust optimization with aperture blocks using VPMC. Results . In the water phantoms, 3D gamma passing rate (2%/2 mm/10%) between VPMC and MCsquare was 99.71 ± 0.23%. In the patient geometries, 3D gamma passing rates (3%/2 mm/10%) between VPMC/MCsquare and RayStation MC were 97.79 ± 2.21%/97.78 ± 1.97%, respectively. Meanwhile, the calculation time was drastically decreased from 112.45 ± 114.08 s (MCsquare) to 8.20 ± 6.42 s (VPMC) with the same statistical uncertainties of ~0.5%. The robustly optimized plans met all the dose–volume-constraints (DVCs) for the targets and OARs per our institutional protocols. The mean calculation time for 13 influence matrices in robust optimization by VPMC was 41.6 s and the subsequent on-the-fly ‘trial-and-error’ optimization procedure took only 71.4 s on average for the selected three patients. Conclusion . VPMC has been successfully enhanced to model aperture blocks in dose calculation and optimization for the PBSPT-based SRS.

AbstractBackgroundDeformable Image Registration (DIR) is an essential technique required in many applications of radiation oncology. However, conventional DIR approaches typically take several minutes to register one pair of 3D CT images and the resulting deformable vector fields (DVFs) are only specific to the pair of images used, making it less appealing for clinical application.PurposeA deep‐learning‐based DIR method using CT images is proposed for lung cancer patients to address the common drawbacks of the conventional DIR approaches and in turn can accelerate the speed of related applications, such as contour propagation, dose deformation, adaptive radiotherapy (ART), etc.MethodsA deep neural network based on VoxelMorph was developed to generate DVFs using CT images collected from 114 lung cancer patients. Two models were trained with the weighted mean absolute error (wMAE) loss and structural similarity index matrix (SSIM) loss (optional) (i.e., the MAE model and the M+S model). In total, 192 pairs of initial CT (iCT) and verification CT (vCT) were included as a training dataset and the other independent 10 pairs of CTs were included as a testing dataset. The vCTs usually were taken 2 weeks after the iCTs. The synthetic CTs (sCTs) were generated by warping the vCTs according to the DVFs generated by the pre‐trained model. The image quality of the synthetic CTs was evaluated by measuring the similarity between the iCTs and the sCTs generated by the proposed methods and the conventional DIR approaches, respectively. Per‐voxel absolute CT‐number‐difference volume histogram (CDVH) and MAE were used as the evaluation metrics. The time to generate the sCTs was also recorded and compared quantitatively. Contours were propagated using the derived DVFs and evaluated with SSIM. Forward dose calculations were done on the sCTs and the corresponding iCTs. Dose volume histograms (DVHs) were generated based on dose distributions on both iCTs and sCTs generated by two models, respectively. The clinically relevant DVH indices were derived for comparison. The resulted dose distributions were also compared using 3D Gamma analysis with thresholds of 3 mm/3%/10% and 2 mm/2%/10%, respectively.ResultsThe two models (wMAE and M+S) achieved a speed of 263.7±163 / 265.8±190 ms and a MAE of 13.15±3.8 / 17.52±5.8 HU for the testing dataset, respectively. The average SSIM scores of 0.987±0.006 and 0.988±0.004 were achieved by the two proposed models, respectively. For both models, CDVH of a typical patient showed that less than 5% of the voxels had a per‐voxel absolute CT‐number‐difference larger than 55 HU. The dose distribution calculated based on a typical sCT showed differences of ≤2cGy[RBE] for clinical target volume (CTV) D95 and D5, within ±0.06% for total lung V5, ≤1.5cGy[RBE] for heart and esophagus Dmean, and ≤6cGy[RBE] for cord Dmax compared to the dose distribution calculated based on the iCT. The good average 3D Gamma passing rates (> 96% for 3 mm/3%/10% and > 94% for 2 mm/2%/10%, respectively) were also observed.ConclusionA deep neural network‐based DIR approach was proposed and has been shown to be reasonably accurate and efficient to register the initial CTs and verification CTs in lung cancer.

AbstractBackgroundIn proton therapy dose calculation, Monte Carlo (MC) simulations are superior in accuracy but more time consuming, compared to analytical calculations. Graphic processing units (GPUs) are effective in accelerating MC simulations but may suffer thread divergence and racing condition in GPU threads that degrades the computing performance due to the generation of secondary particles during nuclear reactions.PurposeA novel concept of virtual particle (VP) MC (VPMC) is proposed to avoid simulating secondary particles in GPU‐accelerated proton MC dose calculation and take full advantage of the computing power of GPU.MethodsNeutrons and gamma rays were ignored as escaping from the human body; doses of electrons, heavy ions, and nuclear fragments were locally deposited; the tracks of deuterons were converted into tracks of protons. These particles, together with primary and secondary protons, are considered to be the realistic particles. Histories of primary and secondary protons were replaced by histories of multiple VPs. Each VP corresponded to one proton (either primary or secondary). A continuous‐slowing‐down‐approximation model, an ionization model, and a large angle scattering event model corresponding to nuclear interactions were developed for VPs by generating probability distribution functions (PDFs) based on simulation results of realistic particles using MCsquare. For efficient calculations, these PDFs were stored in the Compute Unified Device Architecture textures. VPMC was benchmarked with TOPAS and MCsquare in phantoms and with MCsquare in 13 representative patient geometries. Comparisons between the VPMC calculated dose and dose measured in water during patient‐specific quality assurance (PSQA) of the selected 13 patients were also carried out. Gamma analysis was used to compare the doses derived from different methods and calculation efficiencies were also compared.ResultsIntegrated depth dose and lateral dose profiles in both homogeneous and inhomogeneous phantoms all matched well among VPMC, TOPAS, and MCsquare calculations. The 3D–3D gamma passing rates with a criterion of 2%/2 mm and a threshold of 10% was 98.49% between MCsquare and TOPAS and 98.31% between VPMC and TOPAS in homogeneous phantoms, and 99.18% between MCsquare and TOPAS and 98.49% between VPMC and TOPAS in inhomogeneous phantoms, respectively. In patient geometries, the 3D–3D gamma passing rates with 2%/2 mm/10% between dose distributions from VPMC and MCsquare were 98.56 ± 1.09% in patient geometries. The 2D–3D gamma analysis with 3%/2 mm/10% between the VPMC calculated dose distributions and the 2D measured planar dose distributions during PSQA was 98.91 ± 0.88%. VPMC calculation was highly efficient and took 2.84 ± 2.44 s to finish for the selected 13 patients running on four NVIDIA Ampere GPUs in patient geometries.ConclusionVPMC was found to achieve high accuracy and efficiency in proton therapy dose calculation.

AbstractPurposeTo develop an online graphic processing unit (GPU)‐accelerated Monte Carlo‐based adaptive radiation therapy (ART) workflow for pencil beam scanning (PBS) proton therapy to address interfraction anatomical changes in patients treated with PBS.Methods and MaterialsA four‐step workflow was developed using our in‐house developed GPU‐accelerated Monte Carlo‐based treatment planning system to implement online Monte Carlo‐based ART for PBS. The first step conducts diffeomorphic demon‐based deformable image registration (DIR) to propagate contours on the initial planning CT (pCT) to the verification CT (vCT) to form a new structure set. The second step performs forward dose calculation of the initial plan on the vCT with the propagated contours after manual approval (possible modifications involved). The third step triggers a reoptimization of the plan depending on whether the verification dose meets the clinical requirements or not. A robust evaluation will be done for both the verification plan in the second step and the reopotimized plan in the third step. The fourth step involves a two‐stage (before and after delivery) patient‐specific quality assurance (PSQA) of the reoptimized plan. The before‐delivery PSQA is to compare the plan dose to the dose calculated using an independent fast open‐source Monte Carlo code, MCsquare. The after‐delivery PSQA is to compare the plan dose to the dose recalculated using the log file (spot MU, spot position, and spot energy) collected during the delivery. Jaccard index (JI), dice similarity coefficients (DSCs), and Hausdorff distance (HD) were used to assess the quality of the propagated contours in the first step. A commercial plan evaluation software, ClearCheck™, was integrated into the workflow to carry out efficient plan evaluation. 3D Gamma analysis was used during the fourth step to ensure the accuracy of the plan dose from reoptimization. Three patients with three different disease sites were chosen to evaluate the feasibility of the online ART workflow for PBS.ResultsFor all three patients, the propagated contours were found to have good volume conformance [JI (lowest‐highest: 0.833–0.983) and DSC (0.909–0.992)] but suboptimal boundary coincidence [HD (2.37–20.76 mm)] for organs‐at‐risk. The verification dose evaluated by ClearCheck™ showed significant degradation of the target coverage due to the interfractional anatomical changes. Reoptimization on the vCT resulted in great improvement of the plan quality to a clinically acceptable level. 3D Gamma analyses of PSQA confirmed the accuracy of the plan dose before delivery (mean Gamma index = 98.74% with a threshold of 2%/2 mm/10%), and after delivery based on the log files (mean Gamma index = 99.05% with a threshold of 2%/2 mm/10%). The average time cost for the complete execution of the workflow was around 858 s, excluding the time for manual intervention.ConclusionThe proposed online ART workflow for PBS was demonstrated to be efficient and effective by generating a reoptimized plan that significantly improved the plan quality.

AbstractPurposeDue to the employment of quadratic programming using soft constraints to implement dose volume constraints and the “trial‐and‐error” procedure needed to achieve a clinically acceptable plan, conventional dose volume constraints (upper limit) are not adequately effective in controlling small and isolated hot spots in the dose/linear energy transfer (LET) distribution. Such hot spots can lead to adverse events. In order to mitigate the risk of brain necrosis, one of the most clinically significant adverse events in patients receiving intensity‐modulated proton therapy (IMPT) for base of skull (BOS) cancer, we propose per‐voxel constraints to minimize hot spots in LET‐guided robust optimization.Methods and materialsTen BOS cancer patients treated with IMPT were carefully selected by meeting one of the following conditions: (1) diagnosis of brain necrosis during follow‐up; and (2) considered high risk for brain necrosis by not meeting dose constraints to the brain. An optimizing structure (BrainOPT) and an evaluating structure (BrainROI) that both contained the aforementioned hot dose regions in the brain were generated for optimization and evaluation, respectively. Two plans were generated for every patient: one using conventional dose‐only robust optimization, the other using LET‐guided robust optimization. The impact of LET was integrated into the optimization via a term of extra biological dose (xBD). A novel optimization tool of per‐voxel constraints to control small and isolated hot spots in either the dose, LET, or combined (dose/LET) distribution was developed and used to minimize dose/LET hot spots of the selected structures. Indices from dose‐volume histogram (DVH) and xBD dose‐volume histogram (xBDVH) were used in the plan evaluation. A newly developed tool of the dose‐LET‐volume histogram (DLVH) was also adopted to illustrate the underlying mechanism. Wilcoxon signed‐rank test was used for statistical comparison of the DVH and xBDVH indices between the conventional dose‐only and the LET‐guided robustly optimized plans.ResultsPer‐voxel constraints effectively and efficiently minimized dose hot spots in both dose‐only and LET‐guided robust optimization and LET hot spots in LET‐guided robust optimization. Compared to the conventional dose‐only robust optimization, the LET‐guided robust optimization could generate plans with statistically lower xBD hot spots in BrainROI (VxBD,50 Gy[RBE], p = 0.009; VxBD,60 Gy[RBE], p = 0.025; xBD1cc, p = 0.017; xBD2cc, p = 0.022) with comparable dose coverage, dose hot spots in the target, and dose hot spots in BrainROI. DLVH analysis indicated that LET‐guided robust optimization could either reduce LET at the same dose level or redistribute high LET from high dose regions to low dose regions.ConclusionPer‐voxel constraint is a powerful tool to minimize dose/LET hot spots in IMPT. The LET‐guided robustly optimized plans outperformed the conventional dose‐only robustly optimized plans in terms of xBD hot spots control.

Purpose To compare the dosimetric performances of small‐spot three‐dimensional (3D) and four‐dimensional (4D) robustly optimized intensity‐modulated proton (IMPT) plans in the presence of uncertainties and interplay effect simultaneously for distal esophageal carcinoma. Method and Materials Thirteen (13) patients were selected and re‐planned with small‐spot ( ~ 2–6 mm) 3D and 4D robust optimization in IMPT, respectively. The internal clinical target volumes (CTV high3d , CTV low3d ) were used in 3D robust optimization. Different CTVs (CTV high4d , CTV low4d ) were generated by subtracting an inner margin of the motion amplitudes in three cardinal directions from the internal CTVs and used in 4D robust optimization. All patients were prescribed the same dose to CTVs (50 Gy[RBE] for CTV high3d /CTV high4d and 45 Gy[RBE] for CTV low3d /CTV low4d ). Dose–volume histogram (DVH) indices were calculated to assess plan quality. Comprehensive plan robustness evaluations that consisted of 300 perturbed scenarios (10 different motion patterns to consider irregular motion (sampled from a Gaussian distribution) and 30 different uncertainties scenarios (sampled from a 4D uniform distribution) combined), were performed to quantify robustness to uncertainties and interplay effect simultaneously. Wilcoxon signed‐rank test was used for statistical analysis. Results Compared to 3D robustly optimized plans, 4D robustly optimized plans had statistically improved target coverage and better sparing of lungs and heart (heart D mean , P = 0.001; heart V 30Gy[RBE] , P = 0.001) in the nominal scenario. 4D robustly optimized plans had better robustness in target dose coverage (CTV high4d V 100% , P = 0.002) and the protection of lungs and heart (heart D mean , P = 0.001; heart V 30Gy[RBE] , P = 0.001) when uncertainties and interplay effect were considered simultaneously. Conclusions Even with small spots in IMPT, 4D robust optimization outperformed 3D robust optimization in terms of normal tissue protection and robustness to uncertainties and interplay effect simultaneously. Our findings support the use of 4D robust optimization to treat distal esophageal carcinoma with small spots in IMPT.

Abstract Purpose To compare the dosimetric performances of intensity‐modulated proton therapy (IMPT) plans generated with two different beam angle configurations (the Right–Left oblique posterior beams and the Superior–Inferior oblique posterior beams) for the treatment of distal esophageal carcinoma in the presence of uncertainties and interplay effect. Methods and Materials Twenty patients’ IMPT plans were retrospectively selected, with 10 patients treated with the R‐L oblique posterior beams (Group R‐L) and the other 10 patients treated with the S‐I oblique posterior beams (Group S‐I). Patients in both groups were matched by their clinical target volumes (CTVs—high and low dose levels) and respiratory motion amplitudes. Dose‐volume‐histogram (DVH) indices were used to assess plan quality. DVH bandwidth was calculated to evaluate plan robustness. Interplay effect was quantified using four‐dimensional (4D) dynamic dose calculation with random respiratory starting phase of each fraction. Normal tissue complication probability (NTCP) for heart, liver, and lung was calculated, respectively, to estimate the clinical outcomes. Wilcoxon signed‐rank test was used for statistical comparison between the two groups. Results Compared with plans in Group R‐L, plans in Group S‐I resulted in significantly lower liver D mean and lung V 30Gy[RBE] with slightly higher but clinically acceptable spinal cord D max . Similar plan robustness was observed between the two groups. When interplay effect was considered, plans in Group S‐I performed statistically better for heart D mean and V 30Gy[RBE] , lung Dmean and V 5Gy[RBE] , and liver D mean , with slightly increased but clinically acceptable spinal cord D max . NTCP for liver was significantly better in Group S‐I. Conclusions IMPT plans in Group S‐I have better sparing of liver, heart, and lungs at the slight cost of spinal cord maximum dose protection, and are more interplay‐effect resilient compared to IMPT plans in Group R‐L. Our study supports the routine use of the S‐I oblique posterior beams for the treatments of distal esophageal carcinoma.