A preliminary study on a multiresolution‐level inverse planning approach for Gamma Knife radiosurgery

Deep reinforcement learning in radiation therapy planning optimization: A comprehensive review

PURPOSE

The formulation and optimization of radiation therapy plans are complex and time-consuming processes that heavily rely on the expertise of medical physicists. Consequently, there is an urgent need for automated optimization methods. Recent advancements in reinforcement learning, particularly deep reinforcement learning (DRL), show great promise for automating radiotherapy planning. This review summarizes the current state of DRL applications in this field, evaluates their effectiveness, and identifies challenges and future directions.

METHODS

A systematic search was conducted in Google Scholar, PubMed, IEEE Xplore, and Scopus using keywords such as "deep reinforcement learning", "radiation therapy", and "treatment planning". The extracted data were synthesized for an overview and critical analysis.

RESULTS

The application of deep reinforcement learning in radiation therapy plan optimization can generally be divided into three categories: optimizing treatment planning parameters, directly optimizing machine parameters, and adaptive radiotherapy. From the perspective of disease sites, DRL has been applied to cervical cancer, prostate cancer, vestibular schwannoma, and lung cancer. Regarding types of radiation therapy, it has been used in HDRBT, IMRT, SBRT, VMAT, GK, and Cyberknife.

CONCLUSIONS

Deep reinforcement learning technology has played a significant role in advancing the automated optimization of radiation therapy plans. However, there is still a considerable gap before it can be widely applied in clinical settings due to three main reasons: inefficiency, limited methods for quality assessment, and poor interpretability. To address these challenges, significant research opportunities exist in the future, such as constructing evaluators, parallelized training, and exploring continuous action spaces.

A study on inter-planner plan quality variability using a manual planning- or Lightning dose optimizer-approach for single brain lesions treated with the Gamma Knife® Icon™

PURPOSE

The purpose of this study is to investigate inter-planner plan quality variability using a manual forward planning (MFP)- or fast inverse planning (FIP, Lightning)-approach for single brain lesions treated with the Gamma Knife® (GK) Icon™.

METHODS

Thirty patients who were previously treated with GK stereotactic radiosurgery or radiotherapy were selected and divided into three groups (post-operative resection cavity, intact brain metastasis, and vestibular schwannoma [10 patients per group]). Clinical plans for the 30 patients were generated by multiple planners using FIP only (1), a combination of FIP and MFP (12), and MFP only (17). Three planners (Senior, Junior, and Novice) with varying experience levels re-planned the 30 patients using MFP and FIP (two plans per patient) with planning time limit of 60 min. Statistical analysis was performed to compare plan quality metrics (Paddick conformity index, gradient index, number of shots, prescription isodose line, target coverage, beam-on-time (BOT), and organs-at-risk doses) of MFP or FIP plans among three planners and to compare plan quality metrics between each planner's MFP/FIP plans and clinical plans. Variability in FIP parameter settings (BOT, low dose, and target max dose) and in planning time among the planners was also evaluated.

RESULTS

Variations in plan quality metrics of FIP plans among three planners were smaller than those of MFP plans for all three groups. Junior's MFP plans were the most comparable to the clinical plans, whereas Senior's and Novice's MFP plans were superior and inferior, respectively. All three planners' FIP plans were comparable or superior to the clinical plans. Differences in FIP parameter settings among the planners were observed. Planning time was shorter and variations in planning time among the planners were smaller for FIP plans in all three groups.

CONCLUSIONS

The FIP approach is less planner dependent and more time-honored than the MFP approach.

Global optimization for spot-based treatment planning

PURPOSE

Many radiotherapy modalities can deliver concentrated radiation in the form of spots, such as Gamma Knife (GK), GammaPod (GP), intensity-modulated proton therapy, and brachytherapy, and can be generalized as spot-based treatments. These treatments have a great therapeutic advantage of creating potent target dose while sparing the surrounding normal tissues. However, global optimization to determine the spot positions, shapes, and intensities is an intractable combinatorial problem for any real 3D problem. The conventional approach adopts heuristic spot selection and intensity optimization in a sequential manner to mitigate the problem complexity. In this work, we propose a novel framework that enables global optimization of spot-based treatment planning.

METHODS

The framework is based on kernel decomposition (KD) dose calculation, which models each spot dose as a scaled shift-invariant kernel, with the reference kernels and scales pre-calculated. During optimization, the framework incorporates Fast Fourier Transform (FFT) for objective and derivative evaluations and accommodates all spot candidates in optimization search with a temporal complexity of O(N3 log N) as opposed to O(N6 ) complexity in the conventional beamlet framework for volume dimensions of N × N × N. We demonstrated the FFT framework using simulations with different objectives. The framework's planning performance was illustrated using clinical GK and GP cases.

RESULTS

Pre-processing involves only a small number of reference kernels and a scale map for the KD model with marginal spatial and temporal overheads. For simulations with 512 × 512 image dimensions, plan optimization finished in ∼2 seconds with FFT, whereas it took 100× longer with the beamlet approach. For clinical cases, the FFT attained solutions within a minute with improved plan quality compared to clinical plans: better conformity and less integral dose because of using a global fine search space for optimal spots.

CONCLUSIONS

The scaled shift-invariance and FFT framework opens a new paradigm for spot-based treatment planning, as it can substantially reduce both the spatial and temporal complexities. The framework makes global optimization for spot-based treatment planning clinically feasible.

Clinical validation of novel lightning dose optimizer for gamma knife radiosurgery of irregular-shaped arteriovenous malformations and pituitary adenomas

PURPOSE

To demonstrate the clinical feasibility of a novel treatment planning algorithm via lightning dose optimizer (LDO) on Leksell Gamma Knife (LGK) GammaPlan with significantly faster planning times for stereotactic radiosurgery (SRS) of the complex and difficult arteriovenous malformations (AVMs) and pituitary adenomas.

METHODS AND MATERIALS

After completing the in-house end-to-end phantom testing and independent dose verification of the recently upgraded LDO algorithm on GammaPlan using the MD Anderson's IROC anthropomorphic SRS head phantom irradiation credentialing, 20 previously treated GK-SRS patients (10 AVM, average volume 3.61 cm³ and 10 pituitary adenomas, average volume 0.86 cm³ ) who underwent manual forward planning on GammaPlan were retrospectively replanned via LDO. These pathologies were included because of the need for adequate dose delivery with organs at risk in very close proximity. LDO finds the target curvature boundary by well-formulated linear programing objectives and inversely optimizes the GK-SRS plan by isocenter placement, optimization, and sequencing. For identical target coverage, the LDO and original manual plans were compared for target conformity, gradient index, dose to critical organs, and surrounding normal brain. Additionally, various treatment delivery parameters, including beam-on time were recorded.

RESULTS

For both patient cohorts, LDO provided similar target coverage with better dose conformity, tighter radiosurgical dose distribution with a lower value of gradient indices (all p < 0.001), and lower dose to critical organs. For AVMs, there was a significant reduction of normal brain V_10Gy , V_12Gy , and V_14Gy by 4.74, 3.67, and 2.67 cm³ (all p < 0.001). LDO had twice the number of shots (p < 0.001), and longer beam-on time (p = 0.012) by a factor of 1.44. For pituitary adenomas, LDO provided systematically lower values of V_10Gy , V_12Gy , and V_14Gy by 1.08, 0.86, and 0.68 cm³ (all p < 0.001), and lower maximum dose to optic pathway by 0.7 Gy (p = 0.005), but had almost twice the numbers of shots (p < 0.001) and increased beam-on time (p = 0.005) by a factor of 1.2. However, for both patient groups, the average planning time for the LDO was <5 min, compared to the estimated 30-90 min of manual planning times.

CONCLUSION

GK-SRS treatment on Leksell Perfexion GammaPlan using the LDO provided highly conformal target coverage with a steep dose gradient, spared critical organs, and significantly reduced normal brain dose for complex targets at the cost of slightly higher treatment times. LDO generated high-quality treatment plans and could significantly reduce planning time. If available, the LDO algorithm is suggested for validation and clinical use for complex and difficult GK cases.

Clinical Evaluation of the Inverse Planning System Utilized in Gamma Knife Lightning

Objectives

The purpose of this study is to independently compare the performance of the inverse planning algorithm utilized in Gamma Knife (GK) Lightning Treatment Planning System (TPS) to manual forward planning, between experienced and inexperienced users, for different types of targets.

Materials and Methods

Forty patients treated with GK stereotactic radiosurgery (SRS) for pituitary adenoma (PA), vestibular schwannoma (VS), post-operative brain metastases (pBM), and intact brain metastases (iBM) were randomly selected, ten for each site. Three inversely optimized plans were generated for each case by two experienced planners (OptExp1 and OptExp2) and a novice planner (OptNov) using GK Lightning TPS. For each treatment site, the Gradient Index (GI), the Paddick Conformity Index (PCI), the prescription percentage, the scaled beam-on time (sBOT), the number of shots used, and dosimetric metrics to OARs were compared first between the inversely optimized plans and the manually generated clinical plans, and then among the inversely optimized plans. Statistical analyses were performed using the Student’s t-test and the ANOVA followed by the post-hoc Tukey tests.

Results

The GI for the inversely optimized plans significantly outperformed the clinical plans for all sites. PCIs were similar between the inversely optimized and clinical plans for PA and VS, but were significantly improved in the inversely optimized plans for iBM and pBM. There were no significant differences in the sBOT between the inversely optimized and clinical plans, except for the PA cases. No significant differences were observed in dosimetric metrics, except for lower brain V_12Gy and PTV D_98% in the inversely optimized plans for iBM. There were no noticeable differences in plan qualities among the inversely optimized plans created by the novice and experienced planners.

Conclusion

Inverse planning in GK Lightning TPS produces GK SRS plans at least equivalent in plan quality and similar in sBOT compared to manual forward planning in this independent validation study. The automatic workflow of inversed planning ensures a consistent plan quality regardless of a planner’s experience.

Automatic Inverse Treatment Planning of Gamma Knife Radiosurgery via Deep Reinforcement Learning

PURPOSE

Several inverse planning algorithms have been developed for Gamma Knife (GK) radiosurgery to determine a large number of plan parameters via solving an optimization problem, which typically consists of multiple objectives. The priorities among these objectives need to be repetitively adjusted to achieve a clinically good plan for each patient. This study aimed to achieve automatic and intelligent priority-tuning, by developing a deep reinforcement learning (DRL) based method to model the tuning behaviors of human planners.

METHODS

We built a priority-tuning policy network using deep convolutional neural networks. Its input was a vector composed of multiple plan metrics that were used in our institution for GK plan evaluation. The network can determine which tuning action to take, based on the observed quality of the intermediate plan. We trained the network using an end-to-end DRL framework to approximate the optimal action-value function. A scoring function was designed to measure the plan quality to calculate the received reward of a tuning action.

RESULTS

Vestibular schwannoma was chosen as the test bed in this study. The number of training, validation and testing cases were 5, 5, and 16, respectively. For these three datasets, the average scores of the initial plans obtained with a same initial priority set were 3.63 ± 1.34, 3.83 ± 0.86 and 4.20 ± 0.78, respectively, while can be improved to 5.28 ± 0.23, 4.97 ± 0.44 and 5.22 ± 0.26 through manual priority tuning by human expert planners. Our network achieved competitive results with 5.42 ± 0.11, 5.10 ± 0. 42, 5.28 ± 0.20, respectively.

CONCLUSIONS

Our network can generate GK plans of comparable or slightly higher quality comparing with the plans generated by human planners via manual priority tuning for vestibular schwannoma cases. The network can potentially be incorporated into the clinical workflow as a planning assistance to improve GK planning efficiency and help to reduce plan quality variation caused by inter-planner variability. We also hope that our method can reduce the workload of GK planners and allow them to spend more time on more challenging cases. This article is protected by copyright. All rights reserved.

Dosimetry

The dosimeters used to measure radiation dose produce a value which has to be calibrated to be in keeping with the values in an approved laboratory, which will be one of an international network of such laboratories at the center of which is the Bureau International des Poids et Mesures in France (BIPM). Dosimeters work by producing a quantitatively proportional change in status to the intensity of the radiation being measure. Amongst the techniques in use are thermoluminescent devices, radiographic film, radiochromic film, semiconductors, ionization chambers, silicon diodes and gel dosimeters. The Gamma Knife radiation has been difficult to measure directly because the beams have been to fine for accurate measurement by commonly available dosimeters. For more modern dosimeters this is less of a problem. During the treatment of a patient, a variety of indices are recorded to assist in the standardization and accuracy of treatment. Having determined the dose in the beams, it is necessary to calculate how much energy is lost during the passage of radiation from the source to the target. There has been a steady evolution of these calculations to make them more accurate.

Systematic evaluation and plan quality assessment of the Leksell® gamma knife® lightning dose optimizer

To compare stereotactic radiosurgery (SRS) plan quality metrics of manual forward planning (MFP) and Elekta Fast Inverse Planning™ (FIP)-based inversely optimized plans for patients treated with Gamma Knife®. Clinically treated, MFP SRS plans for 100 consecutive patients (115 lesions; 67 metastatic and 48 benign) were replanned with the FIP dose optimizer based on a convex linear programming formulation. Comparative plans were generated to match or exceed the following metrics in order of importance: Target Coverage (TC), Paddick Conformity Index (PCI), beam-on time (BOT), and Gradient Index (GI). Plan quality metrics and delivery parameters between MFP and FIP were compared for all lesions and stratified into subgroups for further analysis. Additionally, performance of FIP for multiple punctate (<4 mm) metastatic lesions on a subset of cases was investigated. A Wilcoxon signed-rank test for non-normal distributions was used to assess the statistical differences between the MFP and FIP treatment plans. Overall, 76% (87/115) of FIP plans showed a statistically significant improvement in plan quality compared to MFP plans. As compared to MFP, FIP plans demonstrated an increase in the median PCI by 1.1% (p<0.01), a decrease in GI by 3.7% (p< 0.01), and an increase in median number of shots by 74% (p< 0.01). TC and BOT were not statistically significantly different between MFP and FIP plans (p>0.05). FIP plans showed a statistically significant increase in use of 16 mm (p< 0.01) and blocked shots (p< 0.01), with a corresponding decrease in 4 mm shots (p< 0.01). Use of multiple shots per coordinate was significantly higher in FIP plans (p<0.01). The FIP optimizer failed to generate a clinically acceptable plan in 4/115 (3.5%) lesions despite optimization parameter changes. The mean optimization time for FIP plans was 5.0 min (Range: 1.0 - 10.0 min). In the setting of multiple punctate lesions, PCI for FIP was significantly improved (p<0.01) by changing the default low-dose/BOT penalty optimization setting from a default of 50/50 to 75-85/40. FIP offers a significant reduction in manual effort for SRS treatment planning while achieving comparable plan quality to an expert planner-substantially improving overall planning efficiency. FIP plans employ a non-intuitive increased use of blocked sectors and shot-in-shot technique to achieve high quality plans. Several FIP plans failed to achieve clinically acceptable treatments and warrant further investigation.

Evaluating the impact of prescription isodose line on plan quality using Gamma Knife inverse planning

The impact of selection of prescription isodose line (IDL) on plan quality has not been well evaluated during inverse planning (IP). In this study, a total of 180 IP plans at five levels of IDL were generated for 30 brain metastases (BMs). For each BM, one round of IP was performed with typical IP settings, followed by a quick fine‐tuning to ensure the same target coverage and comparable conformality index. The impact of the IDL on the quality metrics (selectivity, gradient index [GI], and treatment time) was evaluated. The decrease of selectivity and increase of GI meant inferior target dose conformality and more dose spillage. Additionally, a metric directly correlated to the treatment time was proposed. For all cases, the mean GI decreased monotonically as IDL decreased from 70% to 30%, and the decreasing rate was significantly different based on tumor size. The mean selectivity and number of shots decreased monotonically as IDL decreased for all the tumors. From 70% to 30% IDL, the decreasing rate of the mean selectivity was 2.8% (p = 0.020), 7.7% (p = 0.005), and 15.4% (p = 0.020) and that of the number of shots was 75.4% (p = 0.001), 73.2% (p = 0.001), and 50.7% (p = 0.009), for the large, medium, and small tumors, respectively. For the medium and small tumor groups, the mean treatment time increased monotonically when IDLs decreased (increasing rate was 80.0% [p = 0.002] for medium tumors [p = 0.001] and 130.8% [p = 0.001] for small tumors from 70% to 30%). For the large tumors, the mean treatment time was the shortest at 50% IDL (59.0 min) and higher at 70% (65.9 min) and 30% (71.9 min). Overall, the GammaPlan chose smaller sectors for plans with lower IDLs except for the large size group.

Inverse planning in Gamma Knife radiosurgery: A comparative planning study

PURPOSE

To determine the advantages of inverse planning using a prerelease version of Leksell Gamma Knife® (LGK) Lightning (Elekta AB, Sweden) compared to manual forward planning.

METHODS

Thirty-eight patients with metastases (MET, n = 15), vestibular schwannomas (VS, n = 11) and meningiomas (MEN, n = 12), treated with LGK Icon™ at our institution, were analyzed retrospectively. For each case, an inverse (inv) and inverse full coverage (fc) treatment plan was generated using LGK Lightning and compared to the clinical plans. Several dosimetry and efficiency characteristics were compared for each indication. The mean, median difference and interquartile range were reported and the significance was assessed with a paired-sample Wilcoxon test (significance level < 0.05). Further, the inter operator variability was analyzed for multiple users.

RESULTS

Inv and fc treatment plans show improved target coverage (up to 3.6%) for all analyzed paradigms. For inv plans, the selectivity is enhanced (MET: 2.9%; VS: 1.8%; MEN: 1%) and the organ at risk doses are significantly reduced (VS: up to 4.5%; MEN: up to 17.5%). For inv and fc plans, the beam on time (BOT) is shortened (MET: up to 7.9%; benign tumors: 49.5%). The inter operator variability analysis shows similar treatment plan quality with small differences in plan efficiency (difference in BOT: 1-3.3 min).

CONCLUSIONS

LGK Lightning allows to generate improved LGK treatment plans regarding plan quality with reduced BOT compared to manual forward plans. The inter operator variability showed that multiple users with different experiences can generate similar treatment plan quality using LGK Lightning.

Multiplatform Radiosurgery for Intracranial Meningiomas and Dose to the Dural Tail

Introduction

Meningiomas are extra-axial central nervous system tumors. Complete resection is often curative with macroscopically complete removal of the tumor, excision of its dural attachment, and any abnormal bone. Radiosurgery is also an option for high-risk patients or in patients with surgically residual disease. Dural tail is a typical radiological sign on contrast-enhanced MRI; it can contain tumor cells or be a reaction due to vascular congestion and edema. Radiosurgical planning treatment varies regarding the identification and coverage of the dural tail. This study aimed to retrospectively analyze a series of 143 patients with WHO Grade I meningiomas treated with different radiosurgical platforms, and dosing parameters focused on planning and dose delivery to the dural tail.

Methods

From February 2011 to July 2020, 143 patients with histologically confirmed or radiologically assumed WHO Grade I meningiomas were treated using rotating gamma-ray Infini™ (Gamma [MASEP Medical Science Technology Development Co., Shenzhen, China]), TomoTherapy® (Tomo [Accuray Inc., Sunnyvale, CA]), and CyberKnife® (CK [Accuray Inc.]). All plans were retrospectively reviewed to establish the maximum distance (MaxDis) from the prescription dose to the end of the dural tail and the minimum dose at the dural tail (MinDoseT) at this point. We also established the midpoint distance (MPDis) from the prescription dose to MaxDis and the dose at this point (MPDose). Plans were further distinguished when the physician intended to cover the dural tail versus when not. Patients and tumor response were assessed by imaging and clinical and phone call evaluations.

Results

Of the 143 patients, 81 were treated using Gamma, 34 using Tomo, and 28 using CK. Eighty patients were eligible for follow-up, of whom 58 (72.5%) had an unmistakable dural tail sign. Median follow-up was 1,118 days (range 189-3,496), mean age was 54.5 (range 19-90), and 61 were women, and 19 were men. Overall tumor volume was 6.5 cc (range 0.2-59); mean tumor volumes by different platforms were 2.4, 9.45, and 8 cc; dose prescribed and mean tumor coverage were 14 Gy and 92%, 14.5 Gy and 95%, and 14 Gy and 95.75% with Gamma, Tomo, and CK, respectively. The dural tail was drawn and planned with an attempt to treat in 18 patients (31%); the mean MaxDis, MinDoseT, MPDis, and MPDose were 9.0 mm, 2 Gy, 4.5 mm, and 10.6 Gy, respectively. At last follow-up, tumor control was achieved in 96% of patients for the whole series, and there were no statistical variations regarding tumor volume, dose, conformality, or control when stereotactic radiosurgery was used to cover the dural tail versus when it was not (p=0.105). One patient experienced a Grade 4 Radiation Therapy Oncology Group toxicity as an adverse radiation effect that required surgery, and 11 (7.6%) experienced a Grade 1 toxicity.

Conclusions

This is our preliminary report regarding the efficacy of radiosurgery for meningiomas using diverse platforms at three years of follow-up; the results regarding tumor control are in accordance with the published literature as of this writing. A conscious pursuit of the dural tail with the prescription dose has not proven to provide better tumor control than not doing so - even small areas of the tumor uncovered by the prescription dose did not alter tumor control at current follow-up. The doses delivered to these uncovered areas are quite significant; further follow-up is necessary to validate these findings.

IntuitivePlan inverse planning performance evaluation for Gamma Knife radiosurgery of AVMs

Purpose

To compare planning indices achieved using manual and inverse planning approaches for Gamma Knife radiosurgery of arterio‐venous malformations (AVMs).

Methods and materials

For a series of consecutive AVM patients, treatment plans were manually created by expert planners using Leksell GammaPlan (LGP). Patients were re‐planned using a new commercially released inverse planning system, IntuitivePlan. Plan quality metrics were calculated for both groups of plans and compared.

Results

Overall, IntuitivePlan created treatment plans of similar quality to expert planners. For some plan quality metrics statistically significant higher scores were achieved for the inversely generated plans (Coverage 96.8% vs 96.3%, P = 0.027; PCI 0.855 vs 0.824, P = 0.042), but others did not show statistically significant differences (Selectivity 0.884 vs 0.856, P = 0.071; GI 2.85 vs 2.76, P = 0.096; Efficiency Index 47.0% vs 48.1%, P = 0.242; Normal Brain V₁₂(cc) 5.81 vs 5.79, P = 0.497). Automatic inverse planning demonstrated significantly shorter planning times over manual planning (3.79 vs 11.58 min, P < 10⁻⁶) and greater numbers of isocentres (40.4 vs 10.8, P < 10⁻⁶), with an associated cost of longer treatment times (57.97 vs 49.52 min, P = 0.009). When planning and treatment time were combined, there was no significant difference in the overall time between the two methods (61.76 vs 61.10, P = 0.433).

Conclusions

IntuitivePlan can offer savings on the labor of treatment planning. In many cases, it achieves higher quality indices than those achieved by an “expert planner”.

Tuning-Target-Guided Inverse Planning of Brain Tumors With Abutting Organs at Risk During Gamma Knife Stereotactic Radiosurgery

Purpose We proposed a planning strategy that utilized tuning targets to guide GammaKnife (GK) Inverse Planning (IP) to deliver higher dose to the tumor, while keeping acceptable dose to the abutting organ at risk (OAR). Methods Ten patients with a large portion of brain tumor abutting the OAR previously treated with GK stereotactic radiosurgery (SRS) were selected. For each patient, multiple tuning targets were created by cropping the target contour from three-dimensional (3D) expansions of the OAR. The number of the tuning targets depended on the complexity of the planning process. To demonstrate dose sparing effect, an IP plan was generated for each tuning target after one round of optimization without shot fine-tuning. In the dose enhancement study, a more aggressive target dose was prescribed to the tuning target with a larger margin and one to two shots were filled in the region with missing dose. The resulting plans were compared to the previously approved clinical plans. Results For all 10 patients, a dose sparing effect was observed, i.e. both target coverage and dose to the OARs decreased when the margins of 3D expansion increased. For one patient, a margin of 6 mm was needed to decrease the maximum dose to the optical chiasm and optical nerve by 44.3% and 28.4%, respectively. For the other nine patients, the mean dropping rate of V12Gyto brain stem were 28.2% and 59.5% for tuning targets of 1 and 2 mm margins, respectively. In the dose enhancement study, the tuning-target-guided plans were hotter than the approved treatment plans, while keeping similar dose to the OARs. The mean of the treatment and enhancement dose was 15.6 ± 2.2 Gy and 18.5 ± 3.2 Gy, respectively. The mean coverage of the target by prescription dose was slightly higher in the enhancement plans (96.9 ± 2.6% vs 96.3 ± 3.6%), whereas the mean coverage of the enhancement dose was 20.1% higher in the enhancement plans (89.6 ± 9.0% vs 74.6 ± 19.9%). Conclusions We demonstrated that an inverse planning strategy could facilitate target dose enhancement for challenging GK cases while keeping acceptable OAR dose.