Automated intensity modulated treatment planning: The expedited constrained hierarchical optimization (ECHO) system

A feasibility study of automating radiotherapy planning with large language model agents

Objective.Radiotherapy planning requires significant expertise to balance tumor control and organ-at-risk (OAR) sparing. Automated planning can improve both efficiency and quality. This study introduces GPT-Plan, a novel multi-agent system powered by the GPT-4 family of large language models (LLMs), for automating the iterative radiotherapy plan optimization.Approach.GPT-Plan uses LLM-driven agents, mimicking the collaborative clinical workflow of a dosimetrist and physicist, to iteratively generate and evaluate text-based radiotherapy plans based on predefined criteria. Supporting tools assist the agents by leveraging historical plans, mitigating LLM hallucinations, and balancing exploration and exploitation. Performance was evaluated on 12 lung (IMRT) and 5 cervical (VMAT) cancer cases, benchmarked against the ECHO auto-planning method and manual plans. The impact of historical plan retrieval on efficiency was also assessed.Results.For IMRT lung cancer cases, GPT-Plan generated high-quality plans, demonstrating superior target coverage and homogeneity compared to ECHO while maintaining comparable or better OAR sparing. For VMAT cervical cancer cases, plan quality was comparable to a senior physicist and consistently superior to a junior physicist, particularly for OAR sparing. Retrieving historical plans significantly reduced the number of required optimization iterations for lung cases (p < 0.01) and yielded iteration counts comparable to those of the senior physicist for cervical cases (p = 0.313). Occasional LLM hallucinations have been mitigated by self-reflection mechanisms. One limitation was the inaccuracy of vision-based LLMs in interpreting dose images.Significance.This pioneering study demonstrates the feasibility of automating radiotherapy planning using LLM-powered agents for complex treatment decision-making tasks. While challenges remain in addressing LLM limitations, ongoing advancements hold potential for further refining and expanding GPT-Plan's capabilities.

Descriptive overview of AI applications in x-ray imaging and radiotherapy

Artificial intelligence (AI) is transforming medical radiation applications by handling complex data, learning patterns, and making accurate predictions, leading to improved patient outcomes. This article examines the use of AI in optimising radiation doses for x-ray imaging, improving radiotherapy outcomes, and briefly addresses the benefits, challenges, and limitations of AI integration into clinical workflows. In diagnostic radiology, AI plays a pivotal role in optimising radiation exposure, reducing noise, enhancing image contrast, and lowering radiation doses, especially in high-dose procedures like computed tomography (CT). Deep learning (DL)-powered CT reconstruction methods have already been incorporated into clinical routine. Moreover, AI-powered methodologies have been developed to provide real-time, patient-specific radiation dose estimates. These AI-driven tools have the potential to streamline workflows and potentially become integral parts of imaging practices. In radiotherapy, AI's ability to automate and enhance the precision of treatment planning is emphasised. Traditional methods, such as manual contouring, are time-consuming and prone to variability. AI-driven techniques, particularly DL models, are automating the segmentation of organs and tumours, improving the accuracy of radiation delivery, and minimising damage to healthy tissues. Moreover, AI supports adaptive radiotherapy, allowing continuous optimisation of treatment plans based on changes in a patient's anatomy over time, ensuring the highest accuracy in radiation delivery and better therapeutic outcomes. Some of these methods have been validated and integrated into radiation treatment systems, while others are not yet ready for routine clinical use mainly due to challenges in validation, particularly ensuring reliability across diverse patient populations and clinical settings. Despite the potential of AI, there are challenges in fully integrating these technologies into clinical practice. Issues such as data protection, privacy, data quality, model validation, and the need for large and diverse datasets are crucial to ensuring the reliability of AI systems.

Clinical Experience With an Offline Adaptive Radiation Therapy Head and Neck Program: Dosimetric Benefits and Opportunities for Patient Selection

PURPOSE

The objective of this study was to develop a linear accelerator (LINAC)-based adaptive radiation therapy (ART) workflow for the head and neck that is informed by automated image tracking to identify major anatomic changes warranting adaptation. In this study, we report our initial clinical experience with the program and an investigation into potential trigger signals for ART.

METHODS AND MATERIALS

Offline ART was systematically performed on patients receiving radiation therapy for head and neck cancer on C-arm LINACs. Adaptations were performed at a single time point during treatment with resimulation approximately 3 weeks into treatment. Throughout treatment, all patients were tracked using an automated image tracking system called the Automated Watchdog for Adaptive Radiotherapy Environment (AWARE). AWARE measures volumetric changes in gross tumor volumes (GTVs) and selected normal tissues via cone beam computed tomography scans and deformable registration. The benefit of ART was determined by comparing adaptive plan dosimetry and normal tissue complication probabilities against the initial plans recalculated on resimulation computed tomography scans. Dosimetric differences were then correlated with AWARE-measured volume changes to identify patient-specific triggers for ART. Candidate trigger variables were evaluated using receiver operator characteristic analysis.

RESULTS

In total, 46 patients received ART in this study. Among these patients, we observed a significant decrease in dose to the submandibular glands (mean ± standard deviation: -219.2 ± 291.2 cGy, P < 10-5), parotids (-68.2 ± 197.7 cGy, P = .001), and oral cavity (-238.7 ± 206.7 cGy, P < 10-5) with the adaptive plan. Normal tissue complication probabilities for xerostomia computed from mean parotid doses also decreased significantly with the adaptive plans (P = .008). We also observed systematic intratreatment volume reductions (ΔV) for GTVs and normal tissues. Candidate triggers were identified that predicted significant improvement with ART, including parotid ΔV = 7%, neck ΔV = 2%, and nodal GTV ΔV = 29%.

CONCLUSIONS

Systematic offline head and neck ART was successfully deployed on conventional LINACs and reduced doses to critical salivary structures and the oral cavity. Automated cone beam computed tomography tracking provided information regarding anatomic changes that may aid patient-specific triggering for ART.

Multi-institutional evaluation of a Pareto navigation guided automated radiotherapy planning solution for prostate cancer

BACKGROUND

Current automated planning solutions are calibrated using trial and error or machine learning on historical datasets. Neither method allows for the intuitive exploration of differing trade-off options during calibration, which may aid in ensuring automated solutions align with clinical preference. Pareto navigation provides this functionality and offers a potential calibration alternative. The purpose of this study was to validate an automated radiotherapy planning solution with a novel multi-dimensional Pareto navigation calibration interface across two external institutions for prostate cancer.

METHODS

The implemented 'Pareto Guided Automated Planning' (PGAP) methodology was developed in RayStation using scripting and consisted of a Pareto navigation calibration interface built upon a 'Protocol Based Automatic Iterative Optimisation' planning framework. 30 previous patients were randomly selected by each institution (IA and IB), 10 for calibration and 20 for validation. Utilising the Pareto navigation interface automated protocols were calibrated to the institutions' clinical preferences. A single automated plan (VMATAuto) was generated for each validation patient with plan quality compared against the previously treated clinical plan (VMATClinical) both quantitatively, using a range of DVH metrics, and qualitatively through blind review at the external institution.

RESULTS

PGAP led to marked improvements across the majority of rectal dose metrics, with Dmean reduced by 3.7 Gy and 1.8 Gy for IA and IB respectively (p < 0.001). For bladder, results were mixed with low and intermediate dose metrics reduced for IB but increased for IA. Differences, whilst statistically significant (p < 0.05) were small and not considered clinically relevant. The reduction in rectum dose was not at the expense of PTV coverage (D98% was generally improved with VMATAuto), but was somewhat detrimental to PTV conformality. The prioritisation of rectum over conformality was however aligned with preferences expressed during calibration and was a key driver in both institutions demonstrating a clear preference towards VMATAuto, with 31/40 considered superior to VMATClinical upon blind review.

CONCLUSIONS

PGAP enabled intuitive adaptation of automated protocols to an institution's planning aims and yielded plans more congruent with the institution's clinical preference than the locally produced manual clinical plans.

SISS-MCO: large scale sparsity-induced spot selection for fast and fully-automated robust multi-criteria optimisation of proton plans

Objective.Intensity modulated proton therapy (IMPT) is an emerging treatment modality for cancer. However, treatment planning for IMPT is labour-intensive and time-consuming. We have developed a novel approach for multi-criteria optimisation (MCO) of robust IMPT plans (SISS-MCO) that is fully automated and fast, and we compare it for head and neck, cervix, and prostate tumours to a previously published method for automated robust MCO (IPBR-MCO, van de Water 2013).Approach.In both auto-planning approaches, the applied automated MCO of spot weights was performed with wish-list driven prioritised optimisation (Breedveld 2012). In SISS-MCO, spot weight MCO was applied once for every patient after sparsity-induced spot selection (SISS) for pre-selection of the most relevant spots from a large input set of candidate spots. IPBR-MCO had several iterations of spot re-sampling, each followed by MCO of the weights of the current spots.Main results.Compared to the published IPBR-MCO, the novel SISS-MCO resulted in similar or slightly superior plan quality. Optimisation times were reduced by a factor of 6 i.e. from 287 to 47 min. Numbers of spots and energy layers in the final plans were similar.Significance.The novel SISS-MCO automatically generated high-quality robust IMPT plans. Compared to a published algorithm for automated robust IMPT planning, optimisation times were reduced on average by a factor of 6. Moreover, SISS-MCO is a large scale approach; this enables optimisation of more complex wish-lists, and novel research opportunities in proton therapy.

Toward quantitative intrafractional monitoring in paraspinal SBRT using a proprietary software application: clinical implementation and patient results

Objective.We report on paraspinal motion and the clinical implementation of our proprietary software that leverages Varian's intrafraction motion review (IMR) capability for quantitative tracking of the spine during paraspinal SBRT. The work is based on our prior development and analysis on phantoms.Approach.To address complexities in patient anatomy, digitally reconstructed radiographs (DRR's) that highlight only the spine or hardware were constructed as tracking reference. Moreover, a high-pass filter and first-pass coarse search were implemented to enhance registration accuracy and stability. For evaluation, 84 paraspinal SBRT patients with sites spanning across the entire vertebral column were enrolled with prescriptions ranging from 24 to 40 Gy in one to five fractions. Treatments were planned and delivered with 9 IMRT beams roughly equally distributed posteriorly. IMR was triggered every 200 or 500 MU for each beam. During treatment, the software grabbed the IMR image, registered it with the corresponding DRR, and displayed the motion result in near real-time on auto-pilot mode. Four independent experts completed offline manual registrations as ground truth for tracking accuracy evaluation.Main results.Our software detected ≥1.5 mm and ≥2 mm motions among 17.1% and 6.6% of 1371 patient images, respectively, in either lateral or longitudinal direction. In the validation set of 637 patient images, 91.9% of the tracking errors compared to manual registration fell within ±0.5 mm in either direction. Given a motion threshold of 2 mm, the software accomplished a 98.7% specificity and a 93.9% sensitivity in deciding whether to interrupt treatment for patient re-setup.Significance.Significant intrafractional motion exists in certain paraspinal SBRT patients, supporting the need for quantitative motion monitoring during treatment. Our improved software achieves high motion tracking accuracy clinically and provides reliable guidance for treatment intervention. It offers a practical solution to ensure accurate delivery of paraspinal SBRT on a conventional Linac platform.

Automated planning of stereotactic spine re-irradiation using cumulative dose limits

Background and Purpose

The lack of dedicated tools in commercial planning systems currently restricts efficient review and planning for re-irradiation. The aim of this study was to develop an automated re-irradiation planning framework based on cumulative doses.

Materials and Methods

We performed a retrospective study of 14 patients who received spine SBRT re-irradiation near a previously irradiated treatment site. A fully-automated workflow, DART (Dose Accumulation-based Re-irradiation Tool), was implemented within Eclipse by leveraging a combination of a dose accumulation script and a proprietary automated optimization algorithm. First, we converted the prior treatment dose into equivalent dose in 2 Gy fractions (EQD2) and mapped it to the current anatomy, utilizing deformable image registration. Subsequently, the intersection of EQD2 isodose lines with relevant organs at risk defines a series of optimization structures. During plan optimization, the residual allowable dose at a specified tissue tolerance was treated as a hard constraint.

Results

All DART plans met institutional physical and cumulative constraints and passed plan checks by qualified medical physicists. DART demonstrated significant improvements in target coverage over clinical plans, with an average increase in PTV D99% and V100% of 2.3 Gy [range -0.3-7.7 Gy] and 3.4 % [range -0.4 %-7.6 %] (p < 0.01, paired t-test), respectively. Moreover, high-dose spillage (>105 %) outside the PTV was reduced by up to 7 cm3. The homogeneity index for DART plans was improved by 19 % (p < 0.001).

Conclusions

DART provides a powerful framework to achieve more tailored re-irradiation plans by accounting for dose distributions from the previous treatments. The superior plan quality could improve the therapeutic ratio for re-irradiation patients.

Enhancing the target visibility with synthetic target specific digitally reconstructed radiograph for intrafraction motion monitoring: A proof-of-concept study

BACKGROUND

Intrafraction motion monitoring in External Beam Radiation Therapy (EBRT) is usually accomplished by establishing a correlation between the tumor and the surrogates such as an external infrared reflector, implanted fiducial markers, or patient skin surface. These techniques either have unstable surrogate-tumor correlation or are invasive. Markerless real-time onboard imaging is a noninvasive alternative that directly images the target motion. However, the low target visibility due to overlapping tissues along the X-ray projection path makes tumor tracking challenging.

PURPOSE

To enhance the target visibility in projection images, a patient-specific model was trained to synthesize the Target Specific Digitally Reconstructed Radiograph (TS-DRR).

METHODS

Patient-specific models were built using a conditional Generative Adversarial Network (cGAN) to map the onboard projection images to TS-DRR. The standard Pix2Pix network was adopted as our cGAN model. We synthesized the TS-DRR based on the onboard projection images using phantom and patient studies for spine tumors and lung tumors. Using previously acquired CT images, we generated DRR and its corresponding TS-DRR to train the network. For data augmentation, random translations were applied to the CT volume when generating the training images. For the spine, separate models were trained for an anthropomorphic phantom and a patient treated with paraspinal stereotactic body radiation therapy (SBRT). For lung, separate models were trained for a phantom with a spherical tumor insert and a patient treated with free-breathing SBRT. The models were tested using Intrafraction Review Images (IMR) for the spine and CBCT projection images for the lung. The performance of the models was validated using phantom studies with known couch shifts for the spine and known tumor deformation for the lung.

RESULTS

Both the patient and phantom studies showed that the proposed method can effectively enhance the target visibility of the projection images by mapping them into synthetic TS-DRR (sTS-DRR). For the spine phantom with known shifts of 1 mm, 2 mm, 3 mm, and 4 mm, the absolute mean errors for tumor tracking were 0.11 ± 0.05 mm in the x direction and 0.25 ± 0.08 mm in the y direction. For the lung phantom with known tumor motion of 1.8 mm, 5.8 mm, and 9 mm superiorly, the absolute mean errors for the registration between the sTS-DRR and ground truth are 0.1 ± 0.3 mm in both the x and y directions. Compared to the projection images, the sTS-DRR has increased the image correlation with the ground truth by around 83% and increased the structural similarity index measure with the ground truth by around 75% for the lung phantom.

CONCLUSIONS

The sTS-DRR can greatly enhance the target visibility in the onboard projection images for both the spine and lung tumors. The proposed method could be used to improve the markerless tumor tracking accuracy for EBRT.

ROE (Radiotherapy Outcomes Estimator): An open-source tool for optimizing radiotherapy prescriptions

BACKGROUND AND OBJECTIVES

Radiotherapy prescriptions currently derive from population-wide guidelines established through large clinical trials. We provide an open-source software tool for patient-specific prescription determination using personalized dose-response curves.

METHODS

We developed ROE, a plugin to the Computational Environment for Radiotherapy Research to visualize predicted tumor control and normal tissue complication simultaneously, as a function of prescription dose. ROE can be used natively with MATLAB and is additionally made accessible in GNU Octave and Python, eliminating the need for commercial licenses. It provides a curated library of published and validated predictive models and incorporates clinical restrictions on normal tissue outcomes. ROE additionally provides batch-mode tools to evaluate and select among different fractionation schemes and analyze radiotherapy outcomes across patient cohorts.

CONCLUSION

ROE is an open-source, GPL-copyrighted tool for interactive exploration of the dose-response relationship to aid in radiotherapy planning. We demonstrate its potential clinical relevance in (1) improving patient awareness by quantifying the risks and benefits of a given treatment protocol (2) assessing the potential for dose escalation across patient cohorts and (3) estimating accrual rates of new protocols.

Simulating the Potential of Model-Based Individualized Prescriptions for Ultracentral Lung Tumors

Purpose

The use of stereotactic body radiation therapy for ultracentral lung tumors is limited by increased toxicity. We hypothesized that using published normal tissue complication probability (NTCP) and tumor control probability (TCP) models could improve the therapeutic ratio between tumor control and toxicity. A proposed model-based approach was applied to virtually replan early-stage non-small cell lung cancer (NSCLC) tumors.

Methods and Materials

The analysis included 63 patients with ultracentral NSCLC tumors treated at our center between 2008 and 2017. Along with current clinical constraints, additional NTCP model-based criteria, including for grade 3+ radiation pneumonitis (RP3+) and grade 2+ esophagitis, were implemented using 4 different fractionation schemes. Scaled dose distributions resulting in the highest TCP without violating constraints were selected (optimal plan [Planopt]). Planopt predictions were compared with the observed local control and toxicities.

Results

The observed 2-year local control rate was 72% (95% CI, 57%-88%) compared with 87% (range, 6%-93%) for Planopt TCP. Thirty-nine patients had Planopt with TCP > 80%, and 14 patients had Planopt TCP < 50%. The Planopt NTCPs for RP3+ were reduced by nearly half compared with patients' observed RP3+. The RP3+ NTCP was the most frequent reason for TCP of Planopt < 80% (14/24 patients), followed by grade 2+ esophagitis NTCP (5/24 patients) due to larger tumors (>40 cc vs ≤40 cc; P = .002) or a shorter tumor to esophagus distance (≥5 cm vs <5 cm; P < .001).

Conclusions

We demonstrated the potential for model-based prescriptions to yield higher TCP while respecting NTCP for patients with ultracentral NSCLC. Individualizing treatments based on NTCP- and TCP-driven simulations halved the predicted relative to the observed rates of RP3+. Our simulations also identified patients whose TCP could not be improved without violating NTCP due to larger tumors or a near tumor to esophagus proximity.

Clinical benefit of range uncertainty reduction in proton treatment planning based on dual-energy CT for neuro-oncological patients

OBJECTIVE

Several studies have shown that dual-energy CT (DECT) can lead to improved accuracy for proton range estimation. This study investigated the clinical benefit of reduced range uncertainty, enabled by DECT, in robust optimisation for neuro-oncological patients.

METHODS

DECT scans for 27 neuro-oncological patients were included. Commercial software was applied to create stopping-power ratio (SPR) maps based on the DECT scan. Two plans were robustly optimised on the SPR map, keeping the beam and plan settings identical to the clinical plan. One plan was robustly optimised and evaluated with a range uncertainty of 3% (as used clinically; denoted 3%-plan); the second plan applied a range uncertainty of 2% (2%-plan). Both plans were clinical acceptable and optimal. The dose-volume histogram parameters were compared between the two plans. Two experienced neuro-radiation oncologists determined the relevant dose difference for each organ-at-risk (OAR). Moreover, the OAR toxicity levels were assessed.

RESULTS

For 24 patients, a dose reduction >0.5/1 Gy (relevant dose difference depending on the OAR) was seen in one or more OARs for the 2%-plan; e.g. for brainstem D0.03cc in 10 patients, and hippocampus D40% in 6 patients. Furthermore, 12 patients had a reduction in toxicity level for one or two OARs, showing a clear benefit for the patient.

CONCLUSION

Robust optimisation with reduced range uncertainty allows for reduction of OAR toxicity, providing a rationale for clinical implementation. Based on these results, we have clinically introduced DECT-based proton treatment planning for neuro-oncological patients, accompanied with a reduced range uncertainty of 2%.

ADVANCES IN KNOWLEDGE

This study shows the clinical benefit of range uncertainty reduction from 3% to 2% in robustly optimised proton plans. A dose reduction to one or more OARs was seen for 89% of the patients, and 44% of the patients had an expected toxicity level decrease.

Learning image representations for content-based image retrieval of radiotherapy treatment plans

Objective.In this work, we propose a content-based image retrieval (CBIR) method for retrieving dose distributions of previously planned patients based on anatomical similarity. Retrieved dose distributions from this method can be incorporated into automated treatment planning workflows in order to streamline the iterative planning process. As CBIR has not yet been applied to treatment planning, our work seeks to understand which current machine learning models are most viable in this context.Approach.Our proposed CBIR method trains a representation model that produces latent space embeddings of a patient's anatomical information. The latent space embeddings of new patients are then compared against those of previous patients in a database for image retrieval of dose distributions. All source code for this project is available on github.Main results.The retrieval performance of various CBIR methods is evaluated on a dataset consisting of both publicly available image sets and clinical image sets from our institution. This study compares various encoding methods, ranging from simple autoencoders to more recent Siamese networks like SimSiam, and the best performance was observed for the multitask Siamese network.Significance.Our current results demonstrate that excellent image retrieval performance can be obtained through slight changes to previously developed Siamese networks. We hope to integrate CBIR into automated planning workflow in future works.

A novel automated planning approach for multi-anatomical sites cancer in Raystation treatment planning system

PURPOSE

To develop an automated planning approach in Raystation and evaluate its feasibility in multiple clinical application scenarios.

METHODS

An automated planning approach (Ruiplan) was developed by using the scripting platform of Raystation. Radiotherapy plans were re-generated both automatically by using Ruiplan and manually. 60 patients, including 20 patients with nasopharyngeal carcinoma (NPC), 20 patients with esophageal carcinoma (ESCA), and 20 patients with rectal cancer (RECA) were retrospectively enrolled in this study. Dosimetric and planning efficiency parameters of the automated plans (APs) and manual plans (MPs) were statistically compared.

RESULTS

For target coverage, APs yielded superior dose homogeneity in NPC and RECA, while maintaining similar dose conformity for all studied anatomical sites. For OARs sparing, APs led to significant improvement in most OARs sparing. The average planning time required for APs was reduced by more than 43% compared with MPs. Despite the increased monitor units (MUs) for NPC and RECA in APs, the beam-on time of APs and MPs had no statistical difference. Both the MUs and beam-on time of APs were significantly lower than that of MPs in ESCA.

CONCLUSIONS

This study developed a new automated planning approach, Ruiplan, it is feasible for multi-treatment techniques and multi-anatomical sites cancer treatment planning. The dose distributions of targets and OARs in the APs were similar or better than those in the MPs, and the planning time of APs showed a sharp reduction compared with the MPs. Thus, Ruiplan provides a promising approach for realizing automated treatment planning in the future.

Markerless motion tracking with simultaneous MV and kV imaging in spine SBRT treatment-a feasibility study

Objective. Motion tracking with simultaneous MV-kV imaging has distinct advantages over single kV systems. This research is a feasibility study of utilizing this technique for spine stereotactic body radiotherapy (SBRT) through phantom and patient studies.Approach. A clinical spine SBRT plan was developed using 6xFFF beams and nine sliding-window IMRT fields. The plan was delivered to a chest phantom on a linear accelerator. Simultaneous MV-kV image pairs were acquired during beam delivery. KV images were triggered at predefined intervals, and synthetic MV images showing enlarged MLC apertures were created by combining multiple raw MV frames with corrections for scattering and intensity variation. Digitally reconstructed radiograph (DRR) templates were generated using high-resolution CBCT reconstructions (isotropic voxel size (0.243 mm)3) as the reference for 2D-2D matching. 3D shifts were calculated from triangulation of kV-to-DRR and MV-to-DRR registrations. To evaluate tracking accuracy, detected shifts were compared to known phantom shifts as introduced before treatment. The patient study included a T-spine patient and an L-spine patient. Patient datasets were retrospectively analyzed to demonstrate the performance in clinical settings.Main results. The treatment plan was delivered to the phantom in five scenarios: no shift, 2 mm shift in one of the longitudinal, lateral and vertical directions, and 2 mm shift in all the three directions. The calculated 3D shifts agreed well with the actual couch shifts, and overall, the uncertainty of 3D detection is estimated to be 0.3 mm. The patient study revealed that with clinical patient image quality, the calculated 3D motion agreed with the post-treatment cone beam CT. It is feasible to automate both kV-to-DRR and MV-to-DRR registrations using a mutual information-based method, and the difference from manual registration is generally less than 0.3 mm.Significance. The MV-kV imaging-based markerless motion tracking technique was validated through a feasibility study. It is a step forward toward effective motion tracking and accurate delivery for spinal SBRT.

Direct incorporation of patient-specific efficacy and toxicity estimates in radiation therapy plan optimization

PURPOSE

Current radiation therapy (RT) treatment planning relies mainly on pre-defined dose-based objectives and constraints to develop plans that aim to control disease while limiting damage to normal tissues during treatment. These objectives and constraints are generally population-based, in that they are developed from the aggregate response of a broad patient population to radiation. However, correlations of new biologic markers and patient-specific factors to treatment efficacy and toxicity provide the opportunity to further stratify patient populations and develop a more individualized approach to RT planning. We introduce a novel intensity-modulated radiation therapy (IMRT) optimization strategy that directly incorporates patient-specific dose response models into the planning process. In this strategy, we integrate the concept of utility-based planning where the optimization objective is to maximize the predicted value of overall treatment utility, defined by the probability of efficacy (e.g., local control) minus the weighted sum of toxicity probabilities. To demonstrate the feasibility of the approach, we apply the strategy to treatment planning for non-small cell lung cancer (NSCLC) patients.

METHODS AND MATERIALS

We developed a prioritized approach to patient-specific IMRT planning. Using a commercial treatment planning system (TPS), we calculate dose based on an influence matrix of beamlet-dose contributions to regions-of-interest. Then, outside of the TPS, we hierarchically solve two optimization problems to generate optimal beamlet weights that can then be imported back to the TPS. The first optimization problem maximizes a patient's overall plan utility subject to typical clinical dose constraints. In this process, we facilitate direct optimization of efficacy and toxicity trade-off based on individualized dose-response models. After optimal utility is determined, we solve a secondary optimization problem that minimizes a conventional dose-based objective subject to the same clinical dose constraints as the first stage but with the addition of a constraint to maintain the optimal utility from the first optimization solution. We tested this method by retrospectively generating plans for five previously treated NSCLC patients and comparing the prioritized utility plans to conventional plans optimized with only dose metric objectives. To define a plan utility function for each patient, we utilized previously published correlations of dose to local control and grade 3-5 toxicities that include patient age, stage, microRNA levels, and cytokine levels, among other clinical factors.

RESULTS

The proposed optimization approach successfully generated RT plans for five NSCLC patients that improve overall plan utility based on personalized efficacy and toxicity models while accounting for clinical dose constraints. Prioritized utility plans demonstrated the largest average improvement in local control (16.6%) when compared to plans generated with conventional planning objectives. However, for some patients, the utility-based plans resulted in similar local control estimates with decreased estimated toxicity.

CONCLUSION

The proposed optimization approach, where the maximization of a patient's RT plan utility is prioritized over the minimization of standardized dose metrics, has the potential to improve treatment outcomes by directly accounting for variability within a patient population. The implementation of the utility-based objective function offers an intuitive, humanized approach to biological optimization in which planning trade-offs are explicitly optimized.

Clinical implementation and validation of an automated adaptive workflow for proton therapy

Background and purpose

Treatment quality of proton therapy can be monitored by repeat-computed tomography scans (reCTs). However, manual re-delineation of target contours can be time-consuming. To improve the workflow, we implemented an automated reCT evaluation, and assessed if automatic target contour propagation would lead to the same clinical decision for plan adaptation as the manual workflow.

Materials and methods

This study included 79 consecutive patients with a total of 250 reCTs which had been manually evaluated. To assess the feasibility of automated reCT evaluation, we propagated the clinical target volumes (CTVs) deformably from the planning-CT to the reCTs in a commercial treatment planning system. The dose-volume-histogram parameters were extracted for manually re-delineated (CTV_manual) and deformably mapped target contours (CTV_auto). It was compared if CTV_manual and CTV_auto both satisfied/failed the clinical constraints. Duration of the reCT workflows was also recorded.

Results

In 92% (N = 229) of the reCTs correct flagging was obtained. Only 4% (N = 9) of the reCTs presented with false negatives (i.e., at least one clinical constraint failed for CTV_manual, but all constraints were satisfied for CTV_auto), while 5% (N = 12) of the reCTs led to a false positive. Only for one false negative reCT a plan adaption was made in clinical practice, i.e., only one adaptation would have been missed, suggesting that automated reCT evaluation was possible. Clinical introduction hereof led to a time reduction of 49 h (from 65 to 16 h).

Conclusion

Deformable target contour propagation was clinically acceptable. A script-based automatic reCT evaluation workflow has been introduced in routine clinical practice.

Domain knowledge driven 3D dose prediction using moment-based loss function

Objective. To propose a novel moment-based loss function for predicting 3D dose distribution for the challenging conventional lung intensity modulated radiation therapy plans. The moment-based loss function is convex and differentiable and can easily incorporate clinical dose volume histogram (DVH) domain knowledge in any deep learning (DL) framework without computational overhead. Approach. We used a large dataset of 360 (240 for training, 50 for validation and 70 for testing) conventional lung patients with 2 Gy × 30 fractions to train the DL model using clinically treated plans at our institution. We trained a UNet like convolutional neural network architecture using computed tomography, planning target volume and organ-at-risk contours as input to infer corresponding voxel-wise 3D dose distribution. We evaluated three different loss functions: (1) the popular mean absolute error (MAE) loss, (2) the recently developed MAE + DVH loss, and (3) the proposed MAE + moments loss. The quality of the predictions was compared using different DVH metrics as well as dose-score and DVH-score, recently introduced by the AAPM knowledge-based planning grand challenge. Main results. Model with (MAE + moment) loss function outperformed the model with MAE loss by significantly improving the DVH-score (11%, p < 0.01) while having similar computational cost. It also outperformed the model trained with (MAE + DVH) by significantly improving the computational cost (48%) and the DVH-score (8%, p < 0.01). Significance. DVH metrics are widely accepted evaluation criteria in the clinic. However, incorporating them into the 3D dose prediction model is challenging due to their non-convexity and non-differentiability. Moments provide a mathematically rigorous and computationally efficient way to incorporate DVH information in any DL architecture. The code, pretrained models, docker container, and Google Colab project along with a sample dataset are available on our DoseRTX GitHub (https://github.com/nadeemlab/DoseRTX)

Evaluation of a proprietary software application for motion monitoring during stereotactic paraspinal treatment

PURPOSE

Stereotactic paraspinal treatment has become increasingly popular due to its favorable clinical outcome. An often-overlooked factor that compromises the effectiveness of such treatment is the patients' involuntary intrafractional motion. This work introduces and validates a proprietary software application that quantifies such motion for accurate patient monitoring during treatment.

METHODS

The software uses a separate full-trajectory cone-beam computed tomography (CBCT) after daily patient setup to establish reference projections. Once treatment starts, the software grabs the intrafraction motion review (IMR) image acquired by TrueBeam via the Varian iTools Capture software and compares it against the corresponding reference projection to instantly determine the 2D shifts of the vertebrae being monitored using the classical downhill simplex optimization method. To evaluate its performance, an anthropomorphic phantom was shifted 0, 0.6, 1.2, 1.8, 2.4, 3.0, and 5 mm in three orthogonal directions, immediately after the full-trajectory CBCT but prior to treatment. Depending on the scenario of shift, a nine-field fixed gantry intensity-modulated radiation therapy (IMRT) plan and/or a four partial-posterior-arcs volume-modulated radiation therapy (VMAT) plan were delivered. For the IMRT plan, three IMR images were acquired sequentially every 200 monitor units (MU) at each treatment angle. For the VMAT plan, one IMR image was acquired every 15° of each arc. For each IMR image, the software-reported 2D shift was compared with the ground truth. Certain tests were repeated with 1°, 2°, and 3° of rotation, pitch, and roll, respectively. Some of these tests were also repeated independently on separate days.

RESULTS

Based on the group of tests that involved only the IMRT delivery, the maximum standard deviation of the software-reported shifts for each set of three IMR images was 0.16 mm, with 95th percentile at 0.02 mm. For translational shift, the maximum registration error was 0.44 mm, with 95th percentile at 0.23 mm. Left unaccounted for, rotation and pitch degraded the registration accuracy mainly in the longitudinal direction, while roll degraded it mainly in the lateral direction. The degradation of registration accuracy is positively related to the degree of rotation, pitch, and roll. The maximum registration errors under 3° rotation, pitch, and roll were 2.97, 1.44, 2.72 mm, respectively. Based on the group of tests that compared IMRT delivery with VMAT delivery, the registration errors slightly increased as magnitude of shifts increased; however, they were well under the 0.5-mm threshold. No significant differences in registration errors were observed between IMRT and VMAT deliveries. In addition, the variation in registration errors among different days was limited for both IMRT and VMAT deliveries.

CONCLUSIONS

Our proprietary software has high repeatability, both intrafractionally and interfractionally, and high accuracy in registering IMR images with the reference projections for motion monitoring, regardless of the magnitude of shifts or treatment delivery technique. Rotation, pitch, and roll degrade registration accuracy and need to be accounted for in the future work.

Meta-optimization for fully automated radiation therapy treatment planning

Objective. Radiation therapy treatment planning is a time-consuming process involving iterative adjustments of hyperparameters. To automate the treatment planning process, we propose a meta-optimization framework, called MetaPlanner (MP).Approach. Our MP algorithm automates planning by performing meta-optimization of treatment planning hyperparameters. The algorithm uses a derivative-free method (i.e. parallel Nelder-Mead simplex search) to search for weight configurations that minimize a meta-scoring function. Meta-scoring is performed by constructing a tier list of the relevant considerations (e.g. dose homogeneity, conformity, spillage, and OAR sparing) to mimic the clinical decision-making process. Additionally, we have made our source code publicly available via github.Main results. The proposed MP method is evaluated on two datasets (21 prostate cases and 6 head and neck cases) collected as part of clinical workflow. MP is applied to both IMRT and VMAT planning and compared to a baseline of manual VMAT plans. MP in both IMRT and VMAT scenarios has comparable or better performance than manual VMAT planning for all evaluated metrics.Significance. Our proposed MP provides a general framework for fully automated treatment planning that produces high quality treatment plans. Our MP method promises to substantially reduce the workload of treatment planners while maintaining or improving plan quality.

Deep learning driven predictive treatment planning for adaptive radiotherapy of lung cancer

BACKGROUND AND PURPOSE

To develop a novel deep learning algorithm of sequential analysis, Seq2Seq, for predicting weekly anatomical changes of lung tumor and esophagus during definitive radiotherapy, incorporate the potential tumor shrinkage into a predictive treatment planning paradigm, and improve the therapeutic ratio.

METHODS AND MATERIALS

Seq2Seq starts with the primary tumor and esophagus observed on the planning CT to predict their geometric evolution during radiotherapy on a weekly basis, and subsequently updates the predictions with new snapshots acquired via weekly CBCTs. Seq2Seq is equipped with convolutional long short term memory to analyze the spatial-temporal changes of longitudinal images, trained and validated using a dataset including sixty patients. Predictive plans were optimized according to each weekly prediction and made ready for weekly deployment to mitigate the clinical burden of online weekly replanning.

RESULTS

Seq2Seq tracks structural changes well: DICE between predicted and actual weekly tumor and esophagus were (0.83 ± 0.10, 0.79 ± 0.14, 0.78 ± 0.12, 0.77 ± 0.12, 0.75 ± 0.12, 0.71 ± 0.17), and (0.72 ± 0.16, 0.73 ± 0.11, 0.75 ± 0.08, 0.74 ± 0.09, 0.72 ± 0.14, 0.71 ± 0.14), respectively, while the average Hausdorff distances were within 2 mm. Evaluating dose to the actual weekly tumor and esophagus, a 4.2 Gy reduction in esophagus mean dose while maintaining 60 Gy tumor coverage was achieved with the predictive weekly plans, compared to the plan optimized using the initial tumor and esophagus alone, primarily due to noticeable tumor shrinkage during radiotherapy.

CONCLUSION

It is feasible to predict the longitudinal changes of tumor and esophagus with the Seq2Seq, which could lead to improving the efficiency and effectiveness of lung adaptive radiotherapy.

Artificial Intelligence in Radiation Therapy

Artificial intelligence (AI) has great potential to transform the clinical workflow of radiotherapy. Since the introduction of deep neural networks, many AI-based methods have been proposed to address challenges in different aspects of radiotherapy. Commercial vendors have started to release AI-based tools that can be readily integrated to the established clinical workflow. To show the recent progress in AI-aided radiotherapy, we have reviewed AI-based studies in five major aspects of radiotherapy including image reconstruction, image registration, image segmentation, image synthesis, and automatic treatment planning. In each section, we summarized and categorized the recently published methods, followed by a discussion of the challenges, concerns, and future development. Given the rapid development of AI-aided radiotherapy, the efficiency and effectiveness of radiotherapy in the future could be substantially improved through intelligent automation of various aspects of radiotherapy.

Automated and Clinically Optimal Treatment Planning for Cancer Radiotherapy

Each year, approximately 18 million new cancer cases are diagnosed worldwide, and about half must be treated with radiotherapy. A successful treatment requires treatment planning with the customization of penetrating radiation beams to sterilize cancerous cells without harming nearby normal organs and tissues. This process currently involves extensive manual tuning of parameters by an expert planner, making it a time-consuming and labor-intensive process, with quality and immediacy of critical care dependent on the planner's expertise. To improve the speed, quality, and availability of this highly specialized care, Memorial Sloan Kettering Cancer Center developed and applied advanced optimization tools to this problem (e.g., using hierarchical constrained optimization, convex approximations, and Lagrangian methods). This resulted in both a greatly improved radiotherapy treatment planning process and the generation of reliable and consistent high-quality plans that reflect clinical priorities. These improved techniques have been the foundation of high-quality treatments and have positively impacted over 4,000 patients to date, including numerous patients in severe pain and in urgent need of treatment who might have otherwise required longer hospital stays or undergone unnecessary surgery to control the progression of their disease. We expect that the wide distribution of the system we developed will ultimately impact patient care more broadly, including in resource-constrained countries.

Deep learning auto-segmentation and automated treatment planning for trismus risk reduction in head and neck cancer radiotherapy

Background and Purpose

Reducing trismus in radiotherapy for head and neck cancer (HNC) is important. Automated deep learning (DL) segmentation and automated planning was used to introduce new and rarely segmented masticatory structures to study if trismus risk could be decreased.

Materials and Methods

Auto-segmentation was based on purpose-built DL, and automated planning used our in-house system, ECHO. Treatment plans for ten HNC patients, treated with 2 Gy × 35 fractions, were optimized (ECHO₀). Six manually segmented OARs were replaced with DL auto-segmentations and the plans re-optimized (ECHO₁). In a third set of plans, mean doses for auto-segmented ipsilateral masseter and medial pterygoid (MI_Mean, MPI_Mean), derived from a trismus risk model, were implemented as dose-volume objectives (ECHO₂). Clinical dose-volume criteria were compared between the two scenarios (ECHO₀vs. ECHO₁; ECHO₁vs. ECHO₂; Wilcoxon signed-rank test; significance: p < 0.01).

Results

Small systematic differences were observed between the doses to the six auto-segmented OARs and their manual counterparts (median: ECHO₁ = 6.2 (range: 0.4, 21) Gy vs. ECHO₀ = 6.6 (range: 0.3, 22) Gy; p = 0.007), and the ECHO₁ plans provided improved normal tissue sparing across a larger dose-volume range. Only in the ECHO₂ plans, all patients fulfilled both MI_Mean and MPI_Mean criteria. The population median MI_Mean and MPI_Mean were considerably lower than those suggested by the trismus model (ECHO₀: MI_Mean = 13 Gy vs. ≤42 Gy; MPI_Mean = 29 Gy vs. ≤68 Gy).

Conclusions

Automated treatment planning can efficiently incorporate new structures from DL auto-segmentation, which results in trismus risk sparing without deteriorating treatment plan quality. Auto-planning and deep learning auto-segmentation together provide a powerful platform to further improve treatment planning.

Failure mode and effect analysis for linear accelerator-based paraspinal stereotactic body radiotherapy

Introduction

Paraspinal stereotactic body radiotherapy (SBRT) involves risks of severe complications. We evaluated the safety of the paraspinal SBRT program in a large academic hospital by applying failure modes and effects analysis.

Methods

The analysis was conducted by a multidisciplinary committee (two therapists, one dosimetrist, four physicists, and two radiation oncologists). The paraspinal SBRT workflow was segmented into four phases (simulation, treatment planning, delivery, and machine quality assurance (QA)). Each phase was further divided into a sequence of sub‐processes. Potential failure modes (PFM) were identified from each subprocess and scored in terms of the frequency of occurrence, severity and detectability, and a risk priority number (RPN). High‐risk PFMs were identified based on RPN and were studied for root causes using fault tree analysis.

Results

Our paraspinal SBRT process was characterized by eight simulations, 11 treatment planning, nine delivery, and two machine QA sub‐processes. There were 18, 29, 19, and eight PFMs identified from simulation, planning, treatment, and machine QA, respectively. The median RPN of the PFMs was 62.9 for simulation, 68.3 for planning, 52.9 for delivery, and 22.0 for machine QA. The three PFMs with the highest RPN were: previous radiotherapy outside the institution is not accurately evaluated (RPN: 293.3), incorrect registration between diagnostic magnetic resonance imaging and simulation computed tomography causing incorrect contours (273.0), and undetected patient movement before ExacTrac baseline (217.8). Remedies to the high RPN failures were implemented, including staff education, standardized magnetic resonance imaging acquisition parameters, and an image fusion process, and additional QA on beam steering.

Conclusions

A paraspinal SBRT workflow in a large clinic was evaluated using a multidisciplinary and systematic risk analysis, which led to feasible solutions to key root causes. Treatment planning was a major source of PFMs that systematically affect the safety and quality of treatments. Accurate evaluation of external treatment records remains a challenge.

NeuralDAO: Incorporating neural network generated dose into direct aperture optimization for end-to-end IMRT planning

PURPOSE

Thecurrent practice in intensity-modulated radiation therapy (IMRT) planning almost always includes different dose calculation strategies for plan optimization and final dose verification. The accurate Monte Carlo (MC) dose algorithm is considered to be time-consuming for the optimization. Thus a fast, simplified dose algorithm is used in the optimization. The significant differences between the optimized dose and the delivered dose lead to tediously planning loops and potentially suboptimal solutions. This work aims to develop an IMRT optimization algorithm to minimize the dose discrepancy so that the delivered dose can be optimized in a holistic, end-to-end manner.

METHODS

The proposed algorithm, namely NeuralDAO, integrates a neural dose network into the column generation (CG) direct aperture optimization (DAO) formulation for step-and-shoot IMRT planning. The neural dose network is designed and trained to produce doses of MC-level accuracy within few milliseconds. Its differentiability is fully exploited to compute gradients for identifying potential aperture shapes. A prototype of NeuralDAO was developed in PyTorch and available to the public. Five lung patient cases have been studied. Dosimetric accuracy was compared with the MC dose. Plan quality and time were compared with a state-of-the-art (SoA) dose-correct algorithm. Statistical analysis was performed by Wilcoxon signed-rank test.

RESULTS

The average gamma passing rate at 2 mm/2% is 99.7% between the optimized and delivered doses. The convergence process produced by NeuralDAO is virtually identical to that produced by an MC-based DAO. The average dose calculation time is 12.1 ms for an aperture on GPU. One session of optimization took 10-36 min. Compared with the SoA, better conformity index and homogeneity index were observed for the target. The esophagus was significantly spared. Significant reductions were observed for the replanning number and the planning time.

CONCLUSIONS

A new DAO algorithm based on the neural dose network has been developed. The results suggest that this algorithm minimizes the discrepancy between the optimized and delivered doses, which offers a promising approach to reduce the time and effort required in IMRT planning. This work demonstrates the possibility of applying the neural network in IMRT optimization. It is of great potential to extend this algorithm to other treatment modalities.

Fully automated treatment planning for MLC-based robotic radiotherapy

PURPOSE

To propose and validate a fully automated multicriterial treatment planning solution for a CyberKnife® equipped with an InCise^TM 2 multileaf collimator.

METHODS

The AUTO BAO plans are generated using fully automated prioritized multicriterial optimization (AUTO MCO) of pencil-beam fluence maps with integrated noncoplanar beam angle optimization (BAO), followed by MLC segment generation. Both the AUTO MCO and segmentation algorithms have been developed in-house. AUTO MCO generates for each patient a single, high-quality Pareto-optimal IMRT plan. The segmentation algorithm then accurately mimics the AUTO MCO 3D dose distribution, while considering all candidate beams simultaneously, rather than replicating the fluence maps. Pencil-beams, segment dose depositions, and final dose calculations are performed with a stand-alone version of the clinical dose calculation engine. For validation, AUTO BAO plans were generated for 33 prostate SBRT patients and compared to reference plans (REF) that were manually generated with the commercial treatment planning system (TPS), in absence of time pressure. REF plans were also compared to AUTO RB plans, for which fluence map optimization was performed for the beam angle configuration used in the REF plan, and the segmentation could use all these beams or only a subset, depending on the dosimetry.

RESULTS

AUTO BAO plans were clinically acceptable and dosimetrically similar to REF plans, but had on average reduced numbers of beams ((beams in AUTO BAO)/(beams in REF) (relative improvement): 24.7/48.3 (-49%)), segments (59.5/98.9 (-40%)), and delivery times (17.1/22.3 min. (-23%)). Dosimetry of AUTO RB and REF were also similar, but AUTO RB used on average fewer beams (38.0/48.3 (-21%)) and had on average shorter delivery times (18.6/22.3 min. (-17%)). Delivered Monitor Units (MU) were similar for all three planning approaches.

CONCLUSIONS

A new, vendor-independent optimization workflow for fully automated generation of deliverable high-quality CyberKnife® plans was proposed, including BAO. Compared to manual planning with the commercial TPS, fraction delivery times were reduced by 5.3 min. (-23%) due to large reductions in beam and segment numbers.

Solving the volumetric modulated arc therapy (VMAT) problem using a sequential convex programming method

The volumetric modulated arc therapy (VMAT) problem is highly non-convex and much more difficult than the fixed-field intensity modulated radiotherapy optimization problem. To solve it efficiently, we propose a sequential convex programming algorithm that solves a sequence of convex optimization problems. Beginning by optimizing the aperture weights of many (72) evenly distributed beams using the beam's eye view of the target from each direction as the initial aperture shape, the search space is constrained to allowing the leaves to move within a pre-defined step-size. A convex approximation problem is introduced and solved to optimize the leaf positions and the aperture weights within the search space. The algorithm is equipped with both local and global search strategies, whereby a global search is followed by a local search: a large step-size results in a global search with a less accurate convex approximation, followed by a small step-size local search with an accurate convex approximation. The performance of the proposed algorithm is tested on three patients with three different disease sites (paraspinal, prostate and oligometastasis). The algorithm generates VMAT plans comparable to the ideal 72-beam fluence map optimized plans (i.e. IMRT plans before leaf sequencing) in 14 iterations and 36 mins on average. The algorithm is also tested on a small down-sampled prostate case for which we could computationally afford to obtain the ground-truth by solving the non-convex mixed-integer optimization problem exactly. This general algorithm is able to produce results essentially equivalent to the ground-truth but 12 times faster. The algorithm is also scalable and can handle real clinical cases, whereas the ground-truth solution using mixed-integer optimization can only be obtained for highly down-sampled cases.

Technical Note: Synthetic treatment beam imaging for motion monitoring during spine SBRT treatments - a phantom study

PURPOSE

One of the biggest challenges in applying megavoltage (MV) treatment beam imaging for monitoring spine motion in stereotactic body radiotherapy (SBRT) is the small beam apertures in the images due to strong beam modulations in IMRT planning. The purpose of this study is to investigate the feasibility of a markerless motion tracking method in spine SBRT delivery using a novel enhanced synthetic treatment beam (ESTB) imaging technique.

METHODS

Three clinical spine SBRT plans using 6XFFF beams and sliding window IMRT technique were transferred to a thorax phantom and delivered by a TrueBeam machine. Before delivery, the phantom was aligned to the plan isocenter using CBCT setup and verified with a second CBCT, and then, 2 mm shifts were introduced in both the craniocaudal (CC) and the left-right (LR) directions with the couch. During beam delivery, MV images were continuously taken with an electronic portal imaging device (EPID) and automatically grabbed by Varian iTools Capture software with a frame rate of 11.6 Hz. After preprocessing for scatter correction and beam intensity compensation, every 50 frames of MV images were combined to generate a series of ESTB images for each beam. The ESTB images were then registered to the projections of the verification CBCT at the matched beam angles to detect the 2 mm shifts.

RESULTS

Compared to snapshot MV images, the ESTB images had significantly enlarged fields of view (FOVs) and improved image quality. Based on two-dimensional (2D) rigid registration, the ESTB image to CBCT projection matching showed submillimeter accuracy in detecting motion. Specifically, the root mean square errors in detecting the LR/CC shifts were 0.35/0.28, 0.32/0.35, 0.63/0.44, 0.55/0.51, and 0.69/0.42 mm at gantry angles 180, 160, 140, 120, and 100, respectively.

CONCLUSION

Our results in the phantom study suggest that ESTB images from a sliding window IMRT plan can be used to detect spine motion, with submillimeter precision in the 2D plane perpendicular to the beam.

Automating proton treatment planning with beam angle selection using Bayesian optimization

PURPOSE

To present a fully automated treatment planning process for proton therapy including beam angle selection using a novel Bayesian optimization approach and previously developed constrained hierarchical fluence optimization method.

METHODS

We adapted our in-house automated intensity modulated radiation therapy (IMRT) treatment planning system, which is based on constrained hierarchical optimization and referred to as ECHO (expedited constrained hierarchical optimization), for proton therapy. To couple this to beam angle selection, we propose using a novel Bayesian approach. By integrating ECHO with this Bayesian beam selection approach, we obtain a fully automated treatment planning framework including beam angle selection. Bayesian optimization is a global optimization technique which only needs to search a small fraction of the search space for slowly varying objective functions (i.e., smooth functions). Expedited constrained hierarchical optimization is run for some initial beam angle candidates and the resultant treatment plan for each beam configuration is rated using a clinically relevant treatment score function. Bayesian optimization iteratively predicts the treatment score for not-yet-evaluated candidates to find the best candidate to be optimized next with ECHO. We tested this technique on five head-and-neck (HN) patients with two coplanar beams. In addition, tests were performed with two noncoplanar and three coplanar beams for two patients.

RESULTS

For the two coplanar configurations, the Bayesian optimization found the optimal beam configuration after running ECHO for, at most, 4% of all potential configurations (23 iterations) for all patients (range: 2%-4%). Compared with the beam configurations chosen by the planner, the optimal configurations reduced the mandible maximum dose by 6.6 Gy and high dose to the unspecified normal tissues by 3.8 Gy, on average. For the two noncoplanar and three coplanar beam configurations, the algorithm converged after 45 iterations (examining <1% of all potential configurations).

CONCLUSIONS

A fully automated and efficient treatment planning process for proton therapy, including beam angle optimization was developed. The algorithm automatically generates high-quality plans with optimal beam angle configuration by combining Bayesian optimization and ECHO. As the Bayesian optimization is capable of handling complex nonconvex functions, the treatment score function which is used in the algorithm to evaluate the dose distribution corresponding to each beam configuration can contain any clinically relevant metric.

Automated proton treatment planning with robust optimization using constrained hierarchical optimization

PURPOSE

We present a method for fully automated generation of high quality robust proton treatment plans using hierarchical optimization. To fill the gap between the two common extreme robust optimization approaches, that is, stochastic and worst-case, a robust optimization approach based on the p-norm function is used whereby a single parameter, p , can be used to control the level of robustness in an intuitive way.

METHODS

A fully automated approach to treatment planning using Expedited Constrained Hierarchical Optimization (ECHO) is implemented in our clinic for photon treatments. ECHO strictly enforces critical (inviolable) clinical criteria as hard constraints and improves the desirable clinical criteria sequentially, as much as is feasible. We extend our in-house developed ECHO codes for proton therapy and integrate it with a new approach for robust optimization. Multiple scenarios accounting for both setup and range uncertainties are included (13scenarios), and the maximum/mean/dose-volume constraints on organs-at-risk (OARs) and target are fulfilled in all scenarios. We combine the objective functions of the individual scenarios using the p-norm function. The p-norm with a parameter p = 1 or p = ∞ result in the stochastic or the worst-case approach, respectively; an intermediate robustness level is obtained by employing p -values in-between. While the worst-case approach only focuses on the worst-case scenario(s), the p-norm approach with a large p value ( p ≈ 20 ) resembles the worst-case approach without completely neglecting other scenarios. The proposed approach is evaluated on three head-and-neck (HN) patients and one water phantom with different parameters, p ∈ 1 , 2 , 5 , 10 , 20 . The results are compared against the stochastic approach (p-norm approach with p = 1 ) and the worst-case approach, as well as the nonrobust approach (optimized solely on the nominal scenario).

RESULTS

The proposed algorithm successfully generates automated robust proton plans on all cases. As opposed to the nonrobust plans, the robust plans have narrower dose volume histogram (DVH) bands across all 13 scenarios, and meet all hard constraints (i.e., maximum/mean/dose-volume constraints) on OARs and the target for all scenarios. The spread in the objective function values is largest for the stochastic approach ( p = 1 ) and decreases with increasing p toward the worst-case approach. Compared to the worst-case approach, the p-norm approach results in DVH bands for clinical target volume (CTV) which are closer to the prescription dose at a negligible cost in the DVH for the worst scenario, thereby improving the overall plan quality. On average, going from the worst-case approach to the p-norm approach with p = 20 , the median objective function value across all the scenarios is improved by 15% while the objective function value for the worst scenario is only degraded by 3%.

CONCLUSION

An automated treatment planning approach for proton therapy is developed, including robustness, dose-volume constraints, and the ability to control the robustness level using the p-norm parameter p , to fit the priorities deemed most important.

Integrating soft and hard dose-volume constraints into hierarchical constrained IMRT optimization

PURPOSE

Dose-volume constraints (DVCs) continue to be common features in intensity-modulated radiation therapy (IMRT) prescriptions, but they are non-convex and difficult to incorporate. We propose computationally efficient methods to incorporate dose-volume constraints (DVCs) into automated IMRT planning.

METHODS

We propose a two-phase approach: in phase-1, we solve a convex approximation with DVCs. Although this convex approximation does not guarantee DVC satisfaction, it provides crucial initial information about voxels likely to receive doses below DVC thresholds. Subsequently, phase-2 solves an optimization problem with maximum dose constraints imposed on those subthreshold voxels. We further categorize DVCs into hard- and soft-DVCs, where hard-DVCs are strictly enforced by the optimization and soft-DVCs are encouraged in the objective function. We tested this approach in our automated treatment planning system which is based on hierarchical constrained optimization. Performance is demonstrated on a series of paraspinal, lung, oligometastasis, and prostate cases as well as a small paraspinal case for which we can computationally afford to obtain a ground-truth by solving a non-convex optimization problem.

RESULTS

The proposed algorithm successfully meets all the hard-DVCs while increasing the overall computational time of the baseline planning process (without DVCs) by 20%, 10%, and 11% for paraspinal, oligometastasis, and prostate cases, respectively. For a soft-DVC applied to the lung case, the dose-volume histogram curve moves toward the desired direction and the computational time is increased by 11%. For a low-resolution paraspinal case, the ground-truth solution process using mixed-integer programming methods required 15 h while the proposed algorithm converges in only 2 min with a proximal solution.

CONCLUSIONS

A computationally tractable algorithm to handle hard- and soft-DVCs is developed which is capable of satisfying DVCs without any parameter tweaking. Although the algorithm is demonstrated in our in-house developed automated treatment planning system, it can potentially be used in any constrained optimization framework.

Clinical Experience of Automated SBRT Paraspinal and Other Metastatic Tumor Planning With Constrained Hierarchical Optimization

Purpose

We report on the clinical performance of a fully automated approach to treatment planning based on a Pareto optimal, constrained hierarchical optimization algorithm, named Expedited Constrained Hierarchical Optimization (ECHO).

Methods and materials

From April 2017 to October 2018, ECHO produced 640 treated plans for 523 patients who underwent stereotactic body radiation therapy (RT) for paraspinal and other metastatic tumors. A total of 182 plans were for 24 Gy in a single fraction, 387 plans were for 27 Gy in 3 fractions, and the remainder were for other prescriptions or fractionations. Of the plans, 84.5% were for paraspinal tumors, with 69, 302, and 170 in the cervical, thoracic, and lumbosacral spine, respectively. For each case, after contouring, a template plan using 9 intensity modulated RT fields based on disease site and tumor location was sent to ECHO through an application program interface plug-in from the treatment planning system. ECHO returned a plan that satisfied all critical structure hard constraints with optimal target volume coverage and the lowest achievable normal tissue doses. Upon ECHO completion, the planner received an e-mail indicating the plan was ready for review. The plan was accepted if all clinical criteria were met. Otherwise, a limited number of parameters could be adjusted for another ECHO run.

Results

The median planning target volume size was 84.3 cm³ (range, 6.9-633.2). The median time to produce 1 ECHO plan was 63.5 minutes (range, 11-340 minutes) and was largely dependent on the field sizes. Of the cases, 79.7% required 1 run to produce a clinically accepted plan, 13.3% required 1 additional run with minimal parameter adjustments, and 7.0% required ≥2 additional runs with significant parameter modifications. All plans met or bettered the institutional clinical criteria.

Conclusions

We successfully implemented automated stereotactic body RT paraspinal and other metastatic tumors planning. ECHO produced high-quality plans, improved planning efficiency and robustness, and enabled expedited treatment planning at our clinic.