Model based patient pre-selection for intensity-modulated proton therapy (IMPT) using automated treatment planning and machine learning

The emerging role of Artificial Intelligence in proton therapy: A review

Artificial intelligence (AI) has made a tremendous impact in the space of healthcare, and proton therapy is not an exception. Proton therapy has witnessed growing popularity in oncology over recent decades, and researchers are increasingly looking to develop AI and machine learning tools to aid in various steps of the treatment planning and delivery processes. This review delves into the emergent role of AI in proton therapy, evaluating its development, advantages, intended clinical contexts, and areas of application. Through the analysis of 76 studies, we aim to underscore the importance of AI applications in advancing proton therapy and to highlight their prospective influence on clinical practices.

Reducing the lateral dose penumbra in IMPT by incorporating transmission pencil beams

OBJECTIVE

In intensity-modulated proton therapy (IMPT), Bragg peaks result in steep distal dose fall-offs, while the lateral IMPT dose fall-off is often less steep than in photon therapy. High-energy pristine transmission ('shoot through') pencil beams have no Bragg peak in the patient, but show a sharp lateral penumbra at the target level. We investigated whether combining Bragg peaks with Transmission pencil beams ('IMPT&TPB') could improve head-and-neck plans by exploiting the steep lateral dose fall-off of transmission pencil beams.

APPROACH

Our system for automated multi-criteria IMPT plan optimisation was extended for combined optimisation of BPs and TPBs. The system generates for each patient a Pareto-optimal plan using a generic 'wish-list' with prioritised planning objectives and hard constraints. For eight nasopharynx cancer patients (NPC) and eight oropharynx cancer (OPC) patients, the IMPT&TPB plan was compared to the competing conventional IMPT plan with only Bragg peaks, which was generated with the same optimiser, but without transmission pencil beams.

MAIN RESULTS

Clinical OAR and target constraints were met in all plans. By allowing transmission pencil beams in the optimisation, on average 14 of the 25 investigated OAR plan parameters significantly improved for NPC, and 9 of the 17 for OPC, while only one OPC parameter showed small but significant deterioration. Non-significant differences were found in the remaining parameters. In NPC, cochlea Dmean reduced by up to 17.5 Gy and optic nerve D2% by up to 11.1 Gy.

CONCLUSION

Compared to IMPT, IMPT&TPB resulted in comparable target coverage with overall superior OAR sparing, the latter originating from steeper dose fall-offs close to OARs.

Validation of Fully Automated Robust Multicriterial Treatment Planning for Head and Neck Cancer IMPT

PURPOSE

Our purpose was to compare robust intensity modulated proton therapy (IMPT) plans, automatically generated with wish-list-based multicriterial optimization as implemented in Erasmus-iCycle, with manually created robust clinical IMPT plans for patients with head and neck cancer.

METHODS AND MATERIALS

Thirty-three patients with head and neck cancer were retrospectively included. All patients were previously treated with a manually created IMPT plan with 7000 cGy dose prescription to the primary tumor (clinical target volume [CTV]7000) and 5425 cGy dose prescription to the bilateral elective volumes (CTV5425). Plans had a 4-beam field configuration and were generated with scenario-based robust optimization (21 scenarios, 3-mm setup error, and ±3% density uncertainty for the CTVs). Three clinical plans were used to configure the Erasmus-iCycle wish-list for automated generation of robust IMPT plans for the other 30 included patients, in line with clinical planning requirements. Automatically and manually generated IMPT plans were compared for (robust) target coverage, organ-at-risk (OAR) doses, and normal tissue complication probabilities (NTCP). No manual fine-tuning of automatically generated plans was performed.

RESULTS

For all automatically generated plans, voxel-wise minimum D98% values for the CTVs were within clinical constraints and similar to manual plans. All investigated OAR parameters were favorable in the automatically generated plans (all P < .001). Median reductions in mean dose to OARs went up to 667 cGy for the inferior pharyngeal constrictor muscle, and median reductions in D0.03cm3 in serial OARs ranged up to 1795 cGy for the spinal cord surface. The observed lower mean dose in parallel OARs resulted in statistically significant lower NTCP for xerostomia (grade ≥2: 34.4% vs 38.0%; grade ≥3: 9.0% vs 10.2%) and dysphagia (grade ≥2: 11.8% vs 15.0%; grade ≥3: 1.8% vs 2.8%).

CONCLUSIONS

Erasmus-iCycle was able to produce IMPT dose distributions fully automatically with similar (robust) target coverage and improved OAR doses and NTCPs compared with clinical manual planning, with negligible hands-on planning workload.

Evaluation of an automated clinical decision system with deep learning dose prediction and NTCP model for prostate cancer proton therapy

Objective.To evaluate the feasibility of using a deep learning dose prediction approach to identify patients who could benefit most from proton therapy based on the normal tissue complication probability (NTCP) model.Approach.Two 3D UNets were established to predict photon and proton doses. A dataset of 95 patients with localized prostate cancer was randomly partitioned into 55, 10, and 30 for training, validation, and testing, respectively. We selected NTCP models for late rectum bleeding and acute urinary urgency of grade 2 or higher to quantify the benefit of proton therapy. Propagated uncertainties of predicted ΔNTCPs resulting from the dose prediction errors were calculated. Patient selection accuracies for a single endpoint and a composite evaluation were assessed under different ΔNTCP thresholds.Main results.Our deep learning-based dose prediction technique can reduce the time spent on plan comparison from approximately 2 days to as little as 5 seconds. The expanded uncertainty of predicted ΔNTCPs for rectum and bladder endpoints propagated from the dose prediction error were 0.0042 and 0.0016, respectively, which is less than one-third of the acceptable tolerance. The averaged selection accuracies for rectum bleeding, urinary urgency, and composite evaluation were 90%, 93.5%, and 93.5%, respectively.Significance.Our study demonstrates that deep learning dose prediction and NTCP evaluation scheme could distinguish the NTCP differences between photon and proton treatment modalities. In addition, the dose prediction uncertainty does not significantly influence the decision accuracy of NTCP-based patient selection for proton therapy. Therefore, automated deep learning dose prediction and NTCP evaluation schemes can potentially be used to screen large patient populations and to avoid unnecessary delays in the start of prostate cancer radiotherapy in the future.

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.

Patient selection for proton therapy using Normal Tissue Complication Probability with deep learning dose prediction for oropharyngeal cancer

BACKGROUND

In cancer care, determining the most beneficial treatment technique is a key decision affecting the patient's survival and quality of life. Patient selection for proton therapy (PT) over conventional radiotherapy (XT) currently entails comparing manually generated treatment plans, which requires time and expertise.

PURPOSE

We developed an automatic and fast tool, AI-PROTIPP (Artificial Intelligence Predictive Radiation Oncology Treatment Indication to Photons/Protons), that assesses quantitatively the benefits of each therapeutic option. Our method uses deep learning (DL) models to directly predict the dose distributions for a given patient for both XT and PT. By using models that estimate the Normal Tissue Complication Probability (NTCP), namely the likelihood of side effects to occur for a specific patient, AI-PROTIPP can propose a treatment selection quickly and automatically.

METHODS

A database of 60 patients presenting oropharyngeal cancer, obtained from the Cliniques Universitaires Saint Luc in Belgium, was used in this study. For every patient, a PT plan and an XT plan were generated. The dose distributions were used to train the two dose DL prediction models (one for each modality). The model is based on U-Net architecture, a type of convolutional neural network currently considered as the state of the art for dose prediction models. A NTCP protocol used in the Dutch model-based approach, including grades II and III xerostomia and grades II and III dysphagia, was later applied in order to perform automatic treatment selection for each patient. The networks were trained using a nested cross-validation approach with 11-folds. We set aside three patients in an outer set and each fold consists of 47 patients in training, five in validation and five for testing. This method allowed us to assess our method on 55 patients (five patients per test times the number of folds).

RESULTS

The treatment selection based on the DL-predicted doses reached an accuracy of 87.4% for the threshold parameters set by the Health Council of the Netherlands. The selected treatment is directly linked with these threshold parameters as they express the minimal gain brought by the PT treatment for a patient to be indicated to PT. To validate the performance of AI-PROTIPP in other conditions, we modulated these thresholds, and the accuracy was above 81% for all the considered cases. The difference in average cumulative NTCP per patient of predicted and clinical dose distributions is very similar (less than 1% difference).

CONCLUSIONS

AI-PROTIPP shows that using DL dose prediction in combination with NTCP models to select PT for patients is feasible and can help to save time by avoiding the generation of treatment plans only used for the comparison. Moreover, DL models are transferable, allowing, in the future, experience to be shared with centers that would not have PT planning expertise.

Pencil-beam Delivery Pattern Optimization Increases Dose Rate for Stereotactic FLASH Proton Therapy

PURPOSE

FLASH dose rates >40 Gy/s are readily available in proton therapy (PT) with cyclotron-accelerated beams and pencil-beam scanning (PBS). The PBS delivery pattern will affect the local dose rate, as quantified by the PBS dose rate (PBS-DR), and therefore needs to be accounted for in FLASH-PT with PBS, but it is not yet clear how. Our aim was to optimize patient-specific scan patterns for stereotactic FLASH-PT of early-stage lung cancer and lung metastases, maximizing the volume irradiated with PBS-DR >40 Gy/s of the organs at risk voxels irradiated to >8 Gy (FLASH coverage).

METHODS AND MATERIALS

Plans to 54 Gy/3 fractions with 3 equiangular coplanar 244 MeV proton shoot-through transmission beams for 20 patients were optimized with in-house developed software. Planning target volume-based planning with a 5 mm margin was used. Planning target volume ranged from 4.4 to 84 cc. Scan-pattern optimization was performed with a Genetic Algorithm, run in parallel for 20 independent populations (islands). Mapped crossover, inversion, swap, and shift operators were applied to achieve (local) optimality on each island, with migration between them for global optimality. The cost function was chosen to maximize the FLASH coverage per beam at >8 Gy, >40 Gy/s, and 40 nA beam current. The optimized patterns were evaluated on FLASH coverage, PBS-DR distribution, and population PBS-DR-volume histograms, compared with standard line-by-line scanning. Robustness against beam current variation was investigated.

RESULTS

The optimized patterns have a snowflake-like structure, combined with outward swirling for larger targets. A population median FLASH coverage of 29.0% was obtained for optimized patterns compared with 6.9% for standard patterns, illustrating a significant increase in FLASH coverage for optimized patterns. For beam current variations of 5 nA, FLASH coverage varied between -6.1%-point and 2.2%-point for optimized patterns.

CONCLUSIONS

Significant improvements on the PBS-DR and, hence, on FLASH coverage and potential healthy-tissue sparing are obtained by sequential scan-pattern optimization. The optimizer is flexible and may be further fine-tuned, based on the exact conditions for FLASH.

Trade-off in healthy tissue sparing of FLASH and fractionation in stereotactic proton therapy of lung lesions with transmission beams

PURPOSE AND OBJECTIVE

Besides a dose-rate threshold of 40-100 Gy/s, the FLASH effect may require a dose >3.5-7 Gy. Even in hypofractioned treatments, with all beams delivered in each fraction (ABEF), most healthy tissue is irradiated to a lower fraction dose. This can be circumvented by single-beam-per-fraction (SBPF) delivery, with a loss of healthy tissue sparing by fractionation. We investigated the trade-off between FLASH and loss of fractionation in SBPF stereotactic proton therapy of lung cancer and determined break-even FLASH-enhancement ratios (FERs).

MATERIALS AND METHODS

Treatment plans for 12 patients were generated. GTV delineations were available and a 5 mm GTV-PTV margin was applied. Equiangular arrangements of 3, 5, 7, and 9 244 MeV proton transmission beams were used. To facilitate SBPF, the number of fractions was equal to the number of beams. Iso-effective fractionation schedules with a single field uniform dose prescription were used: D_95%,PTV = 100%D_pres per beam. All plans were evaluated in terms of dose to lung and conformity of dose to target of FLASH-enhanced biologically equivalent dose (EQD2).

RESULTS

Compared to ABEF, SBPF resulted in a median increase of EQD2_mean to healthy lung of 56%, 58%, 55% and 54% in plans with 3, 5, 7 and 9 fractions respectively and of 90%, 108%, 106% and 102% in V_{100% EQD2}, quantifying conformity. This can be compensated for by FERs of at least 1.28, 1.32, 1.30 and 1.23 respectively for EQD2_mean and 1.29, 1.18, 1.28 and 1.15 for V_100%,EQD2.

CONCLUSION

A FLASH effect outweighing the loss of fractionation in SBPF may be achieved in stereotactic lung treatments. The trade-off with fractionation depends on the conditions under which the FLASH effect occurs. Better understanding of the underlying biology and the impact of delivery conditions is needed.

Improving workflow for adaptive proton therapy with predictive anatomical modelling: A proof of concept

PURPOSE

To demonstrate predictive anatomical modelling for improving the clinical workflow of adaptive intensity-modulated proton therapy (IMPT) for head and neck cancer.

METHODS

10 radiotherapy patients with nasopharyngeal cancer were included in this retrospective study. Each patient had a planning CT, weekly verification CTs during radiotherapy and predicted weekly CTs from our anatomical model. Predicted CTs were used to create predicted adaptive plans in advance with the aim of maintaining clinically acceptable dosimetry. Adaption was triggered when the increase in mean dose (Dmean) to the parotid glands exceeded 3 Gy(RBE). We compared the accumulated dose of two adaptive IMPT strategies: 1) Predicted plan adaption: One adaptive plan per patient was optimised on a predicted CT triggered by replan criteria. 2) Standard replan: One adaptive plan was created reactively in response to the triggering weekly CT.

RESULTS

Statistical analysis demonstrates that the accumulated dose differences between two adaptive strategies are not significant (p > 0.05) for CTVs and OARs. We observed no meaningful differences in D95 between the accumulated dose and the planned dose for the CTVs, with mean differences to the high-risk CTV of -1.20 %, -1.23 % and -1.25 % for no adaption, standard and predicted plan adaption, respectively. The accumulated parotid Dmean using predicted plan adaption is within 3 Gy(RBE) of the planned dose and 0.31 Gy(RBE) lower than the standard replan approach on average.

CONCLUSION

Prediction-based replanning could potentially enable adaptive therapy to be delivered without treatment gaps or sub-optimal fractions, as can occur during a standard replanning strategy, though the benefit of using predicted plan adaption over the standard replan was not shown to be statistically significant with respect to accumulated dose in this study. Nonetheless, a predictive replan approach can offer advantages in improving clinical workflow efficiency.

Proton Therapy for Squamous Cell Carcinoma of the Head and Neck: Early Clinical Experience and Current Challenges

Simple Summary

Proton therapy is a promising type of radiation therapy used to destroy tumor cells. It has the potential to further improve the outcomes for patients with head and neck cancer since it allows to minimize the radiation dose to vital structures around the tumor, leading to less toxicity. This paper describes the current experience worldwide with proton therapy in head and neck cancer.

Abstract

Proton therapy (PT) is a promising development in radiation oncology, with the potential to further improve outcomes for patients with squamous cell carcinoma of the head and neck (HNSCC). By utilizing the finite range of protons, healthy tissue can be spared from beam exit doses that would otherwise be irradiated with photon-based treatments. Current evidence on PT for HNSCC is limited to comparative dosimetric analyses and retrospective single-institution series. As a consequence, the recognized indications for the reimbursement of PT remain scarce in most countries. Nevertheless, approximately 100 PT centers are in operation worldwide, and initial experiences for HNSCC are being reported. This review aims to summarize the results of the early clinical experience with PT for HNSCC and the challenges that are currently faced.

Substantial Sparing of Organs at Risk with Modern Proton Therapy in Lung Cancer, but Altered Breathing Patterns Can Jeopardize Target Coverage

Simple Summary

Treatment of locally advanced non-small cell lung cancer (LA-NSCLC) is a fine balance between toxicity and cure. Modern proton therapy might offer a more gentle radiation treatment compared to state-of-the-art photon radiotherapy, but is also more susceptible to the influence of breathing motion and anatomical changes. In this study, the influence of such uncertainties on treatment delivery was thoroughly investigated. Modern proton therapy did indeed show potential to reduce the risk of toxicity for the heart and lungs. This potential was maintained under the influence of anatomical and delivery uncertainties. However, changes in breathing motion jeopardized the target dose distribution in a subset of patients. We therefore recommend imaging at onset or early in treatment to recognize these patients and adapt the treatment.

Abstract

Enhancing treatment of locally advanced non-small cell lung cancer (LA-NSCLC) by using pencil beam scanning proton therapy (PBS-PT) is attractive, but little knowledge exists on the effects of uncertainties occurring between the planning (Plan) and the start of treatment (Start). In this prospective simulation study, we investigated the clinical potential for PBS-PT under the influence of such uncertainties. Imaging with 4DCT at Plan and Start was carried out for 15 patients that received state-of-the-art intensity-modulated radiotherapy (IMRT). Three PBS-PT plans were created per patient: 3D robust single-field uniform dose (SFUD), 3D robust intensity-modulated proton therapy (IMPT), and 4D robust IMPT (4DIMPT). These were exposed to setup and range uncertainties and breathing motion at Plan, and changes in breathing motion and anatomy at Start. Target coverage and dose-volume parameters relevant for toxicity were compared. The organ at risk sparing at Plan was greatest with IMPT, followed by 4DIMPT, SFUD and IMRT, and persisted at Start. All plans met the preset criteria for target robustness at Plan. At Start, three patients had a lack of CTV coverage with PBS-PT. In conclusion, the clinical potential for heart and lung toxicity reduction with PBS-PT was substantial and persistent. Altered breathing patterns between Plan and Start jeopardized target coverage for all PBS-PT techniques.

A Decision Support Tool to Optimize Selection of Head and Neck Cancer Patients for Proton Therapy

Simple Summary

A decision support tool was developed to select head and neck cancer patients for proton therapy. The tool uses delineation data to predict expected toxicity risk reduction with proton therapy and can be used before a treatment plan is created. The positive predictive value of the tool is >90%. This tool significantly reduces delays in commencing treatment and avoid redundant photon vs. proton treatment plan comparison.

Abstract

Selection of head and neck cancer (HNC) patients for proton therapy (PT) using plan comparison (VMAT vs. IMPT) for each patient is labor-intensive. Our aim was to develop a decision support tool to identify patients with high probability to qualify for PT, at a very early stage (immediately after delineation) to avoid delay in treatment initiation. A total of 151 HNC patients were included, of which 106 (70%) patients qualified for PT. Linear regression models for individual OARs were created to predict the D_mean to the OARs for VMAT and IMPT plans. The predictors were OAR volume percentages overlapping with target volumes. Then, actual and predicted plan comparison decisions were compared. Actual and predicted OAR D_mean (VMAT R² = 0.953, IMPT R² = 0.975) and NTCP values (VMAT R² = 0.986, IMPT R² = 0.992) were highly correlated. The sensitivity, specificity, PPV and NPV of the decision support tool were 64%, 87%, 92% and 51%, respectively. The expected toxicity reduction with IMPT can be predicted using only the delineation data. The probability of qualifying for PT is >90% when the tool indicates a positive outcome for PT. This tool will contribute significantly to a more effective selection of HNC patients for PT at a much earlier stage, reducing treatment delay.

Fast, automated knowledge-based treatment planning for selecting patients for proton therapy based on normal tissue complication probabilities

Purpose

Selecting patients who will benefit from proton therapy is laborious and subjective. We demonstrate a novel automated solution for creating high-quality knowledge-based plans (KBPs) using proton and photon beams to identify patients for proton treatment based on their normal tissue complication probabilities (NTCP).

Methods and Materials

Two previously validated RapidPlan PT models for locally advanced head and neck cancer were used in combination with scripting to automatically create proton and photon KBPs for 72 patients with recent oropharynx cancer. NTCPs were calculated for each patient based on the KBPs, and patient selection was simulated according to the current Dutch national protocol.

Results

The photon/proton KBP exhibited good correlation between predicted and achieved organ-at-risk mean doses, with a ≤5 Gy difference in 208/196 out of 215 structures relevant for the head and neck cancer NTCP model. The proton KBPs yielded on average 7.1/6.1/7.6 Gy lower dose to salivary/swallowing structures/oral cavity than the photon KBPs. This reduced average grade 2/3 dysphagia and xerostomia by 7.1/3.3 and 5.5/2.0 percentage points, resulting in 16 of 72 patients (22%) being indicated for proton treatment. The entire automated process took <30 minutes per patient.

Conclusions

Automated support for decision making using KBP is feasible and fast. The planning solution has potential to speed up the planning and patient-selection process significantly without major compromises to the plan quality.

Prediction of liver Dmean for proton beam therapy using deep learning and contour-based data augmentation

The prediction of liver Dmean with 3-dimensional radiation treatment planning (3DRTP) is time consuming in the selection of proton beam therapy (PBT), and deep learning prediction generally requires large and tumor-specific databases. We developed a simple dose prediction tool (SDP) using deep learning and a novel contour-based data augmentation (CDA) approach and assessed its usability. We trained the SDP to predict the liver Dmean immediately. Five and two computed tomography (CT) data sets of actual patients with liver cancer were used for the training and validation. Data augmentation was performed by artificially embedding 199 contours of virtual clinical target volume (CTV) into CT images for each patient. The data sets of the CTVs and OARs are labeled with liver Dmean for six different treatment plans using two-dimensional calculations assuming all tissue densities as 1.0. The test of the validated model was performed using 10 unlabeled CT data sets of actual patients. Contouring only of the liver and CTV was required as input. The mean relative error (MRE), the mean percentage error (MPE) and regression coefficient between the planned and predicted Dmean was 0.1637, 6.6%, and 0.9455, respectively. The mean time required for the inference of liver Dmean of the six different treatment plans for a patient was 4.47±0.13 seconds. We conclude that the SDP is cost-effective and usable for gross estimation of liver Dmean in the clinic although the accuracy should be improved further if we need the accuracy of liver Dmean to be compatible with 3DRTP.

Head and neck proton therapy in France: A missed opportunity or a challenge in front of us?

Following major advances of the best of photon-techniques such as intensity-modulated radiotherapy (IMRT), stereotactic body radiotherapy (SBRT) and, to arrive soon, magnetic resonance (MR)-linac radiotherapy, there are still substantial opportunities in the treatment of head and neck cancers to further reduce the toxicity burden. Proton therapy represents another attractive option in this high-quality and highly competitive precision radiotherapy landscape. Proton therapy holds promises to reduce toxicities and to escalate the dose in radioresistant cases or cases where dose distribution is not satisfactory with photons. However, the selection of patients for proton therapy needs to be done using evidence-based medicine to build arguments in favor of personalized precision radiation therapy. Referral to proton therapy versus IMRT or SBRT should be registered (ProtonShare® platform) and envisioned in a formalized clinical research perspective through randomized trials. The use of an enrichment process using a model-based approach should be done to only randomize patients doomed to benefit from proton. To tackle such great opportunities, the French proton therapy challenge is to collaborate at the national and international levels, and to demonstrate that the extra-costs of treatment are worth clinically and economically in the short, mid, and long-term. In parallel to the clinical developments, there are still preclinical issues to be tackled (e.g., proton FLASH, mini-beams, combination with immunotherapy), for which the French Radiotransnet network offers a unique platform. The current article provides a personal view of the challenges and opportunities with a focus on clinical research and randomized trial requirements as well as the needs for strong collaborations at the national and international levels for PT in squamous cell carcinomas of the head and neck to date.