A Decision-Making Method for Photon/Proton Selection for Nasopharyngeal Cancer Based on Dose Prediction and NTCP

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 andexpertise. 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 andautomatically. 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). 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). 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 planningexpertise.

Simple SummaryThe role of proton therapy as a radiation treatment option for locally advanced sinonasal cancer patients has increased in the last years, showing promising results in terms of clinical outcomes. Definition of strategies to identify patients who would benefit the most from proton therapy in terms of reduced toxicity is highly desirable, due to limited availability and higher costs of this treatment option. The novelty of our in silico study relies on assessing the suitability of a mixed dose volume histograms and normal tissue complication probability model-based approach, aiming at providing, to the scientific community, a patient selection criteria possibly leading the therapeutic choice between proton therapy and advanced photon techniques.(1) Background: In this work, we aim to provide selection criteria based on normal tissue complication probability (NTCP) models and additional explanatory dose-volume histogram parameters suitable for identifying locally advanced sinonasal cancer patients with orbital invasion benefitting from proton therapy. (2) Methods: Twenty-two patients were enrolled, and two advanced radiation techniques were compared: intensity modulated proton therapy (IMPT) and photon volumetric modulated arc therapy (VMAT). Plans were optimized with a simultaneous integrated boost modality: 70 and 56 Gy(RBE) in 35 fractions were prescribed to the high risk/low risk CTV. Several endpoints were investigated, classified for their severity and used as discriminating paradigms. In particular, when NTCP models were already available, a first selection criterion based on the delta-NTCP was adopted. Additionally, an overall analysis in terms of DVH parameters was performed. Furthermore, a second selection criterion based on a weighted sum of the NTCP and DVH was adopted. (3) Results: Four patients out of 22 (18.2%) were suitable for IMPT due to NTCP > 3% for at least one severe toxicity, 4 (18.2%) due to NTCP > 20% for at least three concurrent intermediate toxicities and 16 (72.7%) due to the mixed sum of NTCP and DVH criterion. Since, for some cases, both criteria were contemporary fulfilled, globally 17/22 patients (77.3%) would benefit from IMPT. (4) Conclusions: For this rare clinical scenario, the use of a strategy including DVH parameters and NTCPs when comparing VMAT and IMPT is feasible. We showed that patients affected by sinonasal cancer could profit from IMPT compared to VMAT in terms of optical and central nervous system organs at risk sparing.

Background & purpose –Late radiation-associated dysphagia after head and neck cancer (HNC) significantly impacts patient’s health and quality of life. Conventional normal tissue complication probability (NTCP) models use discrete dose parameters to predict toxicity risk but fail to fully capture the complexity of this side effect. Deep learning (DL) offers potential improvements by incorporating 3D dose data for all anatomical structures involved in swallowing. This study aims to enhance dysphagia prediction with 3D DL NTCP models compared to conventional NTCP models.Materials & methods –A multi-institutional cohort of 1484 HNC patients was used to train and validate a 3D DL model (Residual Network) incorporating 3D dose distributions, organ-at-risk segmentations, and CT scans, with or without patient- or treatment-related data. Predictions of grade ≥2 dysphagia (CTCAEv4) at six months post-treatment were evaluated using area under the curve (AUC) and calibration curves. Results were compared to a conventional NTCP model based on pre-treatment dysphagia, tumour location, and mean dose to swallowing organs. Attention maps highlighting regions of interest for individual patients were assessed.Results –DL models outperformed the conventional NTCP model in both the independent test set (AUC=0.80–0.84 versus 0.76) and external test set (AUC=0.73–0.74 versus 0.63) in AUC and calibration. Attention maps showed a focus on the oral cavity and superior pharyngeal constrictor muscle.Conclusion –DL NTCP models performed better than the conventional NTCP model, suggesting the benefit of using 3D-input over the conventional discrete dose parameters. Attention maps highlighted relevant regions linked to dysphagia, supporting the utility of DL for improved predictions.

The release of a homogeneous high dose to the tumour region has been one of the cornerstones of radiotherapy (RT) treatment since its early days. According to the organ type and to the cancer histology, different doses are required in order to inactivate malignant cells, thus stopping proliferation. However, radiation-induced cell killing is a stochastic process. Tumour control probability (TCP) models have been developed in order to assign a success rate to a given RT treatment. At the same time, there is the need to keep the risks of normal tissue toxicity at an acceptable level. Normal tissue complication probability (NTCP) models provide a means of doing this. Traditionally, TCP and NTCP models combine clinical outcomes with dosimetric information in terms of dose-volume histograms (DVH). Model parameters are derived by mathematical fits to clinical observations and are subsequently used to estimate the risk of tumour relapse or toxicity. In both types of models, all of the patient dosimetric information is condensed into the DVHs, which represents a potential limitation on their descriptive and predictive power. This choice, related to historical and practical reasons, does not allow the full complexity of the 3D dose distribution in the patient to be taken into account. Neglecting these aspects might be relevant in a modern RT setting, which often includes the presence of high dose gradient regions. This has motivated research on ‘advanced’ TCP and NTCP models, able to tackle the problem by looking at a different scale, e.g. in tumour sub-regions or at the single voxel level. This is relevant not only from the purely dosimetric point of view. Increasing evidence is reported on the heterogeneity of cancer tissues, suggesting that non-uniform dose distributions could result in improved survival, for instance if targeted to take into account sub volumes with high clonogen density or hypoxic radioresistant regions. Similarly, radiation-induced side effects are part of a complex biological response, which depends not only on cell killing, but also on the inflammatory response and in some cases on the interplay among different organs. Obviously, conventional NTCP models cannot describe this scenario, and the development of more advanced mathematical tools is needed. This review will be focused on the discussion of recent studies showing possible directions for moving the field of TCP and NTCP modelling forward. Without diminishing the role and usefulness of available models, the aim is to shed light on the benefits that might be achieved by ‘enhanced’ modelling. This could represent an important step in the gradual transition of radiation therapy towards a form of precision medicine.

Integrating photon with proton dose-response data alters a pulmonary complication model-based patient selection in esophageal cancer.

Toward a model-based patient selection strategy for proton therapy: External validation of photon-derived normal tissue complication probability models in a head and neck proton therapy cohort

Objective.Proton therapy can reduce the dose to organs-at-risk (OAR) compared with photon therapy. However, proton dose distributions are more sensitive to respiratory motion and uncertainties. These are probabilistic processes and should be analyzed accordingly regarding impact on dose. This study shows consistent robust planning across free-breathing (FB) and breath-hold (BH) stereotactic body radiotherapy treatments. Further, a method for probabilistic evaluation of proton and photon plans using displacement vector field (DVF) editing to investigate the effects of respiratory motion is presented.Approach.Robust optimization parameters were calculated and the evaluation method was demonstrated for one patient. The goal was to deliver a clinical target volume (CTV)D50%⩾54 Gy in three fractions. A BH CT and a 4DCT were used for planning and additionally deformed with DVF editing to simulate motion and BH variation. Errors in both positioning and proton therapy stopping power ratio were simulated using random sampling.Main results.Both the FB and BH proton plans had a probability of 100% of achieving CTVD50%⩾54 Gy while the two photon plans had probabilities of 97.0% in BH and 94.7% in FB. Doses to the central OARs were lower for the proton plans compared with the photon plans. Spread in CTV dose volume histogram curves was larger for the proton plans compared with the photon plans. Doses for the proton plans were higher compared with the photon plans.Significance.Robust planning, DVF editing and probabilistic evaluation for the comparison of BH and FB proton and photon therapy were established. Spread in CTV dose metrics were larger for proton than for photon therapy but with benefits in dose to OARs for proton therapy. In future studies more patients with different tumor sizes, locations and respiratory motion amplitudes should be investigated.

This work aims to evaluate the predictive strength of the relative seriality, parallel and Lyman–Kutcher–Burman (LKB) normal tissue complication probability (NTCP) models regarding the incidence of radiation pneumonitis (RP), in a group of patients following lung cancer radiotherapy and also to examine their correlation with pulmonary function tests (PFTs). The study was based on 47 patients who received radiation therapy for stage III non-small-cell lung cancer. For each patient, lung dose volume histograms (DVHs) and the clinical treatment outcome were available. Clinical symptoms, radiological findings and pulmonary function tests incorporated in a post-treatment follow-up period of 18 months were used to assess the manifestation of radiation induced complications. Thirteen of the 47 patients were scored as having radiation induced pneumonitis, with RTOG criteria grade ⩾ 3 and 28 of the 47 with RTOG criteria grade ⩾ 2. Using this material, different methods of estimating the likelihood of radiation effects were evaluated, by analysing patient data based on their full dose distributions and associating the calculated complication rates with the clinical follow-up records. Lungs were evaluated as a paired organ as well as individual lungs. Of the NTCP models examined in the overall group considering the dose distribution in the ipsilateral lung, all models were able to predict radiation induced pneumonitis only in the case of grade 2 radiation pneumonitis score, with the LKB model giving the best results (χ2-test: probability of agreement between the observed and predicted results Pχ(χ2) = 0.524 using the 0.05 significance level). The NTCP modelling considering lungs as a paired organ did not give statistically acceptable results. In the case of lung cancer radiotherapy, the application of different published radiobiological parameters alters the NTCP results, but not excessively as in the case of breast cancer radiotherapy. In this relatively small group of lung cancer patients, no positive statistical correlation could be established between the incidence of radiation pneumonitis as estimated by NTCP models and the pulmonary function test evaluation. However, the use of PFTs as markers or predictors for the incidence or severity of radiation induced pneumonitis must be investigated further.

Objective. This study explores the use of neural networks (NNs) as surrogate models for Monte-Carlo (MC) simulations in predicting the dose-averaged linear energy transfer (LET d ) of protons in proton-beam therapy based on the planned dose distribution and patient anatomy in the form of computed tomography (CT) images. As LET d is associated with variability in the relative biological effectiveness (RBE) of protons, we also evaluate the implications of using NN predictions for normal tissue complication probability (NTCP) models within a variable-RBE context. Approach. The predictive performance of three-dimensional NN architectures was evaluated using five-fold cross-validation on a cohort of brain tumor patients (n = 151). The best-performing model was identified and externally validated on patients from a different center (n = 107). LET d predictions were compared to MC-simulated results in clinically relevant regions of interest. We assessed the impact on NTCP models by leveraging LET d predictions to derive RBE-weighted doses, using the Wedenberg RBE model. Main results. We found NNs based solely on the planned dose distribution, i.e. without additional usage of CT images, can approximate MC-based LET d distributions. Root mean squared errors (RMSE) for the median LET d within the brain, brainstem, CTV, chiasm, lacrimal glands (ipsilateral/contralateral) and optic nerves (ipsilateral/contralateral) were 0.36, 0.87, 0.31, 0.73, 0.68, 1.04, 0.69 and 1.24 keV µm−1, respectively. Although model predictions showed statistically significant differences from MC outputs, these did not result in substantial changes in NTCP predictions, with RMSEs of at most 3.2 percentage points. Significance. The ability of NNs to predict LET d based solely on planned dose distributions suggests a viable alternative to compute-intensive MC simulations in a variable-RBE setting. This is particularly useful in scenarios where MC simulation data are unavailable, facilitating resource-constrained proton therapy treatment planning, retrospective patient data analysis and further investigations on the variability of proton RBE.

Cross-modality applicability of rectal normal tissue complication probability models from photon- to proton-based radiotherapy

Multiple competing normal tissue complication probability (NTCP) models have been proposed for predicting symptomatic radiation-induced lung injury in human. In this paper we tested the efficacy of four common NTCP models applied quantitatively to sub-clinical X-ray computed tomography (CT)-density changes in the lung following radiotherapy. Radiotherapy planning datasets and follow-up chest CTs were obtained in eight patients treated for targets within the lung or hilar region. Image pixel-wise radiation dose exposure versus change in observable CT Hounsfield units was recorded for early (2–5 months) and late (6–9 months) time-points. Four NTCP models, Lyman, Logistic, Weibull and Poisson, were fit to the population data. The quality of fits was assessed by five statistical criteria. All four models fit the data significantly (p < 0.05) well at early, late and cumulative time points. The Lyman model fitted best for early effects while the Weibull Model fitted best for late effects. No significant difference was found between the fits of the models and with respect to parameters D50 and γ50. The D50 estimates were more robust than γ50 to image registration error. For analyzing population-based sub-clinical CT pixel intensity-based dose response, all four models performed well.

Optimal Allocation of Proton Therapy Slots in Combined Proton-Photon Radiation Therapy.

Calculation of the uncertainty in complication probability for various dose–response models, applied to the parotid gland

Dosimetric Evaluation of Incorporating Patient Geometric Variations Into Adaptive Plan Optimization Through Probabilistic Treatment Planning in Head and Neck Cancers

Dose–volume toxicity modeling for de-intensified chemo-radiation therapy for HPV-positive oropharynx cancer