Predicting the benefits of proton therapy in lung cancer patients from patient-specific anatomic and pathologic geometry

Abstract

The magnitude of OAR dose sparing feasible with radiation therapy depends upon the patient-specific arrangement of the target volumes in relation to the OARs. When comparing proton beam therapy (PBT) and photon therapy, the evaluation of PBT benefits is currently achieved via resource-intensive treatment planning, which requires specialized software and clinical expertise typically only available at PBT centers. The purpose of this study is to train and validate knowledge-based plan prediction (KPP) models to estimate these benefits for lung cancer patients, requiring fewer resources and no PBT expertise. Our study used a cohort of locally advanced NSCLC patients, enrolled in a trial that randomized treatment between PBT (36 pts) and IMRT (68 pts). KPP models learn geometric patterns from a database of pre-existing treatment plans, and can then predict the feasible OAR DVHs of a new patient. By training independent KPP models for both PBT and IMRT, the feasible benefits of PBT can be predicted for the new patient. Our KPP methodology split each OAR into subvolumes that characterize the dose falloff and coplanar beams. Models were validated by 5-fold cross-validation, by comparing the predicted OAR DVHs to those of the actual treatment plan. DVHs were predicted for the lung (combined lungs excluding GTV), heart and esophagus. For each predicted and planned DVH, the equivalent uniform dose (EUD) and normal tissue complication probability (NTCP) were computed. The table shows the root-mean-square error of the EUD and NTCP predictions, for both the PBT and IMRT models. It also shows the estimated uncertainty in the predicted benefit. Model accuracy depends upon the OAR type and treatment modality, but no significant signs of bias were observed. Pearson’s correlation coefficient was 84% – 98% for PBT and 81% – 96% for IMRT, demonstrating that the KPP models successfully captured the interpatient variation. We developed KPP models for lung cancer patients treated with PBT and an alternative treatment modality (IMRT). By comparing the two predicted plans for a new patient, the patient-specific benefits of PBT can quickly be estimated with high accuracy. This methodology could enable better-informed referral decisions, improving the cost-effectiveness of this expensive therapy and elevating PBT in the era of precision medicine. It could also estimate the expected benefits at an earlier stage in the clinical workflow, or provide a high-throughput patient pre-selection for model-based trials aiming to measure the higher efficacy of PBT.Abstract 3592Prediction Error in EUD [Gy]Prediction Error in NTCP [%]PBTIMRTBenefitsPBTIMRTBenefitsLung2.02.23.02.94.45.3Heart1.93.54.02.14.24.7Esophagus4.35.16.76.87.810.3 Open table in a new tab