Knowledge-based plan library-guided adaptive proton stereotactic ablative radiotherapy (SABR) for localized prostate cancer.
Ultrahypofractionated proton stereotactic ablative radiotherapy (SABR) is an emerging treatment for localized prostate cancer (PCa), with efforts ongoing to further condense treatment regimens to fewer than five fractions. However, proton SABR is highly susceptible to interfractional anatomical variations due to its steep dose gradients, requiring adaptive strategies to ensure robust clinical target volume (CTV) coverage while minimizing dose to organs at risk (OARs). Traditional margin-based approaches introduce unnecessary OAR exposure, while online re-optimization methods can be computationally expensive and resources-intensive.
This study aims to develop and evaluate a knowledge-based (KB) adaptive proton SABR workflow that accounts for prostate interfraction motion and density uncertainty (DU) by selecting the most clinically optimal plan from a set of pre-generated KB plans.
We retrospectively analyzed 42 prostate cancer patients treated with five-fraction proton SABR and 45 treated with 28-fraction proton therapy, using cone-beam CT (CBCT) imaging to evaluate interfraction motion and anatomical variations. Gaussian process regression (GPR) models were trained on these datasets to predict patient-specific prostate motion in the anterior-posterior (AP) and superior-inferior (SI) directions. Three KB treatment plans were generated per patient: KB-Nominal, KB-AS (Anterior-Superior), and KB-PI (Posterior-Inferior), all with 2 mm isotropic setup uncertainty, compared to a clinical plan with 5 mm (3 mm posterior) setup margins. The KB framework was tested on 10 randomly selected patients from the SABR cohort to evaluate plan quality and selection performance. Plans were evaluated using Monte Carlo (MC) dose calculations under nominal and ±3.5% DU conditions. Plan quality was assessed using ProKnow scoring, incorporating CTV coverage, dose conformity (Paddick Conformity Index and D2cm), and OAR doses (bladder, rectum, and bladder neck constraints). The optimal plan per fraction was selected based on the highest weighted-average ProKnow score across DU scenarios.
Across all DU conditions, KB plans reduced bladder and bladder neck dose compared to the clinical plan, while maintaining robust CTV coverage (D98 ≥ prescription dose). Compared to the clinical plan, KB-AS and KB-PI reduced bladder V20.8Gy by 26% (5.3% vs. 7.2%) and rectum V17.6Gy by 17% (2.7% vs. 3.3%), with bladder neck V100%Rx demonstrating the largest reduction at ±3.5% DU. KB plan selection was stable across DU variations, with KB plans consistently achieving higher ProKnow scores than the clinical plan. Benchmarking against Online Adaptive plans confirmed comparable plan quality, further validating the clinical robustness of the KB framework.
This study establishes the feasibility of a KB plan-driven adaptive proton SABR workflow for prostate cancer. By pre-generating motion-informed treatment plans and selecting the most optimal plan using ProKnow scoring, this framework ensures robust target coverage while substantially improving bladder and bladder neck sparing.
This study aims to develop and evaluate a knowledge-based (KB) adaptive proton SABR workflow that accounts for prostate interfraction motion and density uncertainty (DU) by selecting the most clinically optimal plan from a set of pre-generated KB plans.
We retrospectively analyzed 42 prostate cancer patients treated with five-fraction proton SABR and 45 treated with 28-fraction proton therapy, using cone-beam CT (CBCT) imaging to evaluate interfraction motion and anatomical variations. Gaussian process regression (GPR) models were trained on these datasets to predict patient-specific prostate motion in the anterior-posterior (AP) and superior-inferior (SI) directions. Three KB treatment plans were generated per patient: KB-Nominal, KB-AS (Anterior-Superior), and KB-PI (Posterior-Inferior), all with 2 mm isotropic setup uncertainty, compared to a clinical plan with 5 mm (3 mm posterior) setup margins. The KB framework was tested on 10 randomly selected patients from the SABR cohort to evaluate plan quality and selection performance. Plans were evaluated using Monte Carlo (MC) dose calculations under nominal and ±3.5% DU conditions. Plan quality was assessed using ProKnow scoring, incorporating CTV coverage, dose conformity (Paddick Conformity Index and D2cm), and OAR doses (bladder, rectum, and bladder neck constraints). The optimal plan per fraction was selected based on the highest weighted-average ProKnow score across DU scenarios.
Across all DU conditions, KB plans reduced bladder and bladder neck dose compared to the clinical plan, while maintaining robust CTV coverage (D98 ≥ prescription dose). Compared to the clinical plan, KB-AS and KB-PI reduced bladder V20.8Gy by 26% (5.3% vs. 7.2%) and rectum V17.6Gy by 17% (2.7% vs. 3.3%), with bladder neck V100%Rx demonstrating the largest reduction at ±3.5% DU. KB plan selection was stable across DU variations, with KB plans consistently achieving higher ProKnow scores than the clinical plan. Benchmarking against Online Adaptive plans confirmed comparable plan quality, further validating the clinical robustness of the KB framework.
This study establishes the feasibility of a KB plan-driven adaptive proton SABR workflow for prostate cancer. By pre-generating motion-informed treatment plans and selecting the most optimal plan using ProKnow scoring, this framework ensures robust target coverage while substantially improving bladder and bladder neck sparing.
Authors
Shah Shah, Chang Chang, Bohannon Bohannon, McGinnis McGinnis, Zafar Zafar, Diamond Diamond, Dhere Dhere, Patel Patel, Dhabaan Dhabaan, Al-Hallaq Al-Hallaq, Yang Yang, Patel Patel, Zhou Zhou
View on Pubmed