Author: Hania A. Al-Hallaq, Duncan Henry Bohannon, Chih-Wei Chang, Anees H. Dhabaan, Vishal Dhere, H Scott McGinnis, Pretesh Patel, Sagar Patel, Keyur Shah, Xiaofeng Yang, Jun Zhou 👨🔬
Affiliation: Emory University, Department of Radiation Oncology and Winship Cancer Institute, Emory University 🌍
Purpose: Two-fraction proton SABR is an attractive alternative to brachytherapy for localized prostate cancer. However, potential interfractional anatomical changes necessitate online adaptation, especially for treatments with reduced margins. This study evaluates a knowledge-based online adaptive planning strategy using reduced setup uncertainties (2 mm) and Gaussian process regression (GPR)-predicted geometric bounds to account for prostate interfraction motion.
Methods: GPR models were trained and validated using leave-one-out cross-validation on interfraction motion data from 42 five-fraction and 43 28-fraction proton prostate patients. Based on GPR predictions, prostate is shifted to anterior-superior and posterior-inferior positions from their nominal position in the planning CT, and three knowledge-based (KB) plans (KB-Nominal, KB-AS, KB-PI) were generated for three patients. All plans were planned to 26 Gy (13Gyx2) and normalized to CTV V100=98% with two lateral and two anterior beams, using ±2mm setup and ±3.5% range uncertainty. Clinical plans with 5mm (3mm posterior) setup uncertainties served as references. Plans were evaluated on pre-treatment corrected CBCT images under nominal and +3.5% density uncertainty. Metrics analyzed included CTV D98, CTV V100, and OAR doses (bladder V20.8Gy, rectum V17.6Gy, and bladder neck V100). Paddick's conformity index (pCI) and R50 were calculated.
Results: Compared to clinical plans, KB plans demonstrated improved plan quality with a higher conformity index (pCI = 0.76 vs. 0.69) and a sharper dose gradient (R50 = 3.99 vs. 4.41). KB plans resulted in a bladder V20.8Gy reduction of 26% (mean 5.3% vs. 7.2%) and rectum V17.6Gy reduction of 17% (mean 2.7% vs. 3.3%). The bladder neck V100 showed a slight improvement (61.0% vs. 65.2%). CTV V100 for KB plans was 96.1% compared to 94.9% for clinical plans.
Conclusion: A knowledge-based adaptive strategy with reduced margins maintains robust target coverage while substantially improving OAR sparing. This clinically feasible approach supports the implementation of two-fraction proton SABR for prostate cancer.