Author: Shahed Badiyan, Chien-Yi Liao, Mu-Han Lin, Dan Nguyen, Justin D. Visak, Hui Ju Wang, Brien Timothy Washington, Kenneth Westover, Yuanyuan Zhang 👨🔬
Affiliation: Medical Artificial Intelligence and Automation (MAIA) Lab & Department of Radiation Oncology, UT Southwestern Medical Center, Department of Radiation Oncology, UT Southwestern Medical Center, Dallas, TX, Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center, UT Southwestern Medical Center, Department of Radiation Oncology, UT Southwestern Medical Center 🌍
Purpose: Knowledge-based planning (KBP) plays a crucial role in improving treatment plans by leveraging previous clinical data to guide new cases. KBP is applied to the Ethos 2.0 Intelligent Optimization Engine (IOE) with its high-fidelity (HF) mode to assess its effectiveness in enhancing online treatment quality for lung SABR. The aim of this study is to evaluate the potential benefits of integrating KBP in online treatment workflows to help maximize plan quality and decrease sensitivity to anatomical changes.
Methods: Five adaptive lung SABR patients who received 40-50Gy in 5 fractions were collected and replanned with identical treatment geometry. Two plans were generated using Ethos 2.0 with HF mode enabled: one plan incorporated an internally trained lung KBP model (HF+KBP), while the other used HF mode only. A cone-beam-CT was extracted for the planning image in the online emulator. Using physician-approved contours, the workflow was emulated with Varian Ethos 2.0 emulator with both reference plans. The plans were then compared based on DVH metrics of targets, the Conformity Index (CI), the Gradient Index (GI), and organ-at-risk (OAR) dose assessment.
Results: Compared to HF only, the HF+KBP plans demonstrated superior plan quality, with significantly improved GI (4.57 ± 0.56 vs. 4.22 ± 0.44, p = 0.007) and CI (1.06 ± 0.59 vs. 1.04 ± 0.05, p = 0.025). For most OARs, HF+KBP achieved dose reductions, with the esophagus showing a statistically significant improvement (D0.03cc: 3579 cGy vs. 3506 cGy, p = 0.006). However, HF+KBP resulted in higher D2cm coverage and reduced PTV hotspot.
Conclusion: The integration of KBP with the Ethos 2.0 HF mode is compatible and demonstrates potential to enhance plan quality in lung SABR. Benefits may further increase with emerging personalized ultra-fractioned regimes. While improvements were achieved with KBP, slight conflict was observed between the KBP model and HF objectives.