Author: Jacob S. Buatti, Mu-Han Lin, Dominic Moon, David D.M. Parsons, David Sher, Justin D. Visak, Hui Ju Wang 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, 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, Dallas, TX 🌍
Purpose: Most current adaptive treatment planning systems (TPS) natively utilize static planning goals from the reference plan for online adaptive re-optimization. In complex head-and-neck cancer (HNC), this is undesirable due to high geometric complexity and patient inter-fractional changes. The Ethos2.0 TPS optimization can be aided by a robust knowledge-based planning (KBP) model, which allows for dynamically online optimized plans by recalculating DVH estimates based on daily anatomical changes. Herein, we evaluate the strength of the KBP model in an online setting, comparing adaptive HNC plan quality using population-based goals driven with and without a KBP model.
Methods: Five HNC patients who received simultaneous boost treatments to 70-52Gy in 33 to 35 driven with prior Ethos treatment sessions that demonstrated significant radiotherapy response were selected. Two new treatment plans were generated in Ethos2.0 using identical population-based objectives and robust priority ordering with and without a KBP model. Prior online cone-beam CT and physician-approved target contours were extracted for online session emulation. Each reference plan strategy was emulated and dose to high-impact OAR were recorded. Statistical significance was assessed using paired student t-test (alpha=0.05).
Results: KBP-enabled optimization generally improved OAR sparing while maintaining target coverage online via dynamic optimization. Due to small sample size, no statistical significance was observed in OAR sparing. For example, the esophagus mean dose was further spared by 7.13±7.18Gy, 2.10±3.01Gy for the submandibular glands and 1.75±2.96Gy for the parotid glands. However, the KBP model’s known maximum dose limitation is evident in the mandible maximum dose which was moderately spared by 0.62±0.66Gy on average.
Conclusion: Dynamic optimization through KBP-enabled optimization in Ethos2.0 has shown potential in improving plan quality for head and neck cancer treatments. The integration of a KBP model can decrease the online planning burden, making the process more efficient and potentially increasing the use of ART.