Dosimetric and Clinical Features Associated with Replan in Proton Head and Neck Therapy.

Purpose/Objective(s) Intensity modulated proton therapy has become increasingly popular in the treatment of head and neck (HN) cancer because of its high normal tissue sparing compared to photon therapy. However, proton plans are very sensitive to anatomy changes such as weight loss or tumor shrinkage that leads to a 25% replan rate at our center. This analysis aims to predict whether a HN patient will need a replan using various dosimetric and clinical features at the onset of treatment. Materials/Methods Data were gathered from 176 HN patients treated at our institution. The most common sites were the oropharynx (40 cases), lip/oral cavity (20 cases), and nasal cavity/sinuses (14 cases). Beam dose heterogeneity index (BHI) was defined as the ratio of the D0.1cc of beam dose to half of the prescription dose. BHI was calculated for every beam and the average BHI for each plan was calculated. Robustness scenarios included a combination of ± 3 mm setup shifts on 3 coordinate directions and ± 3.5% range uncertainty. Plan robustness was evaluated with several features gathered from the highest dose level CTV. These include the fraction of all scenarios where V100 > 95% (passing rate), mean V100, mean V100 deviation from the nominal plan, and mean max dose change from the nominal. 61% of patients were treated to bilateral neck with multi-dose level simultaneous integrated boost plans. A five-beam arrangement (two posterior, two anterior oblique beams, and an AP beam) were used to treat most bilateral HN patients and a 3-beam arrangement was used to treat most unilateral HN patients. Any portion of the target is typically covered by at least two beams. Results Sixty-three patients were replanned at least once in this subpopulation. Plans were normalized at V100 = 98% when possible for the highest dose level. The BHIs from each planner were significantly different from each other, ranging from 1.01 to 1.27 (P = 0.003). Planner re-plan rates (ranging from 11%-42%) significantly correlate with their BHIs, r = 0.747 (P = 0.03). BHI for re-planned patients was higher than non-replan patients:1.11 ± 0.09 vs. 1.06 ± 0.16 (P = 0.017). The only significant robustness feature was the mean max dose change:3.1 ± 1.3% vs. 2.5 ± 1.4% (P = 0.038). Conclusion We have identified several features that correlate strongly with HN replans. A fine-tuned planning protocol is being developed to reduce the BHI for all HN plans. Our future work is to build a deep-learning model based on our current features and others such as patients’ clinical tumor information, planning image, plan conformity, etc., to predict the probability of a replan before treatment.