Purpose: To evaluate the impact of daily online adaptation on normal tissue complication probability (NTCP) in the context of head and neck intensity modulated proton therapy (IMPT).
Methods: IMPT treatment plans are created without setup margins for 10 retrospective head and neck patients with daily CBCTs (median=33). The plans are adapted at each fraction using an in-house developed online adaptation (OA) workflow based on GPU accelerated Monte Carlo dose calculation and a deep learning-based CBCT scatter correction algorithm. OA is performed by re-adjusting the intensity of a subset of highly weighted spots without changing their energy or position. For comparison, a second plan is created for each patient using robust optimization (RO) with a 3 mm set-up uncertainty on which no adaptation is performed during delivery. For both scenarios, dose accumulation is performed using deformable image registration between each daily CBCT and the planning CT. NTCP models for xerostomia, aspiration, dysphagia, oral mucositis, liquid and solid swallowing are applied to the accumulated dose distributions in order to quantify the toxicity risk associated with each treatment technique.
Results: Both methods achieved similar target coverage over the course of treatment. NTCP calculated on the cumulative dose distributions showed a reduction in toxicity risk for aspiration, dysphagia, liquid and solid swallowing in all patients using OA over RO. The most noticeable differences were for aspiration, with median NTCP values of 28.4% and 17.1% for RO and OA respectively. Xerostomia and oral mucositis NTCP were also reduced over the whole patient cohort using OA over RO, with mean reductions of 4.7 and 8.3 percentage point respectively.
Conclusion: This work demonstrates the potential of online adaptive proton therapy to expand the therapeutic window of head and neck IMPT by reducing the risk of long-term side effects while ensuring adequate target coverage.