Applying voxel-based analysis to oropharyngeal cancer proton therapy patients: A correlation study on radiation-induced acute dysphagia.

BACKGROUND: Voxel-based analysis (VBA) is an analytic approach to evaluate correlations between local dose and the development of different toxicities. DVHs are used for toxicity prediction as well. Compared with DVH, no contours are required for VBA technique, and results tell specific voxels that may be related to the toxicity instead of the whole contoured area. The VBA has been used on different cancer sites and for different toxicities. Most of these studies included patients treated with photon; all published studies were based on planned dose, and VBA tools used were developed in-house. In our study, patient cohort was treated with proton, our VBA tool was developed based on RayStation, and doses fed to the VBA tool were delivered doses with constant and two variable RBE models.

PURPOSE: To develop a VBA tool with RayStation scripting to evaluate correlations between spatial dosimetry and acute dysphagia in oropharyngeal cancer patients treated with chemoradiotherapy.

METHODS: Forty-two oropharyngeal cancer patients treated with intensity-modulated proton therapy were included. All patients received bilateral neck irradiation and had no prior HN surgery or radiation. Clinician-graded acute dysphagia was recorded using CTCAE v5.0. Each patient's dose was estimated as a weighted sum of doses based on the planning CT and subsequent quality assurance CTs acquired during treatment for dosimetric effect evaluation related to anatomic changes. A constant and two-variable RBE model provided estimations of biological doses. Each cumulative dose was mapped into a synthetic CT (common coordinate system) that represents this patient cohort. The cohort was split into toxicity and non-toxicity groups according to their toxicity grade. Permutation tests with maximum statistic strategy were performed on mapped doses to find locations with significant dose differences between the two groups. A loosened p-value threshold (0.1) was selected for sensitive cluster identification to conservatively include more area in identified clusters. The non-dosimetric variables (age and sex) on toxicity were studied with the Mann-Whitney U test.

RESULTS: No sensitive cluster was detected with acute dysphagia grade ≥ 3 grouping criterion. For grade ≥ 2, one or two clusters, depending on RBE models (one within the oral cavity and the other at the anterior-inferior part of the larynx), were identified. For non-dosimetric variables, only a marginal correlation (p = 0.07) between age and acute dysphagia was detected with grade ≥ 2.

CONCLUSION: In the patient cohort evaluated in this study, our VBA tool highlighted oral cavity and larynx subregions that may be related to the acute dysphagia development during definitive radiation for oropharyngeal cancer. Optimization of treatment plans to reduce dose to these areas may allow for a reduction in acute dysphagia in future patients.