Classification of the source of treatment deviation in proton therapy using prompt‐gamma imaging information

• DOI: 10.1002/mp.14393
• Publication Date: 2020-07-17T19:19:02Z
• Authors: Chirasak Khamfong...; Jonathan Berthold; Guillaume Janssens; Johannes Petzoldt; Julien Smeets; Guntram Pausch; Christian Richter
• Keywords: Pencil-beam scanning

Purpose Prompt‐gamma imaging (PGI)‐based range verification has been successfully implemented in clinical proton therapy recently and its sensitivity to detect treatment deviations is currently investigated. The cause of can be multiple — for example, computed tomography (CT)‐based prediction, patient setup, anatomical changes. Hence, it would beneficial, if PGI‐based not only a deviation but also able directly identify most probable source. Here, we propose heuristically derived decision tree approach automatically classify the sources pencil‐beam scanning (PBS) treatments head neck tumors based on information obtained with PGI slit camera. Materials methods model was iteratively generated training dataset different complexities, an anthropomorphic phantom CT data (planning control CTs) from five patients. Different prediction errors, setup changes relevant nonrelevant were introduced or CTs, summing up total 98 scenarios. Independent validation performed another scenarios, solely seven PBS plans nominal scenario. For all spots investigated field, profiles simulated using dedicated analytical camera as well error From comparison spot‐wise shift after spot aggregation kernel 7 mm sigma determined each heuristic includes prefiltering suitable verification. validation, accuracy, sensitivity, specificity determined. Results A five‐step consecutive filter developed preselect spots. On average, 25% (1044 spots) remained input classification model. parameters: coefficient determination (R 2 ), slope intercept linear regression between PGI‐derived shifts respectively predicted ranges spots, average standard shifts. With this approach, 94 scenarios could classified correctly (accuracy 96%). 100% 86% reached. Conclusions In simulation study demonstrated that source identified noiseless tumor high specificity. application, refinement, evaluation measured will next step show feasibility classification.