Room: Room 209
Purpose: IMPT offers great dosimetric advantages; however, manually defining the optimal set of constraints within a commercial optimizer can be quite challenging, and often involves a tedious trial and error procedure for complex cases such as bilateral head neck. The purpose of this work is to develop and validate site-specific models from available clinical cases for deriving IMPT treating plans efficiently for new patients.
Methods: IMPT plans of over 20 bilateral head neck patients treated with SIB prescriptions were created under a commercial planning system using multi-field optimization technique with standard 3-field beam arrangement. Then a statistical model was built from sets of geometrical (object shapes, volumes, sizes etc.,) and dosimetric parameters(DVHs) extracted from these plans. Those sets of parameters will be extracted from any new patient not in the model, and be compared with the learned model to automatically derive an estimate of the set of optimization weights. To quantitatively validate the model, leave-one-out technique was used to compare model-based plans against training plans in term of normal tissue doses. Robustness generalization of the model was also comprehensively evaluated.
Results: We demonstrated that the model-based plans compare favorably with manually created clinical plans in terms of doses to OARs. Particularly, with the same target coverage, there are around 5Gy mean dose reduction on parotid glands, around 3Gy reduction on oral cavity. For the leave-one-out tests, it showed that the learned model can achieve good robustness generalization. For high-risk CTV, the averaged V95% at worst-case scenario is 98%Â±1.6%; for low-risk CTV, this value becomes 99%Â±1.1%.
Conclusion: The study demonstrated that statistical model-based algorithm is capable of deriving clinically-acceptable IMPT plans, and the model also generalize well on robustness.
Funding Support, Disclosures, and Conflict of Interest: The research is supported Varian Medical Systems
Not Applicable / None Entered.