Machine learning‐based automatic proton therapy planning: Impact of post‐processing and dose‐mimicking in plan robustness

• DOI: 10.1002/mp.16408
• Publication Date: 2023-04-08T09:34:00Z
• Authors: Elena Borderias‐V...; Margerie Huet Das...; Ana María Barragá...; Rasmus Helander; Mats Holmstrom; Xavier Geets; Edmond Sterpin
• Keywords: Robustness

Automated treatment planning strategies are being widely implemented in clinical routines to reduce inter-planner variability, speed up the optimization process, and improve plan quality. This study aims evaluate feasibility quality of intensity-modulated proton therapy (IMPT) plans generated with four different knowledge-based (KBP) pipelines fully integrated into a commercial system (TPS).A data set containing 60 oropharyngeal cancer patients was split 11 folds, each 47 for training, five validation, testing. A dose prediction model trained on resulting total models. Three were left out order assess if differences introduced between models significant. From voxel-based predictions, we analyze two steps that follow prediction: post-processing predicted mimicking (DM). We focused effect (PP) or no (NPP) combined DM algorithms optimization: one available TPS RayStation (RSM) simpler isodose-based (IBM). Using 55 test (five model), evaluated robustness by proposed KBP (PP-RSM, PP-IBM, NPP-RSM, NPP-IBM). After robust evaluation, dose-volume histogram (DVH) metrics nominal worst-case scenarios compared those manually plans.Nominal doses from showed promising results achieving comparable target coverage improved organs at risk (OARs) manual plans. However, too optimistic applied (i.e. important decrease organs) compromised Even though RSM seemed partially compensate lack PP plans, still 65% did not achieve expected levels. NPP-RSM best trade-off OAR sparing.PP crucial generate acceptable deliverable IMPT ML-predicted doses. Before implementation any pipeline, parameters predefined need be modified accordingly comprehensive feedback loop which final calculations is evaluated. With right choice parameters, have potential within clinically levels dosimetrists.