Description
ObjectiveThe aim of this study was to automatically generate Intensity Modulated Proton Therapy (IMPT) plans for oropharyngeal patients and to test if plan quality was at least equal compared to clinically and ‘manually’ optimized robust IMPT plans.
Method
Machine Learning Optimization (MLO) planning involved training of a model using data of 73 oropharyngeal cancer patients (CT-scans, structures and dose distributions) to predict the dose distribution for new patients. A robust mimicking optimization algorithm using voxel-based mimicking and 21 perturbed scenarios was then used to generate a machine deliverable plan from the predicted dose distributions. Cross-validation was performed with 3x5 validation patients to tune prediction and mimicking settings. Plans were considered clinically acceptable when robust target coverage, conformity and normal tissue doses were within the following limits; clinical target volume D98 voxelwise minimum dose >94% (using multi-scenario dose evaluation strategy), conformity index decreased <10% and Normal Tissue Complication Probability (NTCP) (sum of grade-2 dysphagia and xerostomia) increased <2%, respectively.
Results
In 8/15 plans the MLO resulted in clinically acceptable plans. In these plans, the sum of NTCPs decreased on average with 0.37% (range: -2.3 - 1.8). The target conformity decreased more than 10% in 4/15 plans and the sum NTCP increased by more than 2% in 3/15 plans. MLO plans were generated including robustness evaluation in 67 +/- 8 minutes.
Conclusion
MLO with dose predictions and robust optimization automatically generated clinical acceptable robust IMPT plans for oropharyngeal cancer patients. Future work aims to increase the autoplanning success rate.
| Periode | 9-jun.-2021 |
|---|---|
| Evenementstitel | PTCOG 2021: Satelite Symposium IBA |
| Evenementstype | Conference |
| Locatie | Taipei, TaiwanToon op kaart |
| Mate van erkenning | International |