(Wednesday, 8/1/2018) 4:30 PM - 6:00 PM
Room: Room 207
Purpose: To provide an automated method for IMPT treatment planning based a machine learning model of the multicriteria Pareto surface.
Methods: A machine learning model was developed to model and predict the multicriteria Pareto surface for IMPT with large number of DVH based objectives (physical dose, robust scenarios, and biological dose). The machine learning model was implemented to guide the generation of the multicriteria Pareto plans during the optimization phase and to provide automated way of navigating the Pareto surface and selecting IMPT plans afterword. An initial set of Pareto plans was generated and used to train the machine learning model. The model was used to predict the Pareto surface and guide the generation of a well distributed points on the Pareto surface. After generation of the IMPT Pareto plans database and during the navigation of the plans, the constructed machine learning model was used to predict the DVH points of the Pareto surface for the treatment planar and automatically generate plans of potential interest. The machine learning model parameters were optimized to improve the root mean square error (RMSE). The model was used to predict various DVH objectives and corresponding physical plan variables (proton spots). The machine learning automated way of generating IMPT plans was compared against clinical plans for head & neck and brain tumor cases.
Results: The machine learning prediction accuracy of the Pareto surface was examined and the error (RMSE) in prediction was minimized.
Conclusion: The machine learning method for automated generation of IMPT plans can provide an efficient mapping of multicriteria Pareto surface and assist in selection of the treatment plan based on large number of DVH objectives for IMPT.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
