The Effect of AI Contouring On Head and Neck Patient Treatment Plan Optimization
M Jones1*, P Mavroidis2, T Wagner1, S Stathakis1, (1)Mays Cancer Center - MD Anderson Cancer Center, San Antonio, TX, (2)University of North Carolina, Chapel Hill, NC
Presentations
PO-GePV-T-101 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To investigate the differences of treatment planning optimization using AI generated structures versus structures by experts.
Methods: For the feasibility of the study, ten (n=10) Head and Neck patients were included. The structure sets were initially segmented by an expert radiation oncology. A second set was created using a commercial AI software. For each patient two plans were created using the RayStation treatment planning system. Two full VMAT arcs using 6MV were used for each plan. Both plans had the same objectives and run for the same number of iterations. Our hypothesis is that there is no difference in the plan quality between the two plans of each patient. We evaluated the differences of PTV homogeneity and conformity and the differences of the dose objectives for each of the organs involved in the optimization.
Results: No statistical differences were observed in the metrics for target coverage. The conformity and homogeneity of the dose coverage was within 10% for all patients in the study. The largest differences were observed on cochleas due to the fact that the AI structures are more than 70% smaller. The AI segmented parotids were larger than the ones drawn by the expert but had no effect on the target coverage and plan quality.
Conclusion: Target coverage can be maintained with contours produced by an AI system. The differences in the structure volume do not affect plan quality. A larger number of patient plans in under way to further evaluate possible trends and establish guidelines.
Keywords
Not Applicable / None Entered.
Taxonomy
Contact Email