AI-Based Online Adaptive Radiation Therapy for Lung Cancer Treatment
W Mao*, J Riess, J Kim, S Vance, P Parikh, H Li, B Zhao, B Movsas, I Chetty, A Kretzker, Henry Ford Health System, Detroit, MI
Presentations
PO-GePV-M-94 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: Retrospective studies were performed to evaluate the accuracy of Varian Ethos AI-based automatically mapped structures and dosimetric consequences of daily adaptive radiation therapy (ART) for lung cancer treatment.
Methods: Ten locally advanced non-small cell lung cancer patients (prescribed by 2Gyx30) with 297 fractions of treatment were selected for this retrospective study on an Ethos research emulator. All patient data were copied and simulated twice: AutomaticART – automatic contours were utilized without modification with the fully automatic workflow and SupervisedART – automatic contours were modified manually by physicians. All daily structure sets and doses were exported for dose accumulation and further analysis. Dosimetric results of scheduled plan, AutomaticART, and SupervisedART were compared as ratios of relative reference plan results and grouped as lower constraints for target coverage and upper constraint for sensitive structures and hot spots of targets.
Results: 290 (of 297) fractions were successfully analyzed. Comparing target volumes between AutomaticART and SupervisedART, Dice similarity coefficient was 0.93±0.05, mean contour distance was 1.5±1.2 mm, and Hausdorff distance was 4.0±2.3 mm. AutomaticART and SupervisedART significantly improved minimum doses of planning treatment volumes by 4.85±3.03 Gy and 4.46±8.99 Gy, respectively, with a maximum increase of 22.62 Gy for one patient. For analysis over daily results of 290 fractions of treatment, average target dose coverages were improved from 0.96±0.04 (scheduled plan) to 1.00±0.02 and 1.02±0.04 and average upper dose/volume constraints were lowered from 1.01±0.11 to 0.94±0.10 and 0.93±0.12 for AutomaticART and SupervisedART, respectively. For cumulative dose comparison over 10 patients, target dose coverages were improved from 0.98±0.02 (scheduled plan) to 1.01±0.03 and 1.01±0.02, and upper dose/volume constraints were lowered from 0.99±0.03 to 0.96±0.02 and 0.96±0.02 for AutomaticART and SupervisedART, respectively.
Conclusion: Accuracy of Ethos AI-based automatic contouring is generally acceptable. The online adaptive radiation therapy improves target coverage and spares sensitive structures.
Funding Support, Disclosures, and Conflict of Interest: Portion of this project has been supported by a Varian Research grant.
ePosters
Keywords
Taxonomy
TH- External Beam- Photons: adaptive therapy
Contact Email