Robustness of AI Treatment Planning to Account for Inter-Fractional Changes in Targets and OARs
T Gray1*, W Wang2, C Liu3, D LaHurd1, E Balagamwala1, A Vassil1, J Wu2, P Xia2, (1) Cleveland Clinic, (2) Duke University, Durham, NC
Presentations
PO-GePV-T-377 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: To evaluate robustness of an AI model developed from a different institution to create adaptive plans for our pancreas patients to account for inter-fractional changes in the targets and organs-at-risk during a course of radiation therapy for pancreatic cancer.
Methods: Fifteen patient CTs, including planning and daily CTs, were contoured using an image analysis software. Major organs-at-risk contoured include the duodenum, stomach, small bowel and colon, along with a GTV and PTV. Treatment plans for each patient CT were planned using an artificial intelligence model developed for pancreas patients. These fifteen patient CTs were not used for the model development. Prescription dose was 33 Gy and 25 Gy to the GTV and PTV, respectively. DVHs were generated and evaluated for each GTV, PTV and each major organ-at-risk. Dosimetric constraints for each organ-at-risk include dose at 0.03 cc < 33 Gy, 0.5 cc < 30 Gy and 5 cc < 25 Gy. Target volumes were evaluated for dose at 95% volume and percent volume at 95% dose.
Results: Overall, the AI model was successful to create adaptive plans with adequate dose coverage to all targets. Most OAR constraints were also met. Average PTV percent volume at 95% dose for the AI model was 96.29 ± 5.20%. Average dose at 95% volume PTVs was 32.21 ± 0.68 Gy.
Conclusion: This study demonstrates that the AI model developed from another institution is robust and can be applied to our patient images for daily adaptive planning.
ePosters
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email