Automated Clinical Target Volume (CTV) Delineation Using Deep 4D Neural Networks with Enhanced OAR Sparing in Radiation Therapy of Non-Small Cell Lung Cancer (NSCLC)
Y Xie1*, K Kang2, Y Wang3, M Khandekar4, H Willers5, F Keane6, T Bortfeld7, (1) Massachusetts General Hospital, Boston, MA, (2) Independent Researcher, ,,(3) Massachusetts General Hospital, Boston, MA, (4) Massachusetts General Hospital, ,,(5) Massachusetts General Hospital, Boston, MA, (6) Massachusetts General Hospital, ,,(7) Massachusetts General Hospital, Boston, MA
Presentations
PO-GePV-M-209 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: Patient-specific CTV delineation can be the most difficult step in the process of defining the treatment volumes; and it accounts for the high workload for most radiation oncology departments worldwide. In this study, we demonstrated that our automated CTV delineation network can learn to avoid organ-at-risk (OAR) by including OAR information to the input of the network.
Methods: Different from our old model which takes patients’ CT and their corresponding iGTV, our new network adds the esophagus and the heart as inputs. The iGTVs were contoured by physicians from the 4D CT simulation scans. From 84 patients with locally advanced NSCLC treated with standard chemotherapy and radiation, 60 datasets were randomly selected to train the network, and the rest 24 were used for testing. Then, we compared the predicted CTVs with the ground truth, physician’s contours, by visual inspection as well as the similarity matrices to the ground truth and the overlap volumes with the OARs.
Results: All CTV predictions visually followed closely with the ground truth. Compared with our old ML-based model, the new model kept the similarity matrices with the ground truth, but with significantly reduced overlap volumes: from 0.4±0.3 to 0.02±0.05 cm3 for the esophagus and 1.2±0.6 to 0.7±0.4 cm3 for the heart.
Conclusion: The automated CTV delineation was utilized for NSCLC patients using the deep learningmethod based on CT images and manually contoured iGTV, the esophagus, and the heart contours. Compared with previous model, the new model demonstrates a significant improvement on the OAR sparing while keeping the overall performance on the intelligent volumetric expansions.
Keywords
Segmentation, Lung, Treatment Planning
Taxonomy
Contact Email