Room: Exhibit Hall | Forum 2
Purpose: To create an improved dose response model that predicts lung ventilation change following radiation therapy in order to produce functional avoidance treatment plans that avoid high functioning regions.
Methods: The dose response model was built using twelve human subjects who underwent radiation therapy (RT) from an IRB-approved trial. For each 4DCT, a ventilation map was created by calculating the local expansion ratio (LER) of the end exhale and inhale phases, referred to as LER3D. The maximum Jacobian determinant to the minimum Jacobian determinant of the deformable image registrations from all phases was referred to as LER4D. A polynomial regression model was created using the LER4D data and dose distributions for each subject, and cross-validated with a leave-one-out method. Validation of the model was performed using thirteen human subjects from a different IRB-approved trial using common statistical operating characteristics.
Results: Cross-validation showed there was a significant increase in the gamma pass rate of the predicted post-RT to the actual post-RT ventilation maps when using the model built using LER4D (p=0.0013). However, the gamma pass rate of the ratio of the predicted post-RT to the actual post-RT ventilation maps was greater for LER4D compared to the LER3D model, but not significantly, using the second dataset (p=0.14). The positive predictive value (p=0.019) and true negative rate (p=0.0018) increased significantly, while the true positive rate decreased significantly (p=0.00004) using the LER4D model compared to the LER3D model. The accuracy of the LER4D model exhibited no significant change from the LER3D model (p=0.61).
Conclusion: A 4D generated Jacobian ventilation model was shown to provide significantly different results. For the subjects considered in this study, LER4D was more accurate at modeling lung tissue elasticity post-RT. Further changes to the model are required to significantly improve statistical operating characteristics and gamma pass rates.
Funding Support, Disclosures, and Conflict of Interest: Funded by NCI Grant 5R01CA166703-05