PROJECT SUMMARY Stereotactic body radiation therapy (SBRT) is a highly effective treatment for early-stage non-small cell lung cancer, but its accuracy can be compromised by multiple factors. There is an interval between simulation and the first day of treatment, the size and position of targets and organs at risk can shift over a course of treatment, and the thorax is in constant multidimensional motion. Adaptive radiation can improve the accuracy of SBRT, but implementing it within the workflow of a busy radiation oncology clinic currently requires re- simulation and re-planning, costing valuable departmental time and resources. Cone beam computed tomography (CBCT) scans are obtained daily prior to the delivery of each fraction, but their utility for adaptive radiation therapy has been limited by their image quality. Processing time also remains a significant barrier for real-time deep learning-based methodologies. The objective of our proposed research is therefore to develop, validate, and test in an early clinical trial the feasibility of using our two-part cone-beam computed tomography- based deep learning method for dose verification based on rapid and accurate generation of high quality synthetic CTs and multi-organ segmentation. In this project, we will pursue two Specific Aims: 1) to develop and refine CBCT-based synthetic CTs for CBCT quality improvement, and 2) to evaluate the clinical feasibility of our synthetic CT-based dose verification. The early clinical trial will prospectively enroll patients with early- stage non-small cell lung cancer receiving definitive SBRT. Validation of the feasibility of this method is a necessary intermediate step towards our longer-term goal of the implementation of real-time lung cancer adaptive radiation, which will allow for increased accuracy of higher dose to target volumes and lower doses to organs at risk, thereby improving local control and decreasing radiation-related risks and toxicities for patients with non-small cell lung cancer.

PROJECT NARRATIVE We propose to develop and validate a two-part deep learning method for dose verification in patients with early-stage non-small cell lung cancer receiving definitive stereotactic body radiation therapy. We will evaluate feasibility in a prospective early-stage clinical trial, a necessary intermediate step prior to the implementation of real-time daily adaptive radiation therapy. Our long-term goal is to use real-time adaptive radiation therapy to improve tumor control and decrease radiation-related toxicities for patients receiving radiation therapy for lung cancer.