SUMMARY

For 4DCT lung cases, it is currently unfeasible for a radiation oncology department to segment tumors and organs at risk, register the contours across all free breathing phases, and calculate dose-volume histograms for the tumors and organs prior to making decisions regarding methods for motion management and the associated types of scans and treatment setup. Instead, relative motions of tumors and organs are considered by the oncologist when determining motion management and treatment setup. Even with an experienced oncologist, this method is biased and may result in suboptimal treatment options. To tackle this shortcoming, this work will use an automatic end-to-end approach for segmentation, registration, and dose estimation for 4DCT lung cases using deep learning so that oncologists can use more objective metrics to determine treatment parameters and improve patient outcomes. Goals for the methods herein include improving realism for the deformations between breathing phases, improving accuracy for contours and registration, and outputting precise dose-volume histograms to aid oncologists in objective treatment creation.