Development of AI-Augmented quality assurance tools for radiation therapy Project summary Quality assurance (QA) is an essential part of radiation therapy (RT) workflow and critically determines the success of patient care. However, current treatment plan QA methods and tools are deficient in multiple aspects and suffer from problems such as limited accountability, labor intensive, and costly. In this project we will leverage the emerging deep learning techniques to create clinically translatable solutions for robust and efficient QA of modern RT. Specifically, we aim to (i) establish a novel framework for using deep learning to verify the machine delivery parameters of an RT treatment plan; (ii) investigate the use of deep learning for RT dose verification; and (iii) evaluate the performance of the QA system and show its potential clinical impact . This research presents the first-of-its-kind treatment plan QA strategy capable of providing both machine delivery parameters (MLC apertures and MUs of the involved IMRT/VMAT fields) and dosimetric distribution on the patient’s treatment geometry. The research will also make it possible to take advantage of the useful features of both deep learning models from Aims 1 and 2 and check the cycle-consistency of a treatment plan (i.e., from the beam parameters of the plan to the corresponding 3D dose distribution, and then from the 3D dose to the beam parameters) for enhanced plan QA. Successful completion of the project will provide urgently needed plan QA tools for safe, efficient and high-quality RT practice, and enable patients to truly benefit from modern RT modalities. Finally, the proposed strategy is quite broad and can be readily generalized for QA of other treatment modalities, such as proton therapy and high-dose rate (HDR) brachytherapy.