End-to-End X-ray-based Cancer Imaging Simulation Toolkit for Technology Evaluation, AI Training, and Virtual Imaging Clinical Trials - PROJECT SUMMARY/ABSTRACT Simulations play three crucial roles in researching and developing new x-ray and computed tomography (CT) technologies, particularly in the context of cancer imaging. 1) New imaging modalities often take more than a decade after conception to become clinically available, due to the cost and time it takes to prototype new imaging systems. Simulations can dramatically shorten this process. 2) Optimizing imaging efficacy over a population of patients and tumors for optimal task-based performance requires extensive patient trials. Virtual imaging trials (VITs), based on simulations, have the potential to partly replace real-world clinical trials. 3) Development of deep learning algorithms for tumor detection, tumor characterization, and dose reduction requires substantial amounts of curated training data. Simulations can provide an excellent source of training data to supplement real-world data. We recently released the open-source and vendor-neutral X-ray-based Cancer Imaging Simulation Toolkit (XCIST), which has already accelerated the work of dozens of research groups. This realistic and accurate x-ray and CT imaging system simulation toolkit is publicly available, includes realistic software phantoms of patients and tumors, and advanced computer models of CT and x-ray imaging systems. In this project, we will develop the next-generation XCIST 2.0, which will include 1) physics models for photon-counting CT, phase-contrast imaging, dose estimation, and deep learning reconstruction; 2) end-to-end capability for performing virtual cancer imaging trials and deep learning training, including image quality phantoms, virtual lesions, and model observers; and 3) major enhancements for improved usability, including a graphical user interface, GPU acceleration, cloud deployment, documentation, and video tutorials. After combining these critical capabilities in one integrated tool, XCIST 2.0 will enable and accelerate the development of emerging CT cancer imaging technologies, enable performance evaluation and optimization of specific CT cancer imaging applications through virtual imaging trials, and provide a user-friendly environment for deep learning training data generation.
