Cervical cancer affects approximately 7.4/100,000 women in the US with minority women being disproportionally affected. High dose-rate (HDR) brachytherapy is an effective treatment modality by irradiating the tumor from inside while simultaneously sparing dose to nearby organs. Multiple randomized trials revealed that including HDR brachytherapy in cervical cancer radiotherapy increases overall survival rates from 46.2% to 58.2%. Yet HDR brachytherapy has seen a downward trend in its use, with complex treatment planning process being one of the major reasons, causing a 13% reduction in cause-specific survival rate. For HDR brachytherapy, the planning has to be accomplished in a very short period of time to reduce patient discomfort and patient motion during the planning process, which could causes substantial deviation of planning geometry from delivery geometry (1.5mm motion leads to >10% change in dose). At present, it takes on average 135 min to manually generate a treatment plan. Due to the time pressure, suboptimal plans are often generated. D2cc, the dosimetric quantity correlated to organ toxicity could be further reduced up to 8.4%, corresponding to 5.1% reduction of complication rate. Therefore, there is a strong need to develop a fully automated treatment planning process to reduce planning time, improve plan quality and therefore outcome, reduce patient discomfort, and most significantly, to promote the use of HDR brachytherapy. Our group has successfully developed and clinically implemented the AutoBrachy system, an automatic planning tool that generates a patient-generic plan in 3 minutes. However, automatic organ segmentation and physician-oriented treatment planning are missing, preventing a fully automated process to generate the patient-specific optimal plan. Current practice relies on human expertise to solve the two problems, taking up to 90 min for segmentation and 30 min for planner and physician to finalize plan. Given the recent advancement of artificial intelligence (AI) in mimicking humans to achieve human-level performance, we believe that AI can be used to empower AutoBrachy to achieve a fully automated planning process. The overall goal of this fellowship project is to develop AI-based organ segmentation and human-like treatment planning modules in AutoBrachy, and to validate the system in real patient cases. We will pursue two specific aims (SAs): SA1: System development. Create an automatic organ segmentation module and a human-like automatic planning module. SA2: System validation. Perform a retrospective study to evaluate the effectiveness of the developments. The innovation is the use of state-of-the-art AI techniques to solve critical problems in HDR brachytherapy. This project is logically built on the candidate?s engineering background and medical physics training. It targets at a disease that is highly significant to the underserved population, for which the candidate is highly motivated to solve. The project also fits perfectly with the candidate's long-term career goal of establishing a high-quality independent research program to develop state-of-the-art treatment methods for cancer radiotherapy.
This project will develop and validate artificial intelligence based computational modules for automatic organ segmentation and human-like treatment planning to achieve a fully automated and streamlined treatment planning process for high-dose-rate brachytherapy treatment of cervical cancer. Cervical cancer is a severe disease, particularly affecting underserved population who typically present at advanced stages. The developed tools will lead to clinical benefits to patients by improving plan quality, planning efficiency, and patient comfort, as well as will reduce the technical burden of high-dose-rate brachytherapy and therefore promote the use of this effective therapeutic modality to yield broad impacts.
