Author: Jiajin Fan, Ulrich Langner, Qiongge Li, Jian Liu, Wei Nie, Edwin Quashie 👨🔬
Affiliation: Brown University Health, Hofstra University Medical Physics Program, Inova Hospital, Inova Schar Cancer Institute, Indiana University School of Medicine, Department of Radiation Oncology 🌍
Purpose:
Chart checking in radiotherapy ensures treatment plans meet clinical and safety standards. For patients in clinical trials, protocol-specific requirements add complexity, making manual reviews time-intensive and error-prone. This project aims to develop a software tool that automates chart checking to improve efficiency and ensure compliance with trial requirements.
Methods:
The tool uses the OpenAI GPT-4.0 mini model to process protocol documents (PDFs), extracting and summarizing key chart-checking requirements, such as dose constraints and imaging protocols. Patient treatment documents (PDFs) are analyzed to extract critical data, including contour names, prescriptions, beam parameters, and approval statuses. Two de-identified patient charts were tested, comparing extracted data with protocol requirements to identify discrepancies. A rule-based algorithm flagged inconsistencies and generated reports for review.
Results:
Initial testing demonstrated the tool’s ability to accurately identify errors. One patient chart had incorrect contour naming, while another had an out-of-range dose prescription. Summaries of protocol requirements were concise and relevant to chart checking. The tool effectively reduced the manual effort required for reviewing trial-specific patient charts while maintaining accuracy in detecting deviations.
Conclusion:
The proposed automation tool streamlines protocol-specific chart checking, improving efficiency and minimizing errors. By reducing manual review time, it enhances compliance with trial requirements and allows medical physicists to focus on critical tasks. Future developments will expand testing to more cases, refine NLP models, and integrate the tool with clinical systems like RayStation and Aria. This innovation aligns with the growing complexity of clinical trials and the need for streamlined workflows in medical physics.