Author: Adnan Jafar, Xun Jia, An Qin 👨🔬
Affiliation: Johns Hopkins University 🌍
Purpose: 3D whole-brain radiotherapy (WBRT) is widely used due to its simplicity and effectiveness. While modern treatment planning systems, like RayStation, offer automated Field-in-Field planning, patient-specific hyperparameters, such as the number of segments and target coverage priority, are still required to derive successful plans. This study presents a deep learning (DL) model to operate a treatment planning system of RayStation like a human for automatic treatment planning.
Methods: A DL-based model was developed to determine the optimal hyperparameter settings. Trained on 56 WBRT cases, the model leverages the geometric features of organs at risk (OARs) and the clinical target volume (CTV) to derive the optimal hyperparameter settings for the Auto Field-in-Field function in RayStation treatment planning system. The entire workflow was fully automated and integrated within RayStation, including retrieving DICOM files, extracting geometric features, executing the DL model, sending the hyperparameters back, and generating the final treatment plan, enabling planning with a single mouse click. We evaluated the DL-generated plans for clinical suitability in 15 independent test cases by a qualified medical physicist using a five-point rating scale. We compared the plans generated with the clinical plans based on clinically relevant metrics.
Results: Our of the 15 independent evaluation cases, 14 DL-optimized plans were clinically acceptable, either as is or with minor edits (i.e., dose normalization to achieve a similar D99(%) as the clinical plan). Only one plan was deemed unacceptable due to more hot spots in the brain and a high lens dose. The DL-method was able to generate plans on average in under 1 minute and total workflow time of 7 minutes with a single mouse click, as compared to 15 minutes for human planners.
Conclusion: The proposed approach significantly increased treatment planning efficiency and has the potential to streamline the creation of high-quality WBRT plans.