Room: AAPM ePoster Library
Purpose: Adaptive plan quality in MRI guided online adaptive radiotherapy (MRgRT) is difficult to assess in relation to the fully optimized, high quality plans traditionally established offline. ANN models were developed to predict 3D dose distributions, enabling the evaluation of online adapted plan quality to better inform adaptive decision-making in MRgRT.
Methods: Over 300 treatment plans from 53 abdominal cancer patients undergoing linac-based MRgRT were analyzed. ANN models were developed to predict per voxel dose inside the GTV using input variables related to patient anatomy and target/OAR relationships. The models were designed to be simple (two nodes in a single hidden layer) in order to avoid overfitting. Beam related variables such as beam angles or fluence are not included as input parameters to enable 3D dose prediction using only target and OAR geometry for guiding treatment planning. Five inferior plans identified by the models were manually re-planned to confirm if plan quality could be improved.
Results: The dose prediction error and the absolute error were 0.1 ± 3.4 Gy (0.1 ± 6.2%) and 3.5 ± 2.4 Gy (6.4 ± 4.3%), respectively. Plan metric prediction errors were -0.1 ± 1.5%, -0.5 ± 2.1%, -0.9 ± 2.2 Gy, and 0.1 ± 2.7 Gy for V95, V100, D95, and Dmean, respectively. Plan metric prediction absolute errors were 1.1 ± 1.1%, 1.5 ± 1.5%, 1.9 ± 1.4 Gy, and 2.2 ± 1.6 Gy. Approximately 10% of the plans studied were clearly identified by the prediction models as inferior quality plans needing further optimization and refinement. Manual replanning of 5 of such inferior plans increased GTV V95% by 5% on average.
Conclusion: The developed ANN models can accurately predict 3D dose distributions and overall plan quality metrics for the adapted plans. The models are useful to identify inferior plans and recommend adjustments in planning optimization.