Room: Exhibit Hall | Forum 8
Purpose: To develop a patient-specific and intellectualized treatment planning optimization framework by guiding with an in-house developed 3D dose distribution prediction engine so as to improve plan quality and planning efficiency.
Methods: This framework mainly contains two parts: a patient-specific optimal 3D-dose-distribution prediction and thereupon prediction-oriented RBE-based plan optimization. The in-house developed deep learning trained geometric-dosimetric prediction model was firstly performed to afford a close-to-optimal plan 3D-dose-distribution, as the starting point to the following IMRT plan optimization. Afterwards, a constrainted voxel-based optimization algorithm was adopted with a goal of achieving the prediction while minimizing their RBEs. Here an SQP method was applied to solve this problem. To evaluate this framework, 25 clinical treated GYN IMRT plans were collected to have 20 of them for modeling and the other 5 for evaluation. Both the plan quality and planning efficiency were investigated by comparing the original plan to the optimized plan in terms of ROI's DVH and detailed dosimetric endpoint.
Results: Both DVH and dose distribution results show comparable target coverage and improved OAR sparing of our proposed optimized plan in comparison with original plans. For V45 of the rectum, bladder, and V30 of the femoral heads, endpoint decreased around 3%-4%, 6%-8%, and 6%-7%, for our proposed optimized plan respectively. For the efficiency, 70-80 iterations were needed for a one time optimization. Comparison of using original plan dose distribution to predicted dose distribution to guide subsequent optimization illustrate promising results of plan quality and efficiency improvement.
Conclusion: We have successfully developed an intelligent 3D Dose distribution prediction guided patient-specific treatment planning optimization framework for intensity modulated radiation therapy. This method can not only raise the routine working efficient, but also assure the plan quality.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by (1) National Natural Science Foundation of China (81601577 and 81571771 ), (2)National Key R&D Program of China (2017YFC0113200), (3) Post-doctoral Science Foundation of China ( 2016M592510), and (4) the Scientific Research Foundation for the Returned Overseas Chinese Scholars of school (LX2016N004).