Room: Exhibit Hall | Forum 7
Purpose: To implement a deep convolutional neural network (CNN) based dose prediction method with incorporation of dosimetric features from PTV-only treatment plan (i.e., the plan with the best PTV coverage by sacrificing the OARs sparing) to facilitate treatment planning workflow.
Methods: Existing prediction models learn geometric/anatomic features for building the dose prediction models and few features that characterize the dosimetric properties of the patients were utilized. In this study we propose a novel method to include the dosimetric features in the construction of a more reliable dose prediction model based on the deep CNN. The deep CNN not only learned the geometric/anatomical features from the contours of critical structures but also captured the dosimetric features from the dose distribution of the PTV-only plans. A database of 60 volumetric modulated arc therapy (VMAT) plans for the prostate cancer patients were used for training. The trained dose prediction model was then tested on a cohort of 10 cases. Performance valuation for the proposed method was conducted by examining the dose difference map, the DVHs derived from the predicted dose distribution, the dosimetric endpoints and the sum of absolute residuals (SARs).
Results: Statistical results showed that the mean SARs for the PTV, bladder and rectum with our method were 0.007Â±0.003, 0.035Â±0.032 and 0.067Â±0.037 respectively, smaller than the data obtained with the existing contours-based method, indicating the great potential of the proposed approach in accurately predicting dose distribution.
Conclusion: We proposed a novel deep CNN based dose prediction framework for VMAT planning, which fully exploited the dosimetric features learned from the dose distribution of the PTV-only plan. Experimental results demonstrated the superior performance of the proposed method than the existing contours-based method, indicating a great potential in clinical application in knowledge-based planning and quality control.
Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by Varian Medical Systems, a gift fund from Huiyihuiying Medical Co, and a Faculty Research Award from Google Inc.
Not Applicable / None Entered.