Program Information
Support Vector Regression Based Dose Distribution Prediction for Intensity Modulated Radiotherapy
F Kong1 , Y Li2 , L Zhou1 , T Song1*, (1) Southern Medical University, Guangzhou, Guangdong, (2) Image Processing Center, Beihang University, Beijing, Beijing
Presentations
TU-C1-GePD-JT-4 (Tuesday, August 1, 2017) 9:30 AM - 10:00 AM Room: Joint Imaging-Therapy ePoster Theater
Purpose: To develop a patient-specific 3D dose distribution prediction method for intensity modulated radiotherapy based on big data and machine learning techniques.
Methods: A patient geometry and plan dosimetry correlation model was established with the voxel dose as output dosimetric feature, while input geometric feature included distance of that voxel to the PTV boundary, ray energy, and beam angle. Thereupon, a support vector regression and radial basis function were adopted to construct the aforementioned relational model with its advantage of high dimensional mapping for nonlinear data. To verify the feasibility of our method, 25 prostate SBRT plans were retrospectively studied with a prescription dose of 45Gy. 21 of them were used for training and the remaining 4 were for validation. Dosimetric difference were investigated for original (clinical) plan and the model prediction with the terms of 3D dose distributions and also DVHs for the PTV, rectum and bladder.
Results: Dosimetric comparison show great agreement between the clinical approved plan and our model prediction, with little dose difference and similar DVH shapes for the bladder of these evaluated 4 cases. More specifically, the average DVH differences between clinical and mode prediction are 1.46±0.13%, 1.21±0.26%, 0.60±0.03% and 0.92±0.03% for each evaluated case, and the maximum dose difference is 24.18% and 27.65%, average dose difference is -1.69±7.94% and 0.71±8.21%, for two of them.
Conclusion: We have successfully developed a method which is able to predict voxel dose distribution for IMRT. This prediction can not only be used as evaluation criterion of the plan quality, but also guiding subsequent planning optimization, laying a foundation of the automatic planning.
Funding Support, Disclosures, and Conflict of Interest: This work is supported by the National Natural Science Foundation of China (No.81601577,81571771), China Postdoctoral Science Foundation and the National Natural Science Foundation of China (No.2016M592510), and Southern Medical University School start-up fund(No.LX2016N0004)
Contact Email: