Program Information
How Many Parameters Are Required to Characterize DVHs in Clinical Circumstance?
Z Chen*, J Wang , W Hu , Fudan university shanghai cancer center, Shanghai, China
Presentations
SU-F-BRB-8 (Sunday, July 12, 2015) 4:00 PM - 6:00 PM Room: Ballroom B
Purpose:The aim of this study is to assess how many parameters are required to characterize the DVH curves by principal component analysis (PCA) in two tumors sites.
Methods:Two tumor sites including nasopharyngeal cancer and lung cancer were enrolled in this study. Each tumor site contains 5 patients. Multiple criteria optimization (MCO) algorithm was used to generate the plan database to cover all the possible optimization solution in RayStation (RaySearch Laboratories AB, Sweden). The constraints and trade-off objectives setting were based on the QUANTEC research and planning experience. The target and normal tissue DVHs were export to MATLAB. Each DVH on plans were resampled into 101 points for principal component analysis. A threshold of 95% was used to decide how many components should be selected to characterize DVH curves. Finally, we evaluate the accuracy of the data extraction by regenerating DVH from the selected principal components (PCs). The D95 of PTV, max-dose of spinal cord and brainstem, mean dose of parotid and other related organs at risk in nasopharyngeal cancer; and the D95 of PTV, the V20Gy of lung-PTV, and mean dose of heart in lung cancer were analyzed.
Results:The PCs used to characterize DVH were 1-3 in nasopharyngeal cancer, 2-3 in lung cancer. The regenerated DVHs were coincided with the original DVHs in both cancers. The mean dose differences of all clinical points between the regenerated and the original were 0, and the maximize deviation is less than 100cGy except some parameters in the nasopharynx cancer, which has the maximal dose difference of 177cGy in spinal cord and 608cGy in brainstem.
Conclusion:It is feasible to characterize the DVH curves by no more than 3 principal independent variables in nasopharyngeal cancer and lung cancer.
Contact Email: