Application of kernel density estimation in predicting bone marrow dose of radiation therapy for gynecological tumors

Xiufeng Cong; Jun Chen; Jingchao Zhang; Xiaoting Zhang; Zaiming LU

Return

Application of kernel density estimation in predicting bone marrow dose of radiation therapy for gynecological tumors

VernacularTitle:应用核密度估计方法预测妇科肿瘤放疗骨髓剂量研究
Author: Xiufeng CONG ; Jun CHEN ; Jingchao ZHANG ; Xiaoting ZHANG ; Zaiming LU
From: Chinese Journal of Radiation Oncology 2021;30(3):262-265
CountryChina
Language:Chinese
Abstract: Objective:To predict the dose of lumbosacral spine (LS) and pelvic bone marrow (PBM) based on kernel density estimation (KDE) in patients with gynecological tumors.Methods:Fifteen patients with gynecological tumors receiving radiotherapy plans with dose limitation for LS and PBM in our hospital were selected as training data for machine learning. Another 10 cases were selected as the data for model validation. The minimum directional distance between the dose point in the organs and the edge of the planned target volume for the LS and PBM was calculated. Model training was performed by KDE. The accuracy of the model prediction was evaluated by the root mean square error. The model was utilized to predict the actual planned doses of the LS and PBM, and a linear fitting was performed on the predicted dose volume histogram (DVH) and actual results. The prediction effect was assessed by the goodness of fit R 2. Results:In terms of the DVH parameters required by the planner, the prediction doses from the model were similar to those of the verification plans: the difference of PBM V 40Gy was 2.0%, the difference of the mean dose was 1.6 Gy, and the difference of LS V 10Gy was -0.4%. In the unrequired DVH parameters, except for the PBM V 10Gy, the predicted values of the model were significantly high. The difference between the DVH predicted by the model and the actual plan was small, and the R 2 of the LS and PBM were 0.988 and 0.995, respectively. Conclusions:The model based on KDE method can accurately predict the doses of the LS and PBM. This model can also be used as a method to ensure the quality of the plan, and improve the consistency and quality of the plan.