A study of automatic treatment planning based on kernel density estimation

Abstract

Objective To develop an automatic algorithm to predict the dose-volume histogram (DVH) and implement it in clinical practice. Methods Based on the prior information in the existing plan, such as dosimetric results of organs at risk (OARs) and OAR-target spatial relationship, a two-dimensional kernel density estimation was implemented to predict the DVH of OARs. The predicted DVH curves were converted into objective functions that would be implemented in the Pinnacle treatment planning system. Comparisons between predicted and actual values and between Auto-plan and manual planning were made by paired t test. Results We applied this algorithm to 10 rectal cancer patients, 10 breast cancer patients, and 10 nasopharyngeal carcinoma patients. The predicted DVH of OARs showed that the deviation between the actual and predicted values at important clinical dose points were within 5%(P>0.05). The re-planning for the 10 breast cancer patients using Auto-plan showed that the heart dose was significantly reduced and the target coverage was increased, which was consistent with the predicted results. Conclusions The method proposed in this study allows for accurat DVH prediction, and, combined with Auto-plan, can be used to generate clinically accepted treatment plans. Key words: Intensity modulated radiation therapy; Dose volume histogram; Kernel density estimation; Automatic treatment planning