Independent check for radiotherapy treatment plans using clustering analysis

• 作 者: PengHuang;YuanTian;ZhihuiHu;WeijieCui;JianrongDai;
• 机 构: Department of Radiation Oncology,Cancer Hospital,Chinese Academy of Medical Science
• 出 处: The 15th Asia-Oceania Congress of Medical Physics (AOCMP2015
• 发表日期: 2015年期

　　Purpose:Many accidents could happen during treatment planning in radiotherapy.Data mining and machine learning techniques could be powerful tools for medical physicist to identify errors in radiotherapy treatment plan.In this study,we proposed a new method to help physicist detect human errors in intensitymodulated radiotherapy(IMRT)treatment plan and improve the efficiency of independent check using clustering analysis.Methods:A computer code was written to extract 835 breast cancer treatment plans using 4-field hybrid intensity-modulated radiotherapy(4F H-IMRT)from Mosaiq.To simplify the model,we considered fractional dose,beam angles and monitor units as the 'features' which composed the data.The data could be divided into different clusters by K-means clustering algorithm based on principal component analysis(PCA).The distance-based outliers were detected,and treatment plans which corresponded with outliers were analyzed manually by senior physicist to verify the accuracy of clustering analysis technique.Results:Treatment plan data were divided into 4 clusters using K-means clustering based on PCA.7 outliers were detected in treatment plan data.3 treatment plans considered as outliers were due to specificity of target,and the other treatment plans in outliers could make some certain improvements.Conclusions:Clustering analysis technique is useful to help physicist to detect potential catastrophic errors and improve the efficiency of human independent check in radiotherapy treatment plan.