IClustering analysis: assistance to verification of radiotherapy treatment plans

Abstract

Objective To use clustering analysis to help physicians detect abnormal parameters in radiotherapy treatment plans and improve the efficiency of plan verification. Methods From 2010 to 2015, 835 breast cancer treatment plans for using 4-field hybrid intensity-modulated radiotherapy from MOSAIQ were collectted. Fractional dose, beam angle, and monitor unit were used as featured parameters of a treatment plan to generate a dataset. The K-means clustering algorithm based on principal component analysis was used to perform a clustering analysis of the dataset and divide the dataset into different clusters. The outliers of clusters were automatically detected based on the distance threshold. The outlier-contained treatment plans were manually verified by physicians to determine the accuracy of clustering analysis in detection of abnormal plans. Results In the clustering analysis, the sample space composed by parameters of treatment plans for breast cancer was divided into 4 clusters, 3 of which had outliers detected. In the targeted treatment plans, 3 plans became outliers because of special target volume and the other 4 plans needed improvement. Conclusions Clustering analysis is effective to help physicians to independently verify treatment plans. Key words: Clustering analysis; Outlier detection; Radiotherapy treatment plan; Independent check