Modeling for knowledge-based plan assisted with Model Analytics

Meijiao WANG,Sha LI,Haizhen YUE,Jian GONG,Xiaoyu XIANG,Wen GUO,Yibao ZHANG

DOI: https://doi.org/10.3969/j.issn.1005-202X.2017.09.002

2017-01-01

Abstract:Objective To reduce the component uncertainty induced by the complexity and subjectivity of outlier processing during the RapidPlan model verification with the use of Varian Model Analytics (MA),and to evaluate the efficiency,statistical and dosimetric performances of MA.Methods An initial RapidPlan model was configured with 81 best-effort plans.The established initial model was uploaded to the MA for automated statistical analysis,and MA suggested that outliers were processed batch by batch.Statistical indexes were compared before and after verification.The dosimetric outcomes of 20 validation cases were re-planned by the RapidPlan models before and after verification,and compared with the clinical manual plans.Results MA took a few minutes to analyze the geometric and dosimetric statistics of the plans.Eight dosimetric outliers were reported in 5 rouads of verifications,which were higher than the estimated upper limits by 11.11％,5.88％,5.56％,5.56％,5.00％,5.26％,5.56％ and 5.88％,respectively.R2 value was improved from 0.32 to 0.45 after statistical verification.Only performing 1 round of verification,we found out all 3 outliers for urinary bladder,where the values of geometric outliers were 62.22％ higher or 55.35％ lower than the majority average values.The value of dosimetric outlier was 3.33％ higher than the estimated upper limit.The R2 value was improved to 0.37 from 0.35 after statistical verification.Validation cases suggested that RapidPlan plan prevails over conventional manual plan significantly (P＜0.05),in which the dose to femoral head was reduced by 23.15％ and 27.55％ using model before and after verification,respectively,and the dose to urinary bladder was reduced by 8.14％ and 6.79％,respectively.Conclusion Statistical features of RapidPlan plans can be acquired quickly using MA which detects outliers hence enhances the efficiency of model configuration for knowledge-based planning.The effect of statistical verification on the dosimetric outcome is marginal.