Click here to


Are you sure ?

Yes, do it No, cancel

Machine Learning for Patient-Specific Quality Assurance of VMAT: Prediction and Classification Accuracy

J Li1*, L Wang2 , X Zhang1 , L Liu1 , J Li1 , M Chan3 , J Sui2 , R Yang1 , (1) Peking University Third Hospital, Beijing, (2) Institute of Automation, Chinese Academy of Sciences, Beijing, (3) Memorial Sloan Kettering Cancer Center, Basking Ridge, NJ


(Thursday, 7/18/2019) 7:30 AM - 9:30 AM

Room: Stars at Night Ballroom 2-3

Purpose: To assess the accuracy of machine learning to predict and classify patient-specific quality assurance (QA) results for volumetric modulated arc therapy (VMAT) plans.

Methods: 255 VMAT plans including 143 gynecological cancer (GYN) and 112 head and neck cancer (H&N) plans were randomly chosen in this study. 54 complexity metrics affecting dose delivery accuracy of VMAT were extracted from the QA plans and considered as inputs. Patient-specific QA was performed, gamma passing rates (GPR) were calculated and used as outputs. One Poisson Lasso (PL) regression model was developed aiming to predict individual GPR and one Random Forest (RF) classification model was developed to classify QA results as “pass� or “fail�. Principal component analysis, random under-sampling and ensemble voting were used to avoid overfitting and rebalance the training data. Nested cross-validation was used for model training and testing. GPR prediction accuracy of PL and classification performance of PL and RF were evaluated.

Results: In GPR prediction using PL, 230 (90.20%) plans had prediction error smaller than 3.5% at 3%/3mm; 233 (91.37%) plans had prediction error smaller than 5% at 3%/2mm. In QA results classification, PL had a higher specificity (accurately identify plans that can pass QA) while RF had a higher sensitivity (accurately identify plans that may fail QA). By using 90% as action limits at 3%/2mm criteria, the specificity of PL and RF were 97.46% (230/236) and 87.71% (207/236), respectively; the sensitivity of PL and RF were 31.58% (6/19) and 100% (19/19), respectively. With 100% sensitivity, the patient-specific QA workload of 81.18% plans (207/255) labeled “pass� by RF could be reduced.

Conclusion: PL model could accurately predict GPR for majority VMAT plans. RF model with 100% sensitivity was preferred for QA results classification. Machine learning is proven to be a powerful tool to assist patient-specific QA and reduce QA workload.

Funding Support, Disclosures, and Conflict of Interest: This study was supported by National natural Science Foundation of China (No.81071237)


Quality Assurance, Classifier Design, Cross Validation


TH- External beam- photons: Quality Assurance - VMAT

Contact Email