The purpose of this study is to evaluate the prediction and classification performances of the gamma passing rate (GPR) for different machine learning models and to select the best model for achieving machine learningbased patient-specific quality assurance (PSQA). Methods: The measurement verification of 356 head-and-neck volumetric modulated arc therapy plans was performed using a diode array phantom (Delta4 Phantom), and GPR values at 2%/2 mm with global normalization and 3%/2 mm with local normalization were calculated. Machine learning models, including ridge regression (RIDGE), random forest (RF), support vector regression (SVR), and stacked generalization (STACKING), were used to predict the GPR. Each machine learning model was trained using 260 plans, and the prediction accuracy was evaluated using the remaining 96 plans. The prediction error between the measured and predicted GPR was evaluated. For the classification evaluation, the lower control limit for the measured GPR and lower control limit for predicted GPR (LCL p ) was defined to identify whether the GPR values represent a "pass" or a "fail." LCL p values with 99% and 99.9% confidence levels were calculated as the upper prediction limits for the GPR estimated from the linear regression between the measured and predicted GPR. Results: There was an overestimation trend of the low measured GPR. The maximum prediction errors for RIDGE, RF, SVR, and STACKING were 3.2%, 2.9%, 2.3%, and 2.2% at the global 2%/2 mm and 6.3%, 6.6%, 6.1%, and 5.5% at the local 3%/2 mm, respectively. In the global 2%/2 mm, the sensitivity was 100% for all the machine learning models except RIDGE when using 99% LCL p . The specificity was 76.1% for RIDGE, RF, and SVR and 66.3% for STACKING; however, the specificity decreased dramatically when 99.9% LCL p was used. In the local 3%/2 mm, however, only STACKING showed 100% sensitivity when using 99% LCL p . The decrease in the specificity using 99.9% LCL p was smaller than that in the global 2%/2 mm, and the specificity for RIDGE, RF, SVR, and STACKING was 61.3%, 61.3%, 72.0%, and 66.8%, respectively. Conclusions: STACKING had better prediction accuracy for low GPR values than other machine learning models. Applying LCL p to a regression model enabled the consistent evaluation of quantitative and qualitative GPR predictions. Adjusting the confidence level of the LCL p helped improve the balance between the sensitivity and specificity. We suggest that STACKING can assist the safe and efficient operation of PSQA.

PurposeCarbon-ion beam (C-beam) has a sharp dose distribution called the Bragg peak. Carbon-ion radiation therapy, such as stereotactic body radiotherapy in photon radiotherapy, can be completed in a short period by concentrating the radiation dose on the tumor while minimizing the dose to organs at-risk. However, the stopping position of C-beam is sensitive to density variations along the beam path and such variations can lower the tumor dose as well as cause the delivery of an unexpectedly high dose to the organs at risk. We evaluated the clinical efficacy of a robust planning technique considering gastrointestinal gas (G-gas) to deliver accurate radiation doses in carbon-ion radiotherapy for pancreatic cancer.Materials and methodsWe focused on the computed tomography (CT) value replacement method. Replacement signifies the overwriting of CT values in the CT images. The most effective replacement method for robust treatment planning was determined by verifying the effects of the three replacement patterns. We selected 10 consecutive patients. Pattern 1 replaces the CT value of the G-gas contours with the value of the region without G-gas (P1). This condition indicates a no-gas state. Pattern 2 replaces each gastrointestinal contour using the mean CT value of each contour (P2). The effect of G-gas was included in the replacement value. Pattern 3 indicates no replacement (P3). We analyzed variations in the target coverage (TC) and homogeneity index (HI) from the initial plan using in-room CT images. We then performed correlation analysis on the variations in G-gas, TC, and HI to evaluate the robustness against G-gas.ResultsAnalysis of variations in TC and HI revealed a significant difference between P1 and P3 and between P2 and P3. Although no statistically significant difference was observed between P1 and P2, variations, including the median, tended to be fewer in P2. The correlation analyses for G-gas, TC, and HI showed that P2 was less likely to be affected by G-gas.ConclusionFor a treatment plan that is robust to G-gas, P2 mean replacement method should be used. This method does not necessitate any particular software or equipment, and is convenient to implement in clinical practice.

No abstract