Program Information
Capturing Different Sparing Decisions in Head-And-Neck IMRT Cases Using Model Trees
Y Sheng1*, Q Wu1 , J Zhang1 , T Xie1 , F Yin1 , J Kirkpatrick1 , Y Ge2 , (1) Duke University Medical Center, Durham, NC, (2) University of North Carolina at Charlotte, Charlotte, NC
Presentations
SU-K-FS1-7 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: Four Seasons 1
Purpose: To develop knowledge models that capture head-and-neck (HN) intensity modulated radiation therapy (IMRT) cases with single-lateral and bi-lateral parotid sparing decisions using model tree.
Methods: Seventy-three HN IMRT cases were included. All parotid glands except 14 right and 14 left given up parotid glands were considered for modeling. A baseline bi-lateral model was trained using 80 bi-laterally spared parotids and a single-lateral model was trained using 23 single-laterally spared parotids. The model tree was trained with the combined set of 80 bi-laterally and 23 single-laterally spared parotids. A standard model was trained using the same dataset as the model tree. The remaining 10 bi-laterally spared parotids and 5 single-laterally spared parotids were used to validate the three models. The experiment was repeated 20 times using bootstrap. The Weighted Sum of Residual (WSR) was used to evaluate the accuracy of dose-volume histogram (DVH) prediction. The difference between predicted parotid D50% and the clinically planned parotid D50% was assessed.
Results: The mean WSR of the validation bi-lateral cases was 0.021,0.034,0.026 and 0.022 for the bi-lateral model, the single-lateral model, the standard model and model tree, respectively. For the single-lateral validation cases, the mean WSR was -0.007,0.010,-0.001 and 0.004 for the bi-lateral model, the single-lateral model, the standard model and model tree, respectively. The mean D50% difference of the bi-lateral model, the single-lateral model, the standard model and model tree were -0.40Gy,-2.00Gy,-1.12Gy and -0.87Gy for bi-lateral cases, and 1.63Gy,-0.13Gy,0.89Gy and 0.60Gy for single-lateral cases.
Conclusion: Results showed the deficiency of using a single regression model for HN IMRT modeling because different parotid sparing decisions were involved in clinical cases. The model tree achieved model prediction accuracy similar to the ideal performance of baseline models. It suggests that the model tree is effective in modeling HN IMRT cases with different parotid sparing decisions.
Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by NIH under grant #R01CA201212 and a master research grant by Varian Medical Systems.
Contact Email: