MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Modeling of Multiple Planning Target Volumes (PTVs) in Knowledge-Based Planning (KBP)

J Zhang1*, Y Sheng2 , C Wang3 , T Xie4 , F Yin5 , Y Ge6 , Q Wu7 , (1) Duke University Medical Center, Durham, NC, (2) Duke University Medical Center, Durham, NC, (3) Duke University Medical Center, Durham, NC, (4) Duke University Medical Center, Durham, NC, (5) Duke University Medical Center, Durham, NC, (6) UNC Charlotte, Charlotte, NC, (7) Duke University Medical Center, Durham, NC

Presentations

(Wednesday, 8/1/2018) 10:15 AM - 12:15 PM

Room: Davidson Ballroom A

Purpose: The purpose of this study is to develop a novel method that enables more accurate and reliable dose volume histogram (DVH) predictions for cases with multiple PTVs.

Methods: We propose a generalized distance-to-target histogram feature to represent the geometric relationship of OAR voxels with respect to two or more PTVs with different prescribed dose levels. A similarity metric that utilizes this information and primary PTV-to-boost PTV dose ratio is used to select a subset of similar cases from training set cases. A local regression model is subsequently built with these selected cases with a few handcrafted features. In addition, we incorporate a novel data augmentation method that effectively increases the number of training cases. Model parameters are tuned with 120 head and neck (HN) sequential boost cases. The model is then evaluated using a fresh validation dataset consisting of 148 2PTV HN cases with five-fold cross validation to avoid positive bias. We compare the proposed model with our previously reported model. The previous model uses two sets of features corresponding to two PTVs and train a model using stepwise multiple regression. Prediction accuracy is measured with root mean squared error (RMSE), which is the L2 norm of differences between predicted DVHs and actual clinical plan DVHs. Significance test is conducted with paired-sample t-test.

Results: The proposed method significantly outperforms the previous model in terms of prediction accuracy for all OARs evaluated, including brainstem (p<0.001), cord (p<0.001), larynx (p=0.004), mandible (p<0.001), oral cavity (p=0.011), parotid (p<0.001), and pharynx (p=0.001).

Conclusion: The proposed model generates accurate and reliable plan DVH predictions when multiple-PTV plans are involved. It is expected that the improvement will translate into better plan quality. The proposed method could potentially serve the purpose of simultaneous integrated boost (SIB) automatic planning, and/or guidance of sequential boost planning.

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.

Keywords

Treatment Planning, Statistical Analysis, Modeling

Taxonomy

TH- Dataset analysis/biomathematics: Machine learning techniques

Contact Email