(Wednesday, 8/1/2018) 10:15 AM - 12:15 PM
Room: Davidson Ballroom A
Purpose: The aim of this study is to improve knowledge-based planning (KBP) robustness and consistency across different datasets.
Methods: Knowledge-based planning models utilize experienced planners’ knowledge embedded in prior cases to estimate optimal achievable dose volume histogram (DVH) of new cases. In the regression-based KBP framework, previous cases’ anatomical features and DVHs are extracted, and prior knowledge is summarized as the regression coefficients that map features to OAR DVH predictions. In our study, we find that in different settings (such as different number of training cases, different number of outliers in the training set), different regression methods perform better. To improve the consistency of KBP models, we propose an ensemble method that combines the strengths of various regression models, including stepwise, lasso, elastic net, and ridge regression. In the ensemble approach, we first obtain individual model prediction metadata using in-training-set leave-one-out cross validation. A constrained optimization is subsequently performed to determine individual model weights. The metadata is also used to filter out impactful training set outliers. We evaluate our method on a fresh set of prostate IMRT cases and head and neck IMRT cases.
Results: The proposed approach is evaluated against individual models in simulated adverse clinical situations including small training set size, wrongly labeled cases, and dosimetric inferior plans. The proposed ensemble method consistently performs better than or similarly well as the best performing individual model for each dataset. Furthermore, the ensemble model significantly outperforms every individual model in at least one situation.
Conclusion: In summary, we propose an ensemble method to build a knowledge-based planning model that is capable of handling less than optimal clinical datasets. With improved robustness, the proposed regression method potentially enables end users to build institution-specific, physician-specific, or even planner specific models.
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, Modeling, Radiation Therapy
Taxonomy
TH- Dataset analysis/biomathematics: Machine learning techniques
Contact Email
