Predicting Objective Function Weights for IMRT Prostate Treatment Planning Using Patient Anatomy
T Lee1*, M Hammad1, T Chan1, T Craig2, M Sharpe2, (1) University of Toronto, Toronto, Ontario, (2) UHN - Princess Margaret Cancer Centre, Toronto, ONSU-E-T-653 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To develop a prediction model for objective function weights for intensity-modulated radiation therapy (IMRT) prostate treatment planning with multiple objectives using geometry information from patient anatomy.
Methods: A previously developed inverse optimization method (IOM) was used to reverse-engineer optimal objective function weights (inverse weights) from an observed treatment plan. We developed a regression model to predict the weights for IMRT prostate treatment planning using patient anatomy from 25 patients. The ratio of the overlap volumes of the rectum and bladder with the planning target volume expanded by 1cm was used to predict the bladder and rectum weights. The femoral head weights were included in the model as a small fixed weight (1%). The model was validated using leave-one-out cross-validation. We evaluated the model by comparing the treatment plans generated through inverse planning using the inverse weights from IOM and the predicted weights from the regression model.
Results: On average, V54Gy for the bladder was 36.1% using the inverse weights and 36.6% using the predicted weights. V70Gy for the bladder was 23.2% (inverse) and 23.5% (predicted). For the rectum, V54Gy was 34.6% (inverse) and 33.9% (predicted), and V70Gy was 22.6% (inverse) and 22.3% (predicted). For each criterion, the difference between the inverse and predicted metrics was not statistically significant. All treatment plans from the predicted weights satisfied the clinical criteria.
Conclusion: Our results show that objective function weights are well-predicted by the regression model. This approach may support the genesis of personalized weights in IMRT treatment planning.
Funding Support, Disclosures, and Conflict of Interest: This research was supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC) and Ontario Graduate Scholarship (OGS).