Encrypted login | home

Program Information

Predictive Modeling of An Adaptive Replanning Physician Decision Support Tool for Stage III Lung Cancer

D Bollinger

D Bollinger*, J Kavanaugh , Washington University in St. Louis, St. Louis, MO


TU-C2-GePD-J(B)-1 (Tuesday, August 1, 2017) 10:00 AM - 10:30 AM Room: Joint Imaging-Therapy ePoster Lounge - B

Purpose: To construct an automated model to predict the necessity of adaptive replanning for a patient based upon changes of the dosimetric quality of the treatment plan.

Methods: Eighteen patients who received a rescan while undergoing intensity modulated radiation therapy for stage III lung cancer were identified. In nine cases analyzed, the rescan resulted in a physician ordering a new plan while for the other nine cases they elected to maintain the original plan for the remaining treatments. For each patient, the dosimetric quality of 2 plans was evaluated: an IMRT RapidPlan generated on the original CT dataset and the same plan recalculated on the rescan CT dataset. Physician specified critical dosimetric end points, including PTV coverage, V20 and Dmean lung dose, and Dmean heart dose were extracted from each plan. The patient population was randomly subdivided into a training set consisting of 16 patients and a 2 patient validation set. Using a boosting based machine learning approach, a model was constructed to predict if a physician would order a replan. The model was based on not only identified dosimetric metrics in both plans, but also the percentage change in these metrics between plans.

Results: On the training set, leave one out cross validation was performed with the algorithm successfully classifying all but two plans as either requiring a replan or maintaining the initial plan. The model was applied to the validation set and accurately predicted the physician’s decision in both cases.

Conclusion: A model capable of predicting a physician’s decision to replan lung cases based on dosimetric metrics on a rescan has been developed. Early results indicate promising potential. The model will now be expanded to include more cases with the ultimate goal of producing a highly accurate decision support tool for physicians considering the need for replanning.

Contact Email: