SU-F-BRD-03: Evaluation of Head-Neck IMPT Plans Quality Using a Knowledge Based Model

Authors


Abstract

Purpose:

To develop and evaluate a model that predicts the dose volume histogram (DVH) of head and neck IMPT plans based on knowledge of prior plans.

Methods:

Total of forty Head-Neck cancer patients underwent IMPT were used in the study. Using thirty patient CT, contours, IMPT treatment plans and calculated dose as input, significant factors that predicts the dose volume histogram (DVH) for organs at risks (OARs) including right/left parotid, oral cavity, esophagus, brain stem, larynx, and spinal cord were identified using principle component analysis (PCA) method. The model was then evaluated by comparing model predicted DVH for with planned DVH for ten additional IMPT plans.

Results:

Combined distance-to-target Histogram (DTH) Principal Components (PCs) were found to be the most important in the first principal component of parotids, oral cavity, and esophagus. The quality of the model prediction is correlated to the knowledge or quantities of the plans that model learned. For patients with tumor and OARs geometry deviates from the training data set, the predicted DVH may be inferior to the actual treatment plans. For patients with tumor and OARs geometry similar to the training data set, the model was able to predict the performance of the actual treatment plan.

Conclusion:

The quality of DVH prediction knowledge base model was investigated using head-neck plans. The results indicated that the performance of the model is dependent on the quality and quantity of the training data set. The prediction model may provide real time evaluations for plan optimization in complex patient geometries such as head and neck cancer with multiple dose levels.

Supported by Research agreement with Varian Medical Systems.

Ancillary