Exhibit Hall | Forum 7
Purpose: Although the knowledge-based dose volume histogram(DVH) prediction has been largely applied and researched in External Beam Radiation Therapy, it’s still far relatively unexplored in the realm of brachytherapy. The purpose of this study is to develop a reliable DVH prediction method for high-dose-rate brachytherapy plans.
Methods: We present a DVH prediction workflow, combing kernel density estimation(KDE) and k-nearest neighbor(kNN) to build the prediction model. The database consists of 79 cervical cancer cases with different applicators inserted. kNN(k=30) is used to select a subset of similar cases from the database for model building. KDE builds a model based on the relationship between distance-to-target (DTH) and the dose in selected cases, which can be subsequently used to estimate the dose probability distribution in the test set. Model performance of bladder and rectum was quantified by ∆D2cc, ∆D1cc, ∆D0.1cc, ∆Dmean, ∆Dmax (mean and standard deviation).
Results: The dose deviation between ground truth and the proposed model were 0.27±0.41, 0.28±0.47, 0.26±0.64, -2.86±1.73, and -0.03±0.24 for D2cc, ∆D1cc, ∆D0.1cc, ∆Dmax, and ∆Dmean in bladder, respectively. For rectum, the deviation for the metrics were -0.05±0.22, -0.07±0.22, -0.3±0.6, -3.51±0.62, -0.096±0.1, respectively. The addition of kNN largely improved the prediction performance. Compared to KDE-only model, the model combines KDE and kNN improves 37.0% for bladder and 37.5% for rectum in D2cc.
Conclusion: In this study, a knowledge based machine learning model was proposed and was demonstrated to accurately predict the DVH for HDR brachytherapy.
Dose Volume Histograms, HDR, Brachytherapy