Room: AAPM ePoster Library
Dose grid size is one of factors that affect accuracy of dose calculation. Although use of small grid (less than 2 mm) can improve the accuracy, it is not typically used in clinic due to hard computation. The purpose of this study is to propose a deep learning strategy to predict the dose of small grid from those of large grid with less time.
Our deep learning architecture consisted of two networks: (1) feature-learning and (2) super-resolution networks. Each network was independently trained using 2D slice-by-slice manner. Doses of 1- and 3-mm grid for 20 patients (training: 16, test: 4), which were calculated by VMAT prostate plan (prescription: 78 Gy) and AXB algorithm, were used. The doses of 1-mm were downsampled to 3-mm to organize two training data pairs: (1) dose of 3-mm/downsampled dose and (2) downsampled dose/dose of 1-mm. The first and second pairs were used to train the feature-learning and super-resolution networks, respectively. The trained networks were connected by using output of the feature-learning network as input of the super-resolution network. Predicted doses by the network were compared with doses of 1-mm using dose-volume histogram (DVH) and dice similarity coefficient (DSC).
The DVH of planning-target-volume (PTV) for the prediction were visually more similar to those for dose of 1-mm than 3-mm grid. Mean/maximum doses in PTV for the prediction were similar to those for 1- and 3-mm. Average minimum dose differences were 1.9±0.4% of the prescription (prediction vs. 1-mm) and 7.7±7.4% (3-mm vs. 1-mm). The DSC between the prediction and doses of 1-mm was more close to 1 compared to those between 3-mm and 1-mm.
Proposed method predicted dose of small grid from those of large grid with less time. The predicted doses were comparable to calculated dose with small grid.
Funding Support, Disclosures, and Conflict of Interest: Funding: This research was supported by Mid-career Researcher Program (No. 2018R1A2B2005343) through the National Research Foundation of Korea funded by the Ministry of Science and ICT. Disclosures and Conflict of Interest: The authors have no conflicts of interest.