Related Experiment Video
Updated: Nov 26, 2025
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
Dose Super-Resolution in Prostate Volumetric Modulated Arc Therapy Using Cascaded Deep Learning Networks
Dong-Seok Shin1,2, Kyeong-Hyeon Kim1,2, Sang-Won Kang1,2
1Department of Biomedical Engineering, Department of Biomedicine and Health Sciences, College of Medicine, The Catholic University of Korea, Seoul, South Korea.
This study introduces a cascaded network model for generating high-resolution radiation therapy doses from low-resolution data, significantly reducing computation time. The model accurately predicts doses, showing improved agreement with baseline values and higher gamma passing rates.
Area of Science:
- Medical Physics
- Radiation Oncology
- Artificial Intelligence in Healthcare
Background:
- Accurate dose calculation is crucial in radiation therapy planning.
- Generating high-resolution dose distributions can be computationally intensive.
- Low-resolution doses may lack the necessary detail for precise treatment evaluation.
Purpose of the Study:
- To develop and validate a cascaded network model for generating high-resolution (1 mm grid) dose distributions from low-resolution (≥3 mm grids) data.
- To significantly reduce the computation time required for high-resolution dose prediction.
- To assess the accuracy and dosimetric agreement of the predicted high-resolution doses compared to baseline high-resolution doses.
Main Methods:
- A cascaded network model comprising a hierarchically densely connected U-net (HD U-net) and a residual dense network (RDN) was employed.
- The model was trained slice-by-slice using dose distributions from volumetric modulated arc therapy plans for 73 prostate cancer patients.
- Performance was evaluated using spatial/dosimetric parameters and gamma analysis (2%/2 mm criterion) against baseline high-resolution doses.
Main Results:
- The model achieved an average computation time of <0.02 seconds per axial dose plane.
- Predicted doses showed improved Dice Similarity Coefficient values and closer agreement with baseline dosimetric parameters compared to low-resolution doses.
- Gamma passing rates for predicted high-resolution doses were higher than for low-resolution doses, with no significant differences observed for most parameters (p > 0.05).
Conclusions:
- The proposed cascaded network model accurately predicts high-resolution doses within the same dose calculation algorithm.
- The model's ability to use only dose data simplifies its application.
- This approach offers a convenient and efficient method for dose super-resolution in radiation therapy.
