(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose:
This study aims to develop a novel deep learning network, Asymmetric Network (A-Net), to predict the optimal 3D dose distribution for lung cancer patients.
Methods:
A-Net was trained and tested with 392 lung cancer cases. All patients were treated with intensity-modulated radiation therapy (IMRT) and the prescription doses were either 50Gy or 60Gy. In A-Net, the encoder and decoder were asymmetric enabling the preservation of input information and adapting the limitation of GPU memory, and squeeze and excitation (SE) units were used to improve the data-fitting ability. The dosimetric differences between clinical dose and predicted dose of the target dose coverage and organs at risk were compared. Also, we compared the prediction performance differences of different models.
Results:
The experimental results demonstrate that A-Net can accurately predict the IMRT dose distribution for lung cancer patients. The differences of PTV are within 4.495% of the prescription dose. Except for Dmax, there was no significant statistical difference in other indicators. The differences of OAR are within 5.524% of the prescription dose, except for Dmax of Spinal cord. Among A-Net, HD Unet and 3D Unet,the performance of A-net was the best.
Conclusion:
A-Net enables an optimization in the current workflow of treatment planning and improves the homogeneity of treatment plans.
Keywords
Taxonomy
Contact Email
