(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: Currently clinicians have no validated mechanisms to determine tumor response to radiotherapy (RT) during the planning process and before treatment delivery. Outcome prediction provides clinicians information regarding the quality of the treatment plan and helps in decision-making. The objective of this study is to use a deep-learning network to predict the treatment outcome in H&N patients using PET images.
Methods: The proposed model is a 3D hierarchically densely connected U-Net (HD U-Net) architecture. The input channels includes the patient CT scan, the pre-RT PET scan, and the RT dose plan. The output channel predicts the post-RT PET images. The patient datasets were divided into 79 training patients, 18 validation patients, and 6 test patients. The model was evaluated by computing the mean squared error between the predicted and the true post-RT PET images.
Results: The initial model has an average error of 1.75±0.36% between the predicted and the true post-RT PET images of the testing cases.
Conclusion: The current model provides a reasonable prediction of the post-RT PET images from the pre-RT PET and RT dose plan. The next step is to perform additional parameter adjustments to the model to further improve the overall accuracy.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
