Click here to

Session: Deep learning in Treatment Planning [Return to Session]

A Deep Learning-Based Framework for Dose Prediction of Pancreatic Stereotactic Body Radiation Therapy

S Momin*, Y Lei, T Wang, J Zhang, J Roper, J Bradley, T Liu, P Patel, X Yang, Department of Radiation Oncology and Winship Cancer Institute, Emory Univ, Atlanta, GA


TH-D-TRACK 6-2 (Thursday, 7/29/2021) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Purpose: Efficacy of deep learning (DL) methods in dose prediction of pancreatic cancer is still to be fully explored. This study aims to implement and demonstrate the feasibility of a dual pyramid networks (DPNs) integrated DL model for predicting dose distributions of pancreatic stereotactic body radiation therapy (SBRT).

Methods: The proposed framework is composed of four parts: CT-only feature-pyramid-network (FPN), contour-only FPN, late-fusion-network (LFN) and an adversarial network. During each phase of the network, combination of mean absolute error, gradient difference error, histogram matching, and adversarial loss was used for supervision. The performance of proposed model was demonstrated for pancreatic cancer SBRT plans with doses prescribed between 33 and 50 Gy across as many as three planning target volumes (PTV) in five fractions. Five-fold cross validation was performed on 30 patients, and 20 patients were used as hold-out tests of trained model. Predicted plans were compared with clinical plans through clinically relevant dose-volume parameters and paired t-test. For the same cross validation and hold-out datasets, our results were compared with the ones predicted by traditional 3D U-Net network architecture.

Results: The proposed framework was able to predict 87% and 91% of clinically relevant dose parameters for cross-validation sets and holdout sets, respectively, without any significant differences (P > 0.05). The proposed model was also able to predict the intentional hotspots as feature characteristics of SBRT plans. In comparison to 3D U-Net, proposed network architecture increased prediction accuracy of 21/26 and 14/22 dose volume parameters for cross-validation and hold-out test, respectively.

Conclusion: We developed a new DPN-based method to predict dose distributions of pancreatic SBRT and demonstrated its feasibility of dose distribution prediction for both single and multiple PTVs. The predicted prior information could help save pancreatic SBRT planning times for multi-level experienced dosimetrists.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email