Click here to

Session: Multi-Disciplinary: Biologically and Functionally-Guided Radiation Therapy [Return to Session]

Powering Predictive Treatment Planning by a Seq2seq Deep Learning Predictor

D Lee*, Y Hu, L KUO, S Alam, E Yorke, A Rimner, P Zhang, Memorial Sloan-Kettering Cancer Center, New York, NY


WE-IePD-TRACK 3-4 (Wednesday, 7/28/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]

Purpose: To develop a novel deep learning algorithm, Seq2Seq, for predicting weekly anatomical changes of locally advanced lung tumor and esophagus during definitive radiotherapy, incorporate the potential tumor shrinkage into a predictive treatment planning paradigm for adaptive radiotherapy, and improve the therapeutic ratio.

Methods: Seq2Seq starts with the tumor and esophagus observed on the planning CT to predict their geometric evolutions during radiotherapy on a weekly basis (week 1 to 6), and subsequently updates the predictions with new snapshots acquired via weekly CBCTs. Seq2Seq is composed of an encoder to parse the changes of images, and a decoder to reconstruct the prediction; both equipped with convolutional long-short term memory for sequence analysis. The longitudinal dataset (planning CT plus 6 consecutive weekly CBCTs) included 55 lung cancer patients, in which we used 11-fold cross-validation to comprehensively evaluate predictions for all patients. Dice coefficient and average Hausdorff distance (d_hd) between the predicted and actual weekly contours were calculated for accuracy evaluation. Simulating a workflow of weekly adaptation, six plans were optimized according to each weekly prediction, and made ready for weekly deployment to mitigate the clinical burden of online replanning.

Results: Seq2Seq tracks structural changes well: DICE between predicted and actual weekly tumor and esophagus were (0.82±0.10, 0.80±0.14, 0.79±0.13, 0.76±0.12, 0.74±0.13, 0.70±0.16), and (0.72±0.16, 0.73±0.11, 0.75±0.08,0.74±0.09, 0.72±0.14, 0.71±0.14), respectively, while d_hd were within 2mm. Evaluating dose to the actual weekly tumor and esophagus, a 5.3Gy reduction in esophagus mean dose while maintaining 60Gy tumor coverage was achieved with the predictive weekly plans, compared to the plan optimized using the initial tumor and esophagus alone, primarily due to noticeable tumor shrinkage during radiotherapy.

Conclusion: It is feasible to predict the longitudinal changes of tumor and esophagus with the Seq2Seq network, which could lead to improving the efficiency and effectiveness of lung adaptive radiotherapy.

Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by Varian Medical Systems.



    Image-guided Therapy, Lung, Pattern Recognition


    TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

    Contact Email