Click here to

Session: AI-Based Auto-Segmentation and Auto-Contouring - I [Return to Session]

Rapid Organ-At-Risk Delineation in Pancreatic CBCT for CBCT-Guided Adaptive Radiotherapy

X Dai*, Y Lei, J Janopaul-naylor, J Wynne, T Wang, J Zhou, J Roper, J Bradley, P Patel, T Liu, X Yang, Emory Univ, Atlanta, GA


TU-E-TRACK 6-5 (Tuesday, 7/27/2021) 3:30 PM - 4:30 PM [Eastern Time (GMT-4)]

Purpose: Rapid re-planning is a fundamental requirement in on-line adaptive radiotherapy. The delineation of organs-at-risk (OARs) is rather time-consuming and labor-intensive in the plan adaptation process. This study aims to develop a deep learning-based multi-organ segmentation to facilitate efficient and accurate re-planning for CBCT-guided adaptive pancreatic radiotherapy.

Methods: Considering the relatively low quality of daily CBCT acquired in the treatment machine, two-stage-in-one model was designed. In the first stage, taking CBCT as input, a cycle-consistent generative adversarial network is used to generate synthetic CT images with improved image quality. Then, in the second stage, the synthetic CT is fed into a mask-scoring regional convolutional neural network (MS-R-CNN) which is constructed by five subnetworks including a backbone, a regional proposal network (RPN), a R-CNN, a mask network, and a mask scoring network for the detection of the positions and shapes of multiple organs simultaneously. The model was trained and tested on a cohort of 40 patients. Metrics including the Dice similarity coefficient (DSC), 95th percentile Hausdorff distance (HD95), mean surface distance (MSD), and residual mean square distance (RMS) were computed to quantitatively assess the proposed method.

Results: A mean DSC of 0.92 (0.89 - 0.97), mean HD95 of 2.90 mm (1.63 mm - 4.19 mm), mean MSD of 0.89 mm (0.61 mm - 1.36 mm), and mean RMS of 1.43 mm (0.90 mm - 2.10 mm) were achieved by our proposed method across eight abdominal OARs including duodenum, large bowel, small bowel, left and right kidneys, liver, spinal cord and stomach. The contours of eight OARs can be obtained on the order of seconds once the model was trained.

Conclusion: We have proposed a novel deep learning-based multi-organ segmentation algorithm for CBCT-guided adaptive pancreatic radiotherapy. It could be implemented in the setting of pancreatic adaptive radiotherapy to rapidly contour OARs.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email