MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

CBCT-Based Synthetic CT Using Deep Learning for Pancreatic Adaptive Radiotherapy

Y Liu*, Y Lei, T Wang, P Patel, T Liu, W Curran, X Yang, Emory University, Atlanta, GA

Presentations

(Sunday, 7/14/2019) 2:00 PM - 3:00 PM

Room: 304ABC

Purpose: Current clinical application of CBCT is limited to patient setup due to its imaging artifacts. In this study, we developed a deep-learning-based approach to improving pancreatic CBCT image quality and Hounsfield unit accuracy for potential extended clinical use in pancreatic adaptive radiotherapy.

Methods: Due to the patient motion during image acquisition, perfectly-registered abdominal CT-CBCT pairs are difficult to obtain. To minimize the mismatch-induced prediction error, a novel residual cycle generative-adversarial-networks (GAN) was employed by supervising both forward and backward projections to achieve one-to-one mapping. Four generators with residual blocks and two discriminators were used in the training networks to enhance the similarity between CT and corrected CBCT (synthetic-CT). Through the residual connection in each residual block, an input bypasses the hidden layers from which specific differences between CBCT and CT are learned. A novel compound loss function was employed to effectively differentiate the structure boundaries. A cohort of 17 pancreas patients with co-registered abdominal CT-CBCT pairs was used to evaluate the algorithm by leave-one-out cross-validation. Initial CT-based plans were transferred to CBCT and synthetic-CT for dosimetric comparison.

Results: Imaging endpoints including mean absolute error (MAE) and peak signal-to-noise ratio (PSNR) were 22.96±7.72 HU and 29.45±4.67 dB. No significant differences (p>0.05) were observed in the PTV and OAR DVH metrics including D10, D50, D95, Dmin, Dmean, and Dmax between the CT- and synthetic-CT-based plans, while significant differences (p<0.05) were found between the CT- and CBCT-based plans.

Conclusion: This work generated abdominal synthetic from routine CBCT images based on our proposed residual-cycle-GAN algorithm. The image similarity and dosimetric agreement between the CT and synthetic-CT warrant further development of pancreatic CBCT-based adaptive radiotherapy to update structural contours and dose calculations. The approach can potentially increase treatment precision and thus minimize gastrointestinal toxicity.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email