MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. A Generative Adversarial Network (GAN)-Based Technique for Synthesizing Realistic Respiratory Motion in the Extended Cardiac-Torso (XCAT) Phantoms

A Generative Adversarial Network (GAN)-Based Technique for Synthesizing Realistic Respiratory Motion in the Extended Cardiac-Torso (XCAT) Phantoms

Y Chang1*, Z Jiang2, K Lafata3, Z Zhang4, P Segars5, J Cai6, F Yin7, L Ren8, (1) Duke University, Durham, NC, (2) Duke Univeristy, Durham, NC, (3) Duke University, Durham, NC, (4) Duke Univeristy, Durham, NC, AF, (5) Duke Univ, Durham, NC, (6) Hong Kong Polytechnic University, Hong Kong, HK, CN, (7) Duke University, Durham, NC, (8) Duke University Medical Center, Cary, NC

Presentations

(Monday, 7/13/2020) 1:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Room: Track 1

Purpose:
To develop a deep-learning technique to synthesize realistic respiratory motion in the XCAT phantom based on real-patient breathing patterns.

Methods:
A deep-learning framework was developed based on Bicycle-GAN to synthesize realistic respiratory motion. The network uses a 3D-CT at the end-of-inhale (EOI) phase as input, and models the deformable-vector-field (DVF) between the EOI phase and other phases, which are then applied to the input 3D-CT to generate 4D images with respiratory motion. The network was trained by matching the generated 4D images with the ground-truth 4D-CT. Analytical phantoms with predefined deformations were first used to train and test the network to investigate its feasibility. In the follow-up patient study, 65 4D-CTs of lung cancer patients were used for model training. For model testing, CTs of 8 patients and various XCAT phantoms were input to the network to generate respiratory motions. Evaluation was performed by comparing ventilation maps of the simulated and ground-truth 4D-CTs using cross-correlation and Spearman’s correlation.

Results:
In the analytical phantom study, the network successfully simulated both stretch and oscillatory deformations. In the patient study, the simulated 4D-CTs presented comparable image quality and respiratory motion compared to ground-truth 4D-CT. Comparing the simulated and ground-truth ventilation maps of the training and testing groups, the mean cross correlation achieved 0.88±0.05 and 0.75±0.06, respectively, and the mean Spearman’s correlation achieved 0.87±0.05 and 0.79±0.05, respectively. The simulated 4D-XCAT phantoms also presented realistic respiratory motion characteristics (i.e., diaphragmatic motion, rib cage expansion, etc.).

Conclusion:
Our results demonstrated the feasibility of synthesizing realistic respiratory motion in the XCAT phantoms using the proposed network. This crucial development greatly enhances the value of the XCAT phantom for various 4D imaging and therapy studies. This model could be further developed to construct patient-specific 4D phantoms and augment 4D phantoms of various realistic respiratory scenarios.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by the National Institutes of Health under Grant No. R01-CA184173 and R01-EB028324.

Keywords

Phantoms, Respiration, Deformation

Taxonomy

IM- CT: Phantoms - digital

Contact Email

Additional Links