Click here to

Session: Deep Learning Image Formation and Motion Management [Return to Session]

Real-Time Liver Tumor Localization Via a Single X-Ray Projection Using Graph Neural Network and Deep Learning-Based Biomechanical Modeling (MeshRegNet-Bio)

H Shao*, T Bai, J Wang, J Chun, J Park, S Jiang, Y Zhang, The University of Texas Southwestern Medical Ctr, Dallas, TX


WE-C930-IePD-F5-1 (Wednesday, 7/13/2022) 9:30 AM - 10:00 AM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 5

Purpose: Real-time imaging provides instantaneous knowledge of patients’ anatomy to achieve the highest tumor targeting accuracy. However, only one or several x-ray projections can be acquired in real time, posing a substantial challenge to localize low-contrast liver tumors in 3D. We proposed a framework combining graph-neural-network(GNN)-based deformable registration with deep learning-based biomechanical modeling to track liver tumor in real time from a single on-board x-ray projection.

Methods: Via MeshRegNet-Bio, real-time liver motion and liver tumor localization were solved via deformable registration between a liver mesh (extracted offline from a prior CT/CBCT) and the liver features encoded in a single x-ray projection. The registration was achieved in two steps: (a) liver surface motion estimation; and (b) intra-liver motion estimation via surface motion propagation. Specifically, in step (a), a patient-specific GNN model was trained to predict a liver boundary deformation-vector-field(DVF) that deforms the prior liver surface mesh to match the liver shape variations encoded in the x-ray projection. In step (b), inspired by biomechanical modeling via finite-element-analysis, a U-Net-based deep learning model was trained to solve the intra-liver DVFs using the GNN-predicted liver surface DVFs as boundary condition. MeshRegNet-Bio was evaluated using 10 patients with liver cancer. The liver tumor localization accuracy and its robustness against different x-ray angles were evaluated.

Results: MeshRegNet-Bio allows marker-less, 3D liver tumor localization with accuracy of <2mm and latency of <120ms. Using 576 testing scenarios for each patient case, the average(±s.d.) Dice coefficients for liver tumor localization were 0.842±0.150, 0.839±0.147, and 0.838±0.137 for x-ray projections acquired at 0°, 45°, and 90°, respectively. The corresponding average 3D center-of-mass-errors were 1.61±1.58 mm, 1.62±1.50 mm, and 1.62±1.49 mm.

Conclusion: MeshRegNet-Bio allows real-time 3D liver tumor localization from a single x-ray projection, making during-treatment plan adaptation, beam gating, and MLC tracking more accessible to clinics with on-board x-ray imaging capability.

Funding Support, Disclosures, and Conflict of Interest: The study was supported by funding from the National Institutes of Health (R01CA240808, R01CA258987) and a seed grant from the Department of Radiation Oncology at the University of Texas Southwestern Medical Center.


X Rays, Registration, Localization


IM/TH- Image Registration Techniques: Machine Learning

Contact Email