Reconstructing moving scenes with computed tomography (4DCT) is a challenging and ill-posed problem with important applications in industrial and medical settings. Dynamic computed tomography (DCT) refers to image reconstruction of moving or non-rigid objects over time while x-ray projections are acquired over a range of angles. Although 4DCT reconstruction is widely applicable to the study of object deformation and dynamics in a number of industrial and clinical applications, it has been a long-standing challenge due to the complexity of the x-ray measurement capturing both spatial and temporal features with the limited data sampling.
The essence of this invention is a method that couples network architecture using neural implicit representations coupled with a novel parametric motion field to perform limited angle 4D-CT reconstruction of deforming scenes. To enable the reconstruction of the scene with high dynamics, the inventors developed a novel method for dynamic 4DCT reconstruction that leverages implicit neural representations with a parametric motion field to reconstruct dynamic scenes as time-varying sequence of 3D volumes. The methods have been demonstrated in experiments that reconstruct dynamic scenes with deformable and periodic motion on physically simulated synthetic data and real data.
Publication:
Reed, A.W., Kim, H., Anirudh, R., Mohan, K.A., Champley, K.M., Kang, J., & Jayasuriya, S. (2021). Dynamic CT Reconstruction from Limited Views with Implicit Neural Representations and Parametric Motion Fields. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2238-2248.
The principal advantages of this invention are:
- This method is an end-to-end optimization approach without the need for any training data;
- This method eliminates the need for fast CT scanners in use cases where the object or scene being scanned is fast moving;
- The hierarchical coarse-to-fine procedure to estimate the motion field enables recovering fine details of the motion scene without suffering from severe artifacts due to poor convergence of the optimization.
CT/CAT (computerized axial tomography) scanner systems
Current stage of technology development: TR-2
LLNL has patent(s) on this invention.
U.S. Patent No. 11,741,643 Reconstruction of dynamic scenes based on differences between collected view and synthesized view published 8/29/2023