ProTCT deep learning method slashes radial artifacts in large-object CT scans
A new reconstruction technique for tangential CT promises clearer images of pipelines and rockets.
Researchers have proposed ProTCT, a deep learning reconstruction method that addresses a fundamental limitation of tangential computed tomography (TCT). TCT is used to image large-diameter objects such as oil pipelines and rocket segments, but its projections are inherently truncated along the detector direction, leading to radial artifacts and degraded slice quality. Existing methods struggle with ill-defined sampling conditions in the projection domain and oversmoothing in the cross-section domain.
ProTCT integrates projection quantification and a fidelity-constraint module that work together across both domains. A deep artifact-suppression network removes radial artifacts while preserving edge details. The authors also analyze sampling conditions to provide practical guidelines for TCT system design. Tests on simulated and real TCT datasets show that ProTCT outperforms prior approaches in structure restoration and detail retention, offering a practical path to high-quality imaging of large-scale industrial components.
- ProTCT combines projection quantification with a fidelity-constraint module to correct for truncated TCT projections.
- A deep artifact-suppression network operates jointly in projection and cross-section domains to remove radial artifacts and restore edges.
- Validated on both simulated and real datasets, showing significant improvements in slice quality for large-diameter objects like oil pipelines and rockets.
Why It Matters
ProTCT enables high-quality non-destructive testing of large infrastructure, improving safety and inspection accuracy for pipelines, rockets, and similar applications.