Abstract: Double Your Views: Exploiting Symmetry in Transmission Imaging

Alexander Preuhs, Andreas K. Maier, Michael Manhart, Javad Fotouhi, Nassir Navab, Mathias UnberathBildverarbeitung für die Medizin 2019 · DOI

Introduction

In C-arm Cone-Beam Computed Tomography (CBCT), often used during surgery, patient motion is a primary cause of image degradation. Even slight movements result in blurry or streaky 3D reconstructions. While mathematical "Consistency Conditions" are typically used to correct these errors, standard circular scan trajectories suffer from a mathematical "blind spot": they cannot robustly detect motion occurring within the plane of rotation. Overcoming this usually requires complex, non-planar scan paths that are often impractical in clinical settings.

Methodology: Symmetric View Augmentation

This research proposes a clever workaround: Symmetric View Augmentation. By exploiting the inherent symmetry of anatomical structures like the human head, researchers can mathematically mirror X-ray projections to create virtual data points.

1. The Virtual Camera Concept

If an object is symmetric, an X-ray taken from one side is theoretically identical to a mirrored X-ray from the opposite side. By identifying the 3D symmetry plane, every real image captured provides a second "virtual" image for free. This effectively doubles the views available for consistency checks.

Symmetric View Augmentation: A real projection (P) and its mirrored virtual counterpart (PF) are used to establish consistency across the symmetry plane (F).

2. Symmetry Plane Estimation and Consistency

The researchers use Epipolar Consistency Conditions (ECC)—based on Grangeat’s theorem—to find the 3D symmetry plane. They search for the specific plane that maximizes the mathematical consistency between the real views and the mirrored virtual views.

Epipolar Consistency: Standard geometry used to compare X-ray projections from different angles to detect motion.

3. Creating the "X-Trajectory"

By mirroring the circular path of the X-ray source across an oblique (tilted) symmetry plane, a virtual trajectory is created. Combined, the real and virtual paths form an "X-shape." This combined trajectory is "Tuy-complete," meaning it captures enough diverse angular information to detect motion in all directions, successfully eliminating the previous "blind spot" for in-plane motion.

Key Insights

  • Symmetry as Redundancy: Symmetry is treated as a source of redundant information that can be traded for geometric stability and motion sensitivity.
  • Breaking Planarity: The critical requirement is that the anatomy (e.g., the head) must be slightly tilted relative to the scanner's rotation. This tilt ensures the mirrored virtual trajectory exists in a different plane than the real one.
  • Robustness to Asymmetry: The human head is symmetric enough for this method to work. The algorithm proved robust even though biological symmetry is never perfect.

Results and Validation

The method was tested using both synthetic phantoms and a realistic anthropomorphic head phantom scanned with a robotic C-arm system.

  • High Precision: In synthetic tests, the symmetry plane was estimated with near-perfect accuracy.
  • Enhanced Sensitivity: The proposed "Mirrored Epipolar Consistency" (MEC) was significantly more sensitive to in-plane rotations and translations than standard methods.
  • 10x Accuracy Boost: In head phantom trials, the mirrored views were approximately 10 times more accurate at identifying the true motion state compared to using standard circular scan data alone.

Conclusion

Exploiting anatomical symmetry provides a powerful mechanism for improving the reliability of motion correction in CBCT. While currently focused on rigid motion in symmetric structures like the head or pelvis, future work aims to extend this to non-rigid motion (e.g., swallowing) and other artifacts like beam hardening or data truncation.

Summary updated

BibTeX
@inproceedings{preuhs2019abstract,
  author = {Alexander Preuhs and Andreas K. Maier and Michael Manhart and Javad Fotouhi and Nassir Navab and Mathias Unberath},
  title = {{Abstract: Double Your Views: Exploiting Symmetry in Transmission Imaging}},
  booktitle = {Bildverarbeitung für die Medizin},
  year = {2019},
  doi = {10.1007/978-3-658-25326-4\_36},
}
Figure