🔢 mathematics

Geodesic X-ray transform and streaking artifacts on simple surfaces or on spaces of constant curvature

This paper investigates the formation of streaking artifacts in geodesic X-ray transforms on nontrapping simple compact Riemannian manifolds, demonstrating that in two-dimensional or constant curvature spaces, these artifacts arise from the propagation of conormal singularities along common tangent geodesics between strictly convex metal regions.

Original authors: Hiroyuki Chihara

Published 2026-02-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hiroyuki Chihara

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Why Do CT Scanners Make "Streaks"?

Imagine you are getting a CT scan of your mouth. The machine shoots X-rays through your jaw to create a picture of your teeth and bones. Usually, this works perfectly. But if you have a metal dental implant (like a silver filling or a screw), the picture gets ruined. Instead of a clean image, you see bright, dark lines shooting out from the metal like lightning bolts. These are called streaking artifacts.

This paper asks a deep mathematical question: Why do these streaks appear exactly where they do, and what shape do they take if we aren't on a flat piece of paper (like a normal CT scan) but on a curved surface (like a sphere or a curved piece of tissue)?

The Cast of Characters

The X-Ray Beam: Think of this as a laser pointer. In a perfect world, it travels in a straight line.
The Metal Regions (The "Bad Guys"): These are the dental implants. They are like shiny, hard islands in a sea of soft tissue.
The "Beam Hardening" Effect: Real X-rays aren't just one color; they are a mix of energies. When they hit metal, the "weak" rays get absorbed, and only the "strong" rays get through. This changes the data the machine collects, causing the math to get confused and create the streaks.
The Common Tangent: This is the most important concept. Imagine two round rocks on the ground. If you roll a marble so it just barely touches both rocks without going through them, that path is a "common tangent."

The Core Discovery: The "Laser Line" Rule

The paper proves a specific rule about where these streaks appear.

The Analogy of the Tightrope:
Imagine two strictly convex metal islands (like two smooth, round boulders). If you try to stretch a tightrope between them so that it touches both but doesn't cut through either, you can only do this along specific lines.

On a Flat Table (Euclidean Space): If you have two round metal implants, the streaks appear along the straight lines that touch both of them.
On a Curved World (The Paper's Innovation): The author asks, "What if the world isn't flat? What if we are on a sphere or a curved surface?"

The paper shows that even on these curved surfaces, the streaks still appear along the geodesics (the curved equivalent of straight lines) that act as "common tangents" to the metal regions.

The "Magic Trick" of the Math

The author had to solve a very difficult puzzle. To understand how the "noise" (the streaks) travels from the metal to the final image, he had to track how tiny ripples in the data move along these curved paths.

He used a strong assumption to make the math work: The world is either 2D (like a flat sheet) or has "constant curvature" (like a perfect sphere or a saddle shape).

Why is this assumption important?
Imagine trying to walk a straight line on a bumpy, irregular hill. It's a nightmare to predict where you'll end up. But if you are walking on a perfect sphere, the rules of the road are simple and uniform. The author used this "perfect world" rule to prove that the "noise" (the streaks) behaves predictably. It travels along the common tangent lines just like a laser beam bouncing between mirrors.

The "Shadow" Metaphor

Think of the metal implants as two large, round boulders in a foggy valley.

The Fog: This is the normal tissue.
The Sun: This is the X-ray beam.
The Streaks: These are the shadows cast by the boulders.

In a flat world, the shadows of two boulders touch along a straight line where the sun's rays graze both of them.
In this paper, the author proves that even if the valley is curved (like the surface of a planet), the "shadows" (the streaking artifacts) still align perfectly along the lines that graze both boulders.

The "Why Should We Care?" Conclusion

You might ask, "Who cares about curved surfaces? CT scans are on flat bodies."

Medical Imaging: While human bodies are mostly flat locally, understanding the math on curved surfaces helps us understand the fundamental limits of the technology. It clarifies why the artifacts happen, which helps engineers design better algorithms to remove them.
Mathematical Beauty: The paper reveals that the "streaks" aren't random errors. They are a geometric necessity. They are the physical manifestation of the "common tangent lines" of the metal objects. The paper clarifies that these artifacts are essentially the "echo" of the geometry of the metal regions traveling along the shortest paths (geodesics) of the space.

Summary in One Sentence

This paper proves that the annoying "streaks" you see in CT scans near metal implants are actually the mathematical shadows of the metal objects, traveling along the specific curved paths that just barely touch both objects, a rule that holds true even if the space itself is curved.

1. Problem Statement and Motivation

The paper addresses the mathematical origin of streaking artifacts in Computed Tomography (CT) imaging, specifically caused by the beam hardening effect.

Physical Context: In standard CT, X-rays are assumed to be monochromatic. However, real X-ray beams have a spectrum of energies. When passing through high-density materials (metal implants, bones), lower-energy photons are absorbed more readily than higher-energy ones, causing the beam's average energy to increase ("harden"). This non-linear attenuation leads to reconstruction errors.
Mathematical Formulation: The artifact is modeled as a non-linear phenomenon where the measured data $P$ deviates from the linear X-ray transform $Rf$. Specifically, if $D$ represents metal regions (strictly convex domains), the error term involves powers of the X-ray transform of the characteristic function of $D$ , i.e., $(R\mathbf{1}_D)^{2l}$ .
The Core Question: Previous work (Park, Choi, Seo; Palacios, Uhlmann, Wang) established that in Euclidean space ( $\mathbb{R}^n$ ), these artifacts manifest as singularities propagating along the common tangent lines of the metal regions. However, the geometric intuition behind why and how these singularities propagate, particularly regarding the role of curvature and the specific nature of the "common tangent" in non-Euclidean settings, remained obscure.
Goal: To clarify the geometric meaning of these artifacts by studying the geodesic X-ray transform on general Riemannian manifolds, specifically under the assumption that the manifold is either 2-dimensional or a space of constant curvature.

2. Methodology

The author employs Microlocal Analysis and the theory of Fourier Integral Operators (FIOs) to analyze the propagation of singularities.

Geometric Setting:
- The domain is a compact Riemannian manifold $(M, g)$ with a smooth boundary.
- Assumption (A0): $(M, g)$ is a simple manifold (no conjugate points, strictly convex boundary, non-trapping).
- Assumption (A1): $(M, g)$ is either 2-dimensional or a space of constant curvature. This is a crucial, seemingly restrictive assumption that simplifies the behavior of Jacobi fields.
Analytical Tools:
- Geodesic X-ray Transform ( $X$ ): Defined as the integral of a function (or half-density) along geodesics.
- Adjoint ( $X^T$ ) and Normal Operator: The reconstruction is modeled via $X^T X$ .
- Lagrangian Distributions: The singularities of the data are analyzed as conormal distributions associated with Lagrangian submanifolds in the cotangent bundle.
- Intersection Calculus: The core of the proof involves analyzing the composition of canonical relations associated with $X$ and $X^T$ with the singular supports of the non-linear terms (products of transforms).
Key Technical Insight (Jacobi Fields):
- Under Assumption (A1), the author proves that Jacobi fields along geodesics decompose into the product of a scalar function and a parallel transport of a tangent vector.
- This decomposition allows for an explicit characterization of the canonical relations and the interaction of singularities at common tangent geodesics, which is difficult to achieve in general Euclidean settings with complex curvature.

3. Key Contributions and Results

A. Geometric Characterization of Singularities

The paper rigorously proves that streaking artifacts arise from the propagation of conormal singularities located on the boundaries of metal regions ( $\partial D_j$ ) along common tangent geodesics.

The Metal Region: Modeled as a disjoint union of strictly convex domains $D = \bigcup D_j$ .
The Artifact Set ( $L$ ): The singularities of the reconstructed error term $f_{MA}$ are supported on a set $L$ of hypersurfaces. These hypersurfaces are generated by the union of all common tangent geodesics between pairs of metal regions $D_j$ and $D_k$ .

B. Classification of Tangent Cases

The author distinguishes between two types of common tangents based on the relationship between the normal vectors at the tangent points:

Case C1 (Parallel Normals): The parallel transport of the normal vector at one tangent point aligns with the normal vector at the other (or its negative). This corresponds to the "common tangent hyperplane" case in Euclidean space.
- Result: Singularities propagate along these geodesics, creating the visible streaking artifacts.
Case C2 (Non-Parallel Normals): The transported normal does not align with the target normal.
- Result: Singularities do not propagate in a way that creates artifacts in the reconstructed image (they are "filtered out" by the inversion process).

C. Main Theorem (Theorem 5.1)

The main result characterizes the regularity of the artifact term $f_{MA}$ :
$f_{MA} \in I^{-1/2 - 3n/4}(N^*(L) \setminus 0)$

$f_{MA}$ is a paired Lagrangian distribution (specifically a conormal distribution) associated with the conormal bundle of the set $L$ (the union of common tangent hypersurfaces).
The principal symbol of the leading term (quadratic in the metal density) is non-vanishing, confirming that these artifacts are the dominant error source.
Higher-order terms in the beam hardening expansion do not introduce new singular supports; they only affect the amplitude of the existing artifacts.

D. Clarification of Euclidean Results

The paper demonstrates that the previously known results in $\mathbb{R}^n$ (specifically the distinction between Case B1 and B2 in Euclidean space) are natural consequences of the geometry of constant curvature or 2D manifolds. The "strange" assumption (A1) is shown to be the precise condition required to make the parallel transport of singularities tractable, thereby revealing the geometric mechanism that was previously "imperceptible" in the Euclidean analysis.

4. Significance

Theoretical Clarity: The paper moves beyond the computational complexity of Euclidean microlocal analysis to provide a clear geometric interpretation of CT artifacts. It explicitly links the existence of streaks to the geometry of Jacobi fields and parallel transport.
Generalization: It extends the theory of streaking artifacts from flat space to curved spaces, which is relevant for medical imaging in non-Euclidean contexts (e.g., imaging on curved surfaces or in theoretical models of spacetime) and provides a unified framework for understanding the phenomenon.
Methodological Advancement: The use of the specific structure of Jacobi fields in constant curvature/2D manifolds to solve the intersection calculus of canonical relations offers a powerful new technique for analyzing non-linear inverse problems in geometric analysis.
Practical Implication: By identifying exactly which geometric configurations (Case C1 vs. C2) lead to artifacts, the work provides a theoretical basis for understanding why certain metal configurations cause severe streaking while others do not, potentially guiding better artifact reduction algorithms.

In summary, Chihara's work successfully demystifies the geometric origin of streaking artifacts in CT by leveraging the specific properties of geodesics in simple, constant-curvature manifolds, proving that these artifacts are the direct result of conormal singularities propagating along common tangent geodesics where the normal vectors align via parallel transport.