Abstract
Tomography has had an important impact on the physical, biological, and medical sciences. To date, most tomographic applications have been focused on 3D scalar reconstructions. However, in some crucial applications, vector tomography is required to reconstruct 3D vector fields such as the electric and magnetic fields. Over the years, several vector tomography methods have been developed. Here, we present the mathematical foundation and algorithmic implementation of REal Space Iterative REconstruction for Vector tomography, termed RESIRE-V. RESIRE-V uses multiple tilt series of projections and iterates between the projections and a 3D reconstruction. Each iteration consists of a forward step using the Radon transform and a backward step using its transpose, then updates the object via gradient descent. Incorporating with a 3D support constraint, the algorithm iteratively minimizes an error metric, defined as the difference between the measured and calculated projections. The algorithm can also be used to refine the tilt angles and further improve the 3D reconstruction. To validate RESIRE-V, we first apply it to a simulated data set of the 3D magnetization vector field, consisting of two orthogonal tilt series, each with a missing wedge. Our quantitative analysis shows that the three components of the reconstructed magnetization vector field agree well with the ground-truth counterparts. We then use RESIRE-V to reconstruct the 3D magnetization vector field of a ferromagnetic meta-lattice consisting of three tilt series. Our 3D vector reconstruction reveals the existence of topological magnetic defects with positive and negative charges. We expect that RESIRE-V can be incorporated into different imaging modalities as a general vector tomography method.
Introduction
While 3D imaging is a cornerstone of modern science—from medical CT scans to atomic-resolution electron microscopy—most applications are limited to scalar tomography. Scalar tomography measures a single value at every point in space, such as density or absorption. However, many fundamental physical forces, such as electricity and magnetism, are vector fields. In these cases, every voxel (3D pixel) possesses both a magnitude and a specific direction.
Reconstructing these 3D vector fields is significantly more challenging than scalar fields. Traditional methods often struggle with "missing wedge" artifacts, where the sample cannot be rotated through a full 180 degrees, and high levels of experimental noise. Furthermore, the mathematical relationship between a 3D vector field and its 2D projections is complex; the $x, y,$ and $z$ components are often coupled in the projection data, making accurate separation difficult.
The RESIRE-V Algorithm
To address these challenges, we developed RESIRE-V (REal Space Iterative REconstruction for Vector tomography). The algorithm utilizes an iterative loop to refine the 3D model based on experimental projections.
As shown in the flowchart above, the process consists of four primary steps per iteration:
- Forward Projection: The current 3D model is projected using the Radon transform to generate calculated 2D projections.
- Residual Calculation: The algorithm calculates the difference (error) between these calculated projections and the actual measured data.
- Gradient/Back-projection: The error is "back-projected" into 3D space using the transpose of the Radon transform.
- Object Update: The 3D model is updated via gradient descent to minimize the error metric.
A key innovation in RESIRE-V is the integration of a 3D support constraint—a mask defining the object's boundaries. This constraint acts as a powerful regularizer, enabling the algorithm to fill in gaps caused by missing data. Additionally, RESIRE-V can iteratively refine experimental tilt angles, correcting for small alignment errors common in high-resolution microscopy.
Results and Validation
The efficacy of RESIRE-V was validated using both simulated and experimental datasets.
Simulated Validation
We first applied the algorithm to a simulated $100 \times 100 \times 100$ pixel 3D magnetization field. This allowed for a direct comparison between the reconstruction and a known "ground truth." Despite the presence of noise and missing wedge artifacts, the reconstructed vector components showed exceptional fidelity. Quantitative analysis yielded normalized cross-correlation coefficients of 94.1% for $M_x$, 93.8% for $M_y$, and 99.1% for $M_z$.
Experimental Application
RESIRE-V was further tested on a real-world ferromagnetic meta-lattice composed of 60 nm silica nanoparticles infiltrated with nickel. Data was acquired via soft X-ray magnetic circular dichroism (XMCD) ptychography at the Advanced Light Source. The algorithm successfully reconstructed the 3D magnetization with a spatial resolution of 10 nm. Notably, the reconstruction revealed the existence of topological magnetic defects—effectively magnetic monopoles—within the nanostructure. These defects are critical for the development of next-generation spintronic and magnetic storage devices.
Discussion and Conclusion
Our analysis highlights several critical insights into vector tomography. First, the $z$-component (aligned with the beam) is typically reconstructed with higher quality than the $x$ and $y$ components due to its reduced coupling in standard projection geometries. Second, while scalar tomography can often function without a support mask, the support constraint is mathematically necessary for vector fields to achieve a unique solution when data is under-determined.
The primary limitation of RESIRE-V is the experimental requirement for multiple tilt series (rotating the sample along different axes) to fully decouple the 3D vector components. However, the high-fidelity results and the ability to identify nanoscale topological defects demonstrate its significant potential. Future work will focus on integrating RESIRE-V with other modalities, such as Lorentz transmission electron microscopy, and optimizing the algorithm for GPU acceleration to handle larger datasets. To encourage adoption, the MATLAB source code and datasets are available on GitHub.
Summary updated