Abstract
The method of common lines is a cornerstone of reconstruction in cryogenic electron microscopy (cryo-EM), enabling the determination of an object's relative orientation from tomographic projections. This paper extends this concept to optical diffraction tomography (ODT). By leveraging the Fourier diffraction theorem, we demonstrate that rigid motions—rotations and translations—can be recovered by identifying "common circles" in Fourier-transformed data. We propose two methods: a direct parameterization approach and an infinitesimal version for smooth motions. Numerical simulations confirm that these methods provide high-precision motion tracking and high-fidelity 3D reconstructions.
Introduction: The Challenge of Diffraction
In traditional X-ray Computed Tomography (CT), imaging assumes that light travels in straight lines. However, when imaging micro-scale biological specimens, light diffracts, necessitating a wave-based approach known as Optical Diffraction Tomography (ODT). A significant hurdle in ODT arises when the target object moves unpredictably during scanning, such as a cell suspended in optical tweezers. Without knowing the object's precise orientation for each frame, accurate internal reconstruction is impossible.
While the "method of common lines" solves this in cryo-EM, it relies on the Fourier Slice Theorem, where data maps to flat planes. In ODT, the Fourier Diffraction Theorem dictates that data maps to curved hemispheres. This geometric shift renders traditional common line methods inapplicable, requiring a new "common circle" framework.
The Common Circle Framework
The core innovation of this work is the exploitation of the intersection of hemispheres in 3D Fourier space. According to the Fourier Diffraction Theorem, each measurement in ODT corresponds to a hemisphere. When an object rotates between two measurements, these two hemispheres intersect along a circular arc—a "common circle."
By comparing data values along these arcs, the relative Euler angles of the object's rotation can be determined. Unlike cryo-EM, which typically requires three images to resolve orientation, the hemispherical geometry of ODT allows for unique rotation recovery using only two images.
Methods for Motion Recovery
The authors introduce two primary strategies based on the Fourier Diffraction Theorem:
- Direct Common Circle Method: This approach searches for the relative rotation that minimizes the difference between datasets along the intersecting circular arcs.
- Infinitesimal Common Circle Method: For smooth, continuous motion, this method utilizes first-order derivatives in time and space to calculate instantaneous angular velocity. Integrating this velocity over time yields the full rotation path.
Furthermore, the paper demonstrates that ODT is uniquely sensitive to 3D translation. In standard ray-transform imaging, moving an object along the beam direction does not change the projection. In ODT, such motion alters the phase of the scattered wave, allowing for the recovery of the full 3D translation vector, provided the scattering potential is real-valued.
Results and Discussion
Numerical proof-of-concept tests using simulated phantoms (such as a biological cell model and the Shepp-Logan phantom) demonstrated the robustness of these methods. For a sequence of 640 time steps with complex, time-dependent rotation and translation, the relative error in rotation matrices remained between $10^{-3}$ and $10^{-5}$.
Reconstructions using the recovered parameters achieved high fidelity, with Peak Signal-to-Noise Ratios (PSNR) exceeding 30 dB. Computationally, the process is efficient, taking approximately 40 seconds for 640 frames on standard hardware.
Conclusion and Future Directions
The common circle method provides a mathematically rigorous and computationally efficient solution for motion tracking in ODT. While currently based on the Born approximation (weak scattering) and real-valued potentials, future work will focus on extending the framework to complex-valued potentials (absorbing objects) and incorporating phase retrieval for experimental data where only intensity is measured. This paves the way for high-resolution imaging of live cells in dynamic environments.
Summary updated