Tensorial tomographic differential phase-contrast microscopy

Shiqi Xu, Xiang Dai, Xi Yang, Kevin C. Zhou, Kanghyun Kim, Vinayak Pathak, Carolyn Glass, Roarke Horstmeyer

Introduction

Most biological specimens are "phase objects," meaning they are nearly transparent and provide little contrast under standard brightfield microscopy. To visualize these structures without toxic chemical stains, researchers utilize Quantitative Phase Imaging (QPI). However, biological samples are not just characterized by their refractive index; many—such as muscle fibers, cell membranes, and protein aggregates—are also anisotropic. This means they affect light differently depending on its polarization state, a property known as birefringence.

Existing methods for 3D mapping of both phase and anisotropy typically rely on complex holographic systems or slow scanning mechanisms. Holographic systems require expensive lasers and are prone to speckle noise, while scanning systems can be slow and potentially damaging to live specimens. To address these challenges, researchers have developed Tensorial Tomographic Differential Phase-Contrast ($T^2DPC$) microscopy, a hardware-simple, computationally robust method for label-free 3D vectorial imaging.

Methodology

Hardware Configuration

The $T^2DPC$ system is designed to be retrofitted onto a standard inverted microscope, making it highly accessible. The setup consists of two primary components:

  1. Programmable LED Matrix: Replacing the traditional condenser, this array allows for precise control over illumination angles. By hitting the sample from various directions, the system captures the diverse scattering information required for 3D reconstruction.
  2. Polarization-Sensitive CMOS Camera: This sensor utilizes a "super-pixel" array containing four linear polarizers (oriented at 0°, 45°, 90°, and 135°). This allows the system to capture the vectorial state of light in a single snapshot, eliminating the need for rotating filters.

The Tensorial Framework

Traditional Differential Phase Contrast (DPC) treats a sample as a scalar quantity (refractive index only). $T^2DPC$ extends this by treating the sample as a permittivity tensor.

  • Forward Model: The researchers derived a mathematical model based on the first Born approximation to describe how a 3D anisotropic sample interacts with partially coherent light.
  • Inverse Problem: To reconstruct the 3D volume, the team employs an iterative optimization approach. Using the Adam optimizer in PyTorch, the system solves for the scattering potential components that best fit the captured intensity data. Total Variation (TV) regularization is applied to ensure the resulting images are sharp and resilient to noise.

Experimental Validation

The method was validated through three distinct experimental stages:

  1. Calibration with MSU Crystals: Monosodium Urate (MSU) crystals were used to verify the system's ability to accurately determine 3D molecular orientation.
  2. Quantitative Benchmarking: By imaging 3-μm polystyrene microspheres immersed in oil, the researchers confirmed that the reconstructed birefringence values matched known physical constants. The system successfully resolved two layers of beads separated by only 8.8 μm, demonstrating superior axial resolution compared to standard focal stacks.
  3. Clinical Application: The system was used to image Congo red-stained human heart tissue. $T^2DPC$ successfully visualized the specific protein fiber alignments characteristic of cardiac amyloidosis. This suggests that the method could serve as a powerful tool for pathologists to examine the 3D molecular architecture of biopsies without complex interferometric setups.

Discussion and Future Outlook

$T^2DPC$ offers a significant "middle way" in microscopy, providing the depth-sectioning capabilities of 3D tomography with the hardware simplicity of LED-based imaging. By using partially coherent light, it avoids the speckle noise inherent in laser-based systems, resulting in cleaner, more biologically relevant images.

However, some limitations remain. The current model operates within the paraxial limit, meaning its accuracy decreases at very high illumination angles. Additionally, the current implementation focuses primarily on "in-plane" (x-y) anisotropy rather than full 3D (z-axis) molecular tilt. Future work will focus on improving reconstruction speeds—potentially through deep learning or variable splitting methods—and extending the model to handle thicker, more highly scattering samples.

In conclusion, $T^2DPC$ represents a major step forward in accessible, high-dimensional microscopy, offering a practical path for researchers to map the complex vectorial nature of biological life in three dimensions.

Summary updated

BibTeX
@inproceedings{xu2022tensorial,
  author = {Shiqi Xu and Xiang Dai and Xi Yang and Kevin C. Zhou and Kanghyun Kim and Vinayak Pathak and Carolyn Glass and Roarke Horstmeyer},
  title = {{Tensorial tomographic differential phase-contrast microscopy}},
  booktitle = {ICCP},
  year = {2022},
  doi = {10.1109/iccp54855.2022.9887674},
}
Figure