Reflection-mode diffraction tomography of multiple-scattering samples on a reflective substrate from intensity images

Tongyu Li, Jiabei Zhu, Yi Shen, Lei TianOptica 2025 · OpenReview

TL;DR

This paper introduces a novel reflection-mode diffraction tomography (DT) technique that simultaneously reconstructs both forward and backward scattering information from intensity-only images. By employing a modified Born series forward model, specialized boundary conditions, and an adjoint method for efficient gradient computation, this technique achieves high-resolution 3D refractive index mapping of scattering samples, even those obscured by other structures.

What Problem Does This Paper Address?

Traditional diffraction tomography (DT) primarily relies on transmission-mode measurements, which are limited in axial resolution and struggle to capture information from lateral interfaces. While optical coherence tomography (OCT) excels at capturing interfacial information through reflection-mode measurements, it lacks the ability to quantitatively reconstruct the refractive index (RI) distribution. This leaves a gap for applications requiring detailed RI mapping in reflection mode, especially for complex, scattering samples. The authors aim to bridge this gap by developing a reflection-mode DT technique that can simultaneously recover both forward scattering (FS) and backward scattering (BS) information, providing a more complete picture of a sample's 3D RI distribution.

Methods

The core of the technique lies in a novel reconstruction algorithm that integrates several key components:

  • Modified Born Series (MBS): This serves as the forward model, capable of accurately simulating multiple scattering interactions, which is crucial for scattering samples. It addresses limitations of traditional Born series by employing a specific strategy for the background medium's refractive index to ensure stability.
  • Bloch Boundary Condition (BC) and Perfect Electric Conductor (PEC) BC: These boundary conditions are essential for accurately modeling the experimental setup. Bloch BC handles oblique incidence, while PEC BC simulates the strong reflections from the highly reflective substrate, a key element of the reflection-mode approach. A novel padding-free implementation of Bloch BC is also introduced to reduce computational overhead.
  • Adjoint Method (AM): For efficient gradient computation in the inverse scattering problem, the authors utilize the adjoint method. This approach significantly reduces memory requirements compared to traditional backpropagation methods, making it more practical for complex simulations.
  • Intensity-Only Measurements: The technique operates on intensity images captured by a microscope, eliminating the need for interferometric setups that are sensitive to environmental vibrations.

The overall workflow involves iteratively updating the RI distribution by performing forward simulations (MBS) to predict intensity and then using the adjoint method to compute the gradient of a loss function (difference between simulated and measured intensities) to guide the update.

Datasets

The paper validates its technique through both simulated data and experimental measurements on various samples.

  • Simulated Data:
    • Simulations of a 3D sphere near a PEC boundary were performed and compared against FDTD results.
  • Simulations of a 2D circular scatterer were used to validate the adjoint method's gradient calculation.
  • Experimental Data:
    • Dual-layer resolution target: A fabricated target with distinct layers was used to assess lateral and axial resolution.
  • 3D randomly distributed beads: Samples of polystyrene beads were imaged to evaluate axial resolution for both FS and BS.
  • Phase spoke target behind random fibers: A resolution target obscured by a scattering fiber layer was used to demonstrate the technique's ability to image through occlusions.
  • Biological samples: Fixed breast cancer cells and fixed C. elegans were imaged to showcase the method's applicability in biological imaging and its ability to differentiate subcellular structures and interfaces.

Key Insights and Ideas

  • Complementary Information from FS and BS: The study clearly demonstrates that forward scattering (FS) primarily captures smooth, volumetric RI information, while backward scattering (BS) is highly sensitive to interfacial details and lateral structures. This complementary nature is key to achieving comprehensive 3D RI reconstruction.
  • Overcoming Limitations of Traditional DT: By incorporating BS information and using a robust forward model for multiple scattering, the technique overcomes the "missing cone problem" and improves axial resolution compared to traditional transmission-mode DT.
  • Imaging Through Occlusions: The ability to reconstruct structures obscured by scattering materials (like fibers) is a significant advancement, opening doors for inspection in challenging industrial and biological scenarios.
  • Efficiency of Adjoint Method: The use of the adjoint method for gradient computation is highlighted as a critical factor for computational efficiency, especially for complex, volumetric reconstructions.

Key Results

  • Accurate 3D RI Reconstruction: The technique successfully reconstructed 3D RI distributions for various samples, including those with high RI contrasts and complex geometries.
  • Resolution Enhancement: The method achieved diffraction-limited lateral resolution, with a synthetic NA of 0.56 demonstrated on a dual-layer target, representing a significant improvement over single-LED brightfield microscopy.
  • Axial Resolution: The measured axial resolution for both FS and BS was found to be approximately 27.0 µm, agreeing well with theoretical predictions.
  • Successful Imaging of Obscured Samples: The technique successfully visualized resolution features of a target hidden behind a layer of scattering fibers, demonstrating its potential for inspection in challenging environments.
  • Biological Sample Imaging: The method effectively differentiated subcellular structures and interfacial features in breast cancer cells, highlighting its utility in biological analysis.

Limitations and Future Work

  • Axial Resolution Limitations: The axial resolution is currently limited by the system's NA, similar to transmission-mode DT. The authors suggest that increasing the NA or using multi-wavelength illumination could further improve axial resolution, particularly for BS.
  • Computational Cost: While the adjoint method improves efficiency, further reductions in computational cost are desirable. The authors suggest exploring simplified models and advanced computational methods like deep learning for faster and more isotropic imaging.
  • Ghost Images: In some cases, ghost images from overlapping layers or scattering effects were observed, particularly in the RIHF reconstruction. Future work aims to suppress these artifacts.
  • Future Directions: The authors propose focusing on improving resolution and computational efficiency, integrating advanced computational methods like deep learning, and exploring applications in mid-infrared photothermal microscopy and other label-free imaging techniques.

Summary updated

BibTeX
@inproceedings{li2025reflection,
  author = {Tongyu Li and Jiabei Zhu and Yi Shen and Lei Tian},
  title = {{Reflection-mode diffraction tomography of multiple-scattering samples on a reflective substrate from intensity images}},
  booktitle = {Optica},
  year = {2025},
  url = {https://openreview.net/forum?id=urpPW8TPkS},
}
Figure