High-fidelity optical diffraction tomography of multiple scattering samples.

Area of Science:

• Computational imaging
• Wave optics
• Tomographic reconstruction

Background:

• Optical diffraction tomography (ODT) enables label-free imaging of transparent samples.
• Existing learning tomography methods often rely on approximations like the beam propagation method (BPM).
• Accurate forward models are crucial for high-fidelity tomographic reconstruction.

Purpose of the Study:

• To develop an improved iterative reconstruction scheme for ODT.
• To leverage the split-step non-paraxial (SSNP) method within a learning tomography framework.
• To enhance the accuracy and resolution of ODT reconstructions, especially for complex biological samples.

Main Methods:

• An iterative reconstruction scheme employing the SSNP method as the forward model in a learning tomography (LT-SSNP) framework.
• Comparison with a previous learning tomography approach using the beam propagation method (LT-BPM).
• Validation using synthetic and experimental measurements, and a novel method using discrete dipole approximation for ground truth generation.

Main Results:

• LT-SSNP demonstrates superior performance and accuracy compared to LT-BPM.
• LT-SSNP successfully resolves structures that are highly distorted or undetectable with LT-BPM.
• The discrete dipole approximation method provides a reliable way to compare reconstruction algorithms without ground truth.
• Learning approaches show potential for data compression by reducing scanning angles.

Conclusions:

• The proposed LT-SSNP method significantly enhances ODT reconstruction accuracy and resolution.
• LT-SSNP offers a more robust approach for imaging complex and distorted structures.
• This work introduces a valuable tool for quantitative analysis in ODT and explores efficient data acquisition strategies.