Three dimensions, two microscopes, one code: Automatic differentiation for x-ray nanotomography beyond the depth of...

Area of Science:

• Computational Imaging
• X-ray Microscopy
• Materials Science

Background:

• Conventional X-ray nanotomography struggles with thick samples due to diffraction and multiple scattering.
• Advances in X-ray nanotomography necessitate new reconstruction methods for accurate imaging of complex, thick specimens.

Purpose of the Study:

• To develop and validate a novel reconstruction method for X-ray nanotomography that addresses diffraction and multiple scattering.
• To demonstrate the method's applicability to both full-field microscopy and ptychography techniques.

Main Methods:

• Modeling scattering and diffraction effects using multislice propagation.
• Employing iterative optimization to retrieve the 3D object function.
• Utilizing TensorFlow for optimization and automatic differentiation.

Main Results:

• The proposed method successfully models and corrects for multiple scattering and diffraction effects.
• The reconstruction technique is validated for both full-field microscopy and ptychography.
• The implementation showcases the flexibility and portability of deep learning tools in computational imaging.

Conclusions:

• The developed iterative reconstruction method offers a robust solution for imaging thick samples in X-ray nanotomography.
• This approach overcomes limitations of conventional algorithms, enabling more accurate 3D reconstructions.
• The use of deep learning frameworks like TensorFlow simplifies complex optimization problems in advanced imaging.