Using automatic differentiation as a general framework for ptychographic reconstruction

Area of Science:

• Optics and Photonics
• Computational Imaging
• Materials Science

Background:

• Coherent diffraction imaging (CDI) offers super-resolution capabilities beyond traditional lens-based microscopy.
• Object reconstruction in CDI relies on minimizing an error metric between measured and simulated diffraction patterns.
• Calculating analytical derivatives of the error metric is often complex and limits method development.

Purpose of the Study:

• To introduce automatic differentiation as a method to simplify and generalize object reconstruction in CDI.
• To enable exploration of novel CDI variations without manual derivative computation.
• To demonstrate the applicability of automatic differentiation across diverse ptychographic imaging models.

Main Methods:

• Implemented automatic differentiation to replace manual calculation of error metric derivatives.
• Utilized physics-based experimental forward models to define the imaging process.
• Applied the method to various complex ptychographic experimental setups.

Main Results:

• Successfully reconstructed objects using automatic differentiation without analytical derivative expressions.
• Demonstrated straightforward implementation for complex ptychographic models.
• Validated the generality and efficiency of the proposed approach.

Conclusions:

• Automatic differentiation provides a powerful and flexible tool for object reconstruction in CDI.
• This method significantly reduces the complexity of developing new imaging algorithms.
• The approach facilitates broader research into advanced coherent diffraction imaging techniques.