Skip to main content
Research Highlight | Materials Science

AI-enhanced single image magnetic phase retrieval

In a study published in NPJ Computational materials, researchers developed a method, SIPRAD, to accurately reconstruct the magnetic phase shift from a single Lorentz transmission electron microscopy image that outperforms previous techniques.

Scientific Achievement

A new method, SIPRAD, was developed to reconstruct and isolate the magnetic phase shift from a single LTEM image.

Significance and Impact

SIPRAD enables qualitatively superior and quantitatively accurate phase reconstructions of time-resolved LTEM data, and the method is publicly available in the open-source software package PyLorentz.

Research Details

  • Automatic differentiation and a deep image prior were used in conjunction with an LTEM forward model of image formation to reconstruct the electron phase shift while suppressing noise.
  • A non-uniform amplitude or electrostatic phase shift induced by the sample can be accounted for, which allows the magnetic component of the phase shift to be effectively isolated.
  • SIPRAD vastly outperforms previous methods in all cases and is especially suited to experimentally relevant imaging conditions.

DOI: 10.1038/s41524-024-01285-8

Download this highlight