Improving Multislice Electron Ptychography with a Generative Prior

2025-07-23

AI Summary

To address the ill-posedness, computational inefficiency, and susceptibility to suboptimal solutions in multi-slice electron ptychographic tomography (MEP), this work proposes a hybrid reconstruction method integrating generative priors. We introduce diffusion models—specifically, a tailored MEP-Diffusion architecture—into MEP-based 3D reconstruction for the first time and train it on large-scale crystallographic structure data. Furthermore, we embed diffusion posterior sampling (DPS) into a conventional iterative optimization framework to jointly leverage physical forward models and learned priors. The proposed approach significantly enhances reconstruction stability and fidelity: on standard benchmarks, it achieves a 90.50% improvement in structural similarity (SSIM) over baseline methods. It enables high-fidelity, atomic-resolution 3D crystal structure reconstruction, establishing a novel, interpretable, and data-driven paradigm for quantitative electron microscopy.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Security and Privacy: Data transparency and provenance

Abstract

Multislice electron ptychography (MEP) is an inverse imaging technique that computationally reconstructs the highest-resolution images of atomic crystal structures from diffraction patterns. Available algorithms often solve this inverse problem iteratively but are both time consuming and produce suboptimal solutions due to their ill-posed nature. We develop MEP-Diffusion, a diffusion model trained on a large database of crystal structures specifically for MEP to augment existing iterative solvers. MEP-Diffusion is easily integrated as a generative prior into existing reconstruction methods via Diffusion Posterior Sampling (DPS). We find that this hybrid approach greatly enhances the quality of the reconstructed 3D volumes, achieving a 90.50% improvement in SSIM over existing methods.

Problem

• Enhance multislice electron ptychography reconstruction quality
• Address time-consuming iterative algorithms in MEP
• Mitigate ill-posed nature of inverse imaging

Innovation

• Uses diffusion model for crystal structures
• Integrates generative prior via DPS
• Improves reconstruction quality significantly