High-speed quantitative X-ray multi-contrast imaging with deep learning based modulated pattern analysis

Area of Science:

• Physics
• Materials Science
• Biomedical Imaging

Background:

• X-ray multi-contrast imaging (absorption, phase, dark-field) enables non-destructive visualization.
• Low efficiency in X-ray pattern analysis limits high-resolution, in situ imaging applications.

Purpose of the Study:

• Introduce the Enhanced Scanning Pattern-based Imaging Neural Network (ESPINNet) for high-speed, high-resolution quantitative imaging.
• Enhance the capabilities of previous neural networks by enabling dark-field image generation.

Main Methods:

• Leveraging scanning patterns for improved resolution and measurement precision.
• Utilizing a neural network architecture (ESPINNet) adaptable to various modulation patterns (sandpaper, coded masks, gratings).

Main Results:

• ESPINNet achieves faster data collection compared to correlation-based methods (XSVT, UMPA).
• Demonstrates balanced performance in resolution and speed, requiring fewer scanning images.
• Enables real-time 2D and 3D multi-contrast imaging.

Conclusions:

• ESPINNet is a transformative solution for high-speed, in situ X-ray imaging.
• Broad applicability in materials science and biomedical research due to its adaptability and enhanced capabilities.