Deep learning for high-resolution and high-sensitivity interferometric phase contrast imaging.

Area of Science:

• Physics
• Optics
• Image Processing

Background:

• Talbot-Lau interferometry faces a trade-off between phase sensitivity and spatial resolution.
• This challenge stems from geometric blur at optimal sample positions and limitations of conventional sources.

Purpose of the Study:

• To develop a deep learning method to simultaneously achieve high phase sensitivity and high spatial resolution in interferometric imaging.
• To overcome the inherent limitations of Talbot-Lau interferometry for neutron and X-ray imaging.

Main Methods:

• A generative adversarial network (GAN) was trained using numerically generated differential phase contrast images.
• The GAN utilized image pairs from two distinct sample positions as input data.
• The trained neural network was applied to real experimental data from a neutron grating interferometer.

Main Results:

• The deep learning method successfully estimated images with both enhanced phase sensitivity and improved spatial resolution.
• The approach effectively mitigated the trade-off between sensitivity and resolution observed in traditional methods.
• Improved imaging quality was demonstrated on experimental neutron interferometry data.

Conclusions:

• Deep learning offers a viable solution to the phase sensitivity-spatial resolution trade-off in Talbot-Lau interferometry.
• The proposed method enhances the performance of interferometric imaging for scientific applications.
• This technique holds promise for advancing neutron and X-ray imaging capabilities.