BNL Home


Data Science Heals Imperfect X-ray Scattering Datasets

What is the scientific achievement?

The team of scientists developed and implemented a ‘physics-aware’ algorithm to correct for missing information in experimental X-ray scattering datasets. Because the algorithm relies on well-understood physics of X-ray scattering, the ‘healing’ operation provides robust and physically-rigorous results and outperforms all other conventional image interpolation methods.

Why does this achievement matter?

Experimental X-ray scattering images always contain missing data and artifacts, which complicate further analysis, especially rapid, automated analysis.  This healing operation is an essential pre-processing step for machine-learning interpretation of scientific data.

What are the details?

Healing X-ray Images

Click on the image to download a high-resolution version. Healing X-ray Images: X-ray scattering images have defects (left side), such as missing data and artifacts, which complicate analysis. A 'physics-aware' algorithm heals datasets in a physically-rigorous way (right side).

X-ray scattering is a powerful way to measure the structure of materials at the molecular- and nano-scale. Scattering images contain features, such as peaks and rings, which encode structural information.  As with most scientific data, collected X-ray scattering images are inevitably ‘incomplete,’ with missing data being due to limits of the measurement, or experimental considerations. These missing data render automated data analysis of the datasets much more difficult. In this work, the team developed an image healing algorithm designed for X-ray scattering/diffraction datasets. Because the algorithm is ‘physics-aware’ (incorporating known properties of an X-ray scattering measurement), it outperforms all other image healing methods when applied to X-ray scattering data. The healed images can then be easily fed into existing data analysis pipelines. Importantly, the image healing is also a crucial pre-processing step for input to machine-learning methods — which would otherwise tend to focus on the high-intensity — but ultimately irrelevant — image defects.

CFN Capabilities:

Experiments to validate the healing algorithm were performed on the CMS beamline at NSLS-II — operated in partnership between CFN and NSLS-II.

Publication Reference

J. Liu, J. Lhermitte, Y. Tian, Z. Zhang, D. Yu, K. G. Yager, Healing X-ray Scattering Images, Journal of the International Union of Crystallography (IUCrJ) 4, 455 (2017).

DOI: 10.1107/S2052252517006212

Acknowledgement of Support

U.S. DOE Office of Science Facilities, operated at Brookhaven National Laboratory under contract no. DE-SC0012704

2017-12587  |  INT/EXT  |  Media & Communications Office