Deep Learning, AI Used to Advance X-ray Data Technology

Scientists from the United States Department of Energy’s (DOE) Argonne National Laboratory are using deep learning and artificial intelligence strategies to upgrade the current Advanced Photon Source (APS) and visualize X-ray data in three dimensions.

Researchers have developed a new computational framework called 3D-CDI-NN. The framework has demonstrated it can create 3D visualizations from data collected at the APS significantly faster than traditional methods.

Coherent diffraction imaging (CDI) is an X-ray technique that bounces ultra-bright X-ray beams off samples. The beams of light then are collected by detectors as data and are turned into images. According to Mathew Cherukara, leader of the Computational X-ray Science group in Argonne’s X-ray Science Division (XSD), the current detectors only capture some of the beam’s information.

Scientists rely on computers to fill in missing data. However, the process can take a significant amount of time. The solution, according to Cherukara, is to train artificial intelligence to recognize objects and the changes they undergo directly from raw data, without having to account for missing information.

The team trained the neural network with simulated X-ray data. The neural network is a series of algorithms designed to teach computers to predict outcomes based on the data it receives.

“We used computer simulations to create crystals of different shapes and sizes, and we converted them into images and diffraction patterns for the neural network to learn,” lead author Henry Chan said in a press release. “The ease of quickly generating many realistic crystals for training is the benefit of simulations.”

Physicist and group leader of Argonne’s Materials Science Division, Stephan Hruszkewycz said it will process accurate data once the network is trained.  However, there will still be room for improvement and efficiency.

“The Materials Science Division cares about coherent diffraction because you can see materials at few-nanometer length scales — about 100,000 times smaller than the width of a human hair — with X-rays that penetrate into environments,” Hruszkewycz said.

“This paper is a demonstration of these advanced methods, and it greatly facilitates the imaging process. We want to know what a material is, and how it changes over time, and this will help us make better pictures of it as we make measurements.”

After testing 3D-CDI-NN’s ability to fill missing information, scientists saw that the network can reconstruct images with less data than typically required to compensate for information that was missed by the detectors.

According to the team, they will need to continue collecting data and having the network learn from it.

“In order to make full use of what the upgraded APS will be capable of, we have to reinvent data analytics,” Cherukara said. “Our current methods are not enough to keep up. Machine learning can make full use and go beyond what is currently possible.”

The incorporation of this new technology has the potential to advance 3D imaging technologies regarding biological structures.