Automation of experiments at synchrotron based X-ray fluorescence nanoprobes using artificial intelligence

Quantitative microscopy can probe complex, heterogeneous systems, but advanced techniques hit fundamental limits in throughput, radiation damage and experimental complexity. Physics sets the ultimate limits, but practical throughput is also constrained by necessary human involvement; we show that AI can reduce these bottlenecks by automating region-of-interest identification, acquisition control, and signal enhancement, thereby bringing experiments closer to those physical limits. We demonstrate this approach on X-ray fluorescence (XRF) imaging in cancer and drug research, where AI assists in determining which regions to measure and optimizing acquisition parameters. A self-supervised denoising strategy, exploiting the multi-element detector configuration, recovers high quality chemical maps from shorter exposures and lower X-rays doses. This automated and optimized workflow increases throughput from a handful of manually selected maps to statistically significant datasets containing hundreds of measurements. This enables synchrotron beamlines to operate in a more autonomous, high-throughput mode, including unattended overnight acquisition, while maintaining high data quality.

You can follow this PhD Defense on Zoom via the link below: 

https://esrf.zoom.us/j/92145269936?pwd=LkLDvbB3bgmKMJsQIjj0XLmzmywIOt.1

Passcode : 322707