Artificial Neural Networks for Analysis of Coherent X-Ray Diffraction Images

We present an application of neural networks to the analysis of diffraction images produced via X-ray photon correlation spectroscopy (XPCS) experiments. The detection apparatus for these experiments is an array of charge-coupled devices (CCDs). Determining the incident location of each photon allows for estimation of the image contrast, which elucidates structural and dynamical properties of a measured system such as diffusion constant and particle radius. Photons incident on the detector in the same vicinity result in additive interference that obscures their individual locations. A neural network classifier was designed and trained on data from an artificial model to scan an image and determine a discrete probability distribution for the number of photons that have been incident within the area of each CCD. These distributions were fit to the contrast and were found to be in agreement with the true underlying value. We have thus demonstrated a promising method for accurately determining the contrast from experimental XPCS images.