1.Building a Foundation Model for Brain Tissue Volumes Acquired with X-ray Nano-Tomography & 2. Deep Learning for Bragg Coherent Diffraction Imaging: Gap In-painting and Phase Retrieval

Coffee will be served from 9.30am

1. Abstract: High-resolution brain tissue volumes are regularly acquired at ESRF. Mapping neural circuits in these volumes could be very useful for understanding how the mammalian brain works.
However, manually mapping such circuits is prohibitively costly. Machine learning has been very successful in addressing this issue, as it can map neural circuits in large volumes by leveraging a small amount of expert annotations. These annotations remain costly to obtain and must be provided for every newly acquired volume. Using self-supervised learning and hundreds of volumes acquired at id16, we propose a new machine learning model able to map neural circuits using an order of magnitude less data...

2. Abstract: Bragg Coherent Diffraction Imaging is a powerful X-ray microscopy technique for the imaging of the strain inside crystalline nano-particles. However, its strength relies partly on computational algorithms for the inversion of the diffraction patterns. Despite the overall robustness, these standard algorithms can struggle with detector gaps affected regions in the diffraction patterns and highly strained particles. In fact, while the former induces artifacts in the reconstructions, the latter often hinder the convergence stability leading to not successful inversions. For these reasons we have developed two different convolutional neural networks for detector gaps in-painting and for the phase retrieval of highly strained particles respectively. The first has been trained to recover the missing intensity inside the gaps and significantly reduces the presence of artifacts in the reconstructions of experimental data. The second has been trained to predict the lost reciprocal space phase and thus estimate the particle complex amplitude, speeding up the phasing process and increasing the rate of successful reconstructions.