Machine Learning for the Automated Selection and Reconstruction of Multi-Modal Nanotomography Data of Bone-Implant Interfaces

Supervisors: Dr. Imke Greving (Hereon), Dr. Berit Zeller-Plumhoff (Hereon), Prof. Jan Baumbach (UHH)

The combination of the different X-ray methods is required to gain in-depth understanding of the structure function relationship in complex hierarchical materials, such as bone surrounding biodegradable implants. One major challange in multimodal and multiscale X-ray imaging is the large amount of data that is obtained and needs to be analysed to establish scientific results. This is the case in particular when looking at biological systems which are highly variable and therefore require the analysis of a large amount of data. One way of reducing the amount of data generated is by selecting defined regions of interest for the analysis based on overview scans and determining the minimal amount of information necessary to be able to tomographically reconstruct the desired 3D volume.
The main goal of the project is to use machine learning to determine the experimental parameters in terms of number of projections and step numbers and sizes for 3D scanning X-ray fluorescence and 3D scanning X-ray scattering based on the sample morphology obtained using near-field holotomography. Moreover, the machine learning algorithm will determine which regions are of highest interest based on an unsupervised learning approach similar to anomaly detection and select the imaging patterns to optimize imaging time vs. feature resolution.