Python-Based Analysis of Multidimensional Electron Diffraction Data in Human Bone

Abstract

The main focus of this thesis is to study the nanoscale crystallographic organization of human cortical bone using detector-based four-dimensional scanning transmission electron microscopy (4D-STEM). Because conventional electron diffraction averages over relatively large areas, it is difficult to relate diffraction features directly to local real-space structure in bone. This limitation is addressed with a previously acquired 4D-STEM dataset and a simultaneously recorded HAADF-STEM image, where py4DSTEM and Dragonfly-based real-space segmentation were used. The analysis was based on ROI selection, mean diffraction pattern (DP) construction, detector shape and position selection, virtual dark-field imaging, real-space masking, thresholding, and diffraction reconstruction from masked and thresholded pixels. The broad diffraction arc commonly associated with the apatite (002) reflection was used to guide the detector placement. Bright-band, dark-band, and extra-fibrillar mineral regions were segmented in Dragonfly using the HAADF-STEM image for real-space masking purposes. The results reveal that the diffraction signal along the selected arc is spatially heterogeneous, with different detector locations highlighting different local mineralized regions. An aggressive positive thresholding analysis separates the pixels that have strong diffraction contributions from a broader weakly contributing background, and the resulting reconstructed DPs of positiveand negative-thresholded DF images display either strong Bragg-like disk features or a broader ring geometry, respectively. Comparisons of different mask-selected regions show that the revealed structural regions do not separate into sharply distinctive diffraction classes, given the detector conditions used. However, this work suffers from some limitations, such as sample thickness, heterogeneity, diffuse scattering, low signal-to-noise ratio (SNR), and beam sensitivity. Nonetheless, the thesis demonstrates that masking 4D-STEM data in both real and reciprocal space can be used to connect local diffraction features to nanoscale structural variations in bone. Further work is needed to improve the above-mentioned limitations and the overall interpretation of our findings.