In a comprehensive study, we demonstrate the performance and typical application scenarios for laboratory-based nano-computed tomography in materials research on various samples. Specifically, we focus on a projection magnification system with a nano focus source. The imaging resolution is quantified with common 2D test structures and validated in 3D applications by means of the Fourier Shell Correlation. As representative application examples from nowadays material research, we show metallization processes in multilayer integrated circuits, aging in lithium battery electrodes, and volumetric of metallic sub-micrometer fillers of composites. Thus, the laboratory system provides the unique possibility to image non-destructively structures in the range of 170–190 nanometers, even for high-density materials.

Strict morphological control over growing polymer particles is an indispensable requirement in many catalytic olefin polymerization processes. In catalysts with mechanically stronger supports, e. g., polymerization‐grade silicas, the emergence of extensive cracks via the sectioning fragmentation mechanism requires severe stress build‐up in the polymerizing catalyst particle. Here, we report on three factors that influence the degree of sectioning in silica‐supported olefin polymerization catalysts. Laboratory‐based X‐ray nano‐computed tomography (nanoCT) and focused ion beam‐scanning electron microscopy (FIB‐SEM) were employed to study catalyst particle morphology and crack propagation in two showcase catalyst systems, i.e., a zirconocene‐based catalyst (i.e., Zr/MAO/SiO₂, with Zr=2,2’‐biphenylene‐bis‐2‐indenyl zirconium dichloride and MAO=methylaluminoxane) and a Ziegler‐Natta catalyst (i.e., TiCl₄/MgCl₂/SiO₂), during slurry‐phase ethylene polymerization. The absence of extensive macropores in some of the catalysts’ larger constituent silica granulates, a sufficient accessibility of the catalyst particle interior at reaction onset, and a high initial polymerization rate were found to favor the occurrence of the sectioning pathway at different length scales. While sectioning is beneficial for reducing diffusion limitations, its appearance in mechanically stronger catalyst supports can indicate a suboptimal support structure or unfavourable reaction conditions.

X-ray full-field microscopy at laboratory sources for photon energies above 10 keV suffers from either long exposure times or low resolution. The photon flux is mainly limited by the objectives used, having a limited numerical aperture NA. We show that this can be overcome by making use of the cone-beam illumination of laboratory sources by imaging the same field of view (FoV) several times under slightly different angles using an array of X-ray lenses. Using this technique, the exposure time can be reduced drastically without any loss in terms of resolution. A proof-of-principle is given using an existing laboratory metal-jet source at the 9.25 keV Ga K_α-line and compared to a ray-tracing simulation of the setup.

Novel and high-performance medical image classification pipelines are heavily utilizing ensemble learning strategies. The idea of ensemble learning is to assemble diverse models or multiple predictions and, thus, boost prediction performance. However, it is still an open question to what extent as well as which ensemble learning strategies are beneficial in deep learning based medical image classification pipelines. In this work, we proposed a reproducible medical image classification pipeline for analyzing the performance impact of the following ensemble learning techniques: Augmenting, Stacking, and Bagging. The pipeline consists of state-of-the-art preprocessing and image augmentation methods as well as 9 deep convolution neural network architectures. It was applied on four popular medical imaging datasets with varying complexity. Furthermore, 12 pooling functions for combining multiple predictions were analyzed, ranging from simple statistical functions like unweighted averaging up to more complex learning-based functions like support vector machines. Our results revealed that Stacking achieved the largest performance gain of up to 13% F1-score increase. Augmenting showed consistent improvement capabilities by up to 4% and is also applicable to single model based pipelines. Cross-validation based Bagging demonstrated significant performance gain close to Stacking, which resulted in an F1-score increase up to +11%. Furthermore, we demonstrated that simple statistical pooling functions are equal or often even better than more complex pooling functions. We concluded that the integration of ensemble learning techniques is a powerful method for any medical image classification pipeline to improve robustness and boost performance.