Yutaka Ohtake

This study presents a practicable reconstruction workflow for cm-scale samples at the beamline BL05XU of SPring-8, utilizing a one-dimensional reflective beam expander (RBE) to expand the vertical field of view (FOV). While the original vertical beam size was limited to 1.3 mm, the RBE broadened it to approximately 5.7mm, enabling a massive 36 x 17 tiled scan of a smartphone in 15.5 hours. To handle the resulting geometry, we developed a hybrid reconstruction process: horizontal stitching in the projection domain followed by vertical stacking of reconstructed sub-volumes. We demonstrate that while a parallel-beam approximation is sufficient near the axis of rotation (AoR), it induces severe artifacts in regions distant from the AoR. Incorporating the RBE’s vertical fan-beam geometry is essential for high-fidelity large-scale imaging, effectively reducing artifacts that occur under a parallel-beam approximation.

Food is essential to all people throughout their lives, and its quality is directly linked to human health and quality of life. With the progression of population aging, there is an increasing demand for foods that are easier for elderly people to chew and swallow. At the same time, from the perspective of producing high value-added products, the development of foods with high palatability is strongly required. This study investigates how effectively X-ray computed tomography (CT), a common three-dimensional non-destructive testing technology in manufacturing industry, can contribute to the objective of identifying geometric features related to food texture. We report the results of a comparison of transmission images and CT volumes acquired using three different X-ray CT systems: lab CT, synchrotron, and X-ray microscopy.

We extend our NeRF-based material decomposition method for spectral X-ray CT (SXCT) to eliminate the need for pre-defined LAC for materials and detector calibration. By introducing a learnable Attenuation coefficient dictionary and an Energy Transformer, the method jointly estimates spatial compositions of the materials and effective detector sensitivity with almost no prior information on materials and detectors. Experimental results on real SXCT data show that our approach achieves sparse-view material decomposition comparable to our previous method that requires extensive prior knowledge.

X-ray phase CT has a lot of applications, including non-destructive testing and field of medicine because it can observe small differences in material density than conventional X-ray absorption CT. However, artifacts unique to phase CT may appear in conventional method. Therefore, we propose a new method for reconstructing CT volume from two-directional differential phase contrast images by solving the Poisson equation. Our method reduces artifacts in air regions and improves image quality in some cases.

Spectral X-ray computed tomography enables material decomposition by leveraging energy-dependent X-ray attenuation properties. However, material decomposition with spectral CT requires a longer acquisition time to obtain sufficient numbers of photons in each energy bin. Sparse-view offers a practical solution to reduce acquisition time, but it introduces ill-posedness, degrading decomposition accuracy. This study introduces a material decomposition framework based on Neural Radiance Fields where material maps are represented using a multilayer perceptron (MLP). The material maps are then optimized through a spectral forward projection process based on the Lambert–Beer’s law, while a partition of unity (PoU) loss ensures the physical constraint on material maps. Our method was evaluated using simulated and real spectral CT datasets and compared with a traditional statistical approach. The results demonstrated that our method performs well in material decomposition under sparse-view conditions. The results suggest that our “neural material composition fields” framework offers accurate material decomposition robust to sparse-view conditions without requiring labeled training data.

There is a growing interest in using X-ray CT scanners to measure industrial products, as CT scan data holds significant potential for reverse engineering and model-based design. To fully utilize CT scan data from assembled parts, it is essential to segment each individual component. The commonly used thresholding method is ineffective when there is no luminance difference at the boundaries between parts. Consequently, segmentation becomes particularly challenging when adjacent parts are made of the same material (i.e., have the same density). When threshold-based segmentation fails, manual interventions such as outlining objects with mouse clicks is often required. However, manual segmentation on a single image may lead to misalignments in surrounding slices, even though contour shapes generally remain similar within a certain region. If artificial intelligence can correct these misalignments and propagate contours across multiple slices, it could enable segmentation in many cases that are currently unfeasible with existing technologies. In this study, we propose a deep learning-based active contour algorithm for applying the manual segmentation performed on a single image to its surrounding slices.

In recent years, large-scale X-ray CT scanners have made it possible to perform CT scans on objects which are the size of automobiles. The resulting CT volume can be utilized for reverse engineering. However, handling them is difficult due to their large data sizes. Furthermore, the segmentation required for reverse engineering is currently performed manually by experts, which is time-consuming. In this study, we propose a semi-automatic method for parts segmentation of the CT scanned assembled car. The method involves automatically segmenting the CT scanned assembled car by thickness and material and then allowing the user to manually select from the automatically segmented parts. The segmentation based on thickness is carried out by the disappearing thin objects when the isosurface is offset. The segmentation based on material is performed by referencing CT values of the input CT volume. As a result, the parts composed of iron and aluminum, such as motors, were properly segmented.

Plant roots influence various soil physical properties by altering the soil structure and pore configuration; however, a detailed understanding of these effects remains limited. In this study, we applied a relatively simple approach for segmenting plant roots and soil constituents using X-ray computed tomography (CT) images to evaluate root-induced changes in soil structure. The method combines manual initialization with a layer-wise automated region-growing approach, enabling the extraction of the root systems of soybean, Italian ryegrass, and Guinea grass. The method utilizes freely available software with a simple interface and does not require advanced image analysis skills, making it accessible to a wide range of researchers. The soil particles, pore water, and pore air were segmented using a Kriging-based thresholding technique. The segmented four-phase images allowed for the quantification of the volume fractions of soil constituents, pore size distributions, and coordination numbers. Furthermore, by separating the rhizosphere and bulk soil, we found that the root presence significantly reduced solid fractions and increased water content, particularly in the upper soil layers. Macropores and fine pores were observed near the roots, highlighting the complex structural impacts of root growth. While further validation is needed to assess the method’s applicability across different soil types and imaging conditions, it provides a practical basis for visualizing and quantifying root–soil interactions, and could contribute to advancing our understanding of how plant roots influence key soil hydraulic and thermal properties.

Point clouds from laser scanners have been widely used in recent research on indoor modeling methods. Currently, particularly in data-driven modeling methods, data preprocessing for dividing structural components and nonstructural components is required before modeling. In this paper, we propose an indoor modeling method without the classification of structural and nonstructural components. A pre-mesh is generated for constructing the adjacency relations of point clouds, and plane components are extracted using planar-based region growing. Then, the distance fields of each plane are calculated, and voxel data referred to as a surface confidence map are obtained. Subsequently, the inside and outside of the indoor model are classified using a graph-cut algorithm. Finally, indoor models with watertight meshes are generated via dual contouring and mesh refinement. The experimental results showed that the point-to-mesh error ranged from approximately 2 mm to 50 mm depending on the dataset. Furthermore, completeness—measured as the proportion of original point-cloud data successfully reconstructed into the mesh—approached 1.0 for single-room datasets and reached around 0.95 for certain multiroom and synthetic datasets. These results demonstrate the effectiveness of the proposed method in automatically removing non-structural components and generating clean structural meshes.

Photon-counting X-ray detectors, in contrast to conventional flat panel detectors, have the capability of distinguishing between photons with different energies, and have been leveraged for material decomposition tasks for materials with similar X-ray attenuation properties. However, as the energy resolution increases, it may suffer from a lack of photons falling into each energy bin, resulting in an inadequate material decomposition. In this study, we demonstrate the effectiveness of recent neural radiance fields (NeRF) for material decomposition tasks using spectral X-ray CT in sparse-view reconstruction scenarios. Particularly, our method exploits the known linear attenuation properties of base materials and reconstructs the fractions of base materials comprising the target object.