Advanced characterisation of 3D structure and porosity of ordinary portland cement (OPC) mortar using plasma focused ion beam tomography and X-ray computed tomography.

The visualisation and quantification of pore networks and main phases have been critical research topics in cementitious materials as many critical mechanical and chemical properties and infrastructure reliability rely on these 3D characteristics. In this study, we realised the mesoscale serial sectioning and analysis up to ∼80 µm by ∼90 µm by ∼60 µm on portland cement mortar using plasma focused ion beam (PFIB) for the first time. The workflow of working with mortar and PFIB was established applying a prepositioned hard silicon mask to reduce curtaining. Segmentation with minimal human interference was performed using a trained neural network, in which multiple types of segmentation models were compared. Combining PFIB analysis at microscale with X-ray micro-computed tomography, the analysis of capillary pores and air voids ranging from hundreds of nanometres (nm) to millimetres (mm) can be conducted. The volume fraction of large capillary pores and air voids are 11.5% and 12.7%, respectively. Moreover, the skeletonisation of connected capillary pores clearly shows fluid transport pathways, which is a key factor determining durability performance of concrete in aggressive environments. Another interesting aspect of the FIB tomography is the reconstruction of anhydrous phases, which could enable direct study of hydration kinetics of individual cement phases.

Scanning Electron Microscopy versus Transmission Electron Microscopy for Material Characterization: A Comparative Study on High-Strength Steels.

The microstructures of quenched and tempered steels have been traditionally explored by transmission electron microscopy (TEM) rather than scanning electron microscopy (SEM) since TEM offers the high resolution necessary to image the structural details that control the mechanical properties. However, scanning electron microscopes, apart from providing larger area coverage, are commonly available and cheaper to purchase and operate compared to TEM and have evolved considerably in terms of resolution. This work presents detailed comparison of the microstructure characterization of quenched and tempered high-strength steels with TEM and SEM electron channeling contrast techniques. For both techniques, similar conclusions were made in terms of large-scale distribution of martensite lath and plates and nanoscale observation of nanotwins and dislocation structures. These observations were completed with electron backscatter diffraction to assess the martensite size distribution and the retained austenite area fraction. Precipitation was characterized using secondary imaging in the SEM, and a deep learning method was used for image segmentation. In this way, carbide size, shape, and distribution were quantitatively measured down to a few nanometers and compared well with the TEM-based measurements. These encouraging results are intended to help the material science community develop characterization techniques at lower cost and higher statistical significance.

Optimization of 3D network topology for bioinspired design of stiff and lightweight bone-like structures.

A truly bioinspired approach to design optimization should follow the energetically favorable natural paradigm of "minimum inventory with maximum diversity". This study was inspired by constructive regression of trabecular bone - a natural process of network connectivity optimization occurring early in skeletal development. During trabecular network optimization, the original excessively connected network undergoes incremental pruning of redundant elements, resulting in a functional and adaptable structure operating at lowest metabolic cost. We have recapitulated this biological network topology optimization algorithm by first designing in silico an excessively connected network in which elements are dimension-independent linear connections among nodes. Based on bioinspired regression principles, least-loaded connections were iteratively pruned upon simulated loading. Evolved networks were produced along this optimization trajectory when pre-set convergence criteria were met. These biomimetic networks were compared to each other, and to the reference network derived from mature trabecular bone. Our results replicated the natural network optimization algorithm in uniaxial compressive loading. However, following triaxial loading, the optimization algorithm resulted in lattice networks that were more stretch-dominated than the reference network, and more capable of uniform load distribution. As assessed by 3D printing and mechanical testing, our heuristic network optimization procedure opens new possibilities for parametric design.

Deep learning for 3D imaging and image analysis in biomineralization research.

Biomineralization research examines structure-function relations in all types of exo- and endo-skeletons and other hard tissues of living organisms, and it relies heavily on 3D imaging. Segmentation of 3D renderings of biomineralized structures has long been a bottleneck because of human limitations such as our available time, attention span, eye-hand coordination, cognitive biases, and attainable precision, amongst other limitations. Since recently, some of these routine limitations appear to be surmountable thanks to the development of deep-learning algorithms for biological imagery in general, and for 3D image segmentation in particular. Many components of deep learning often appear too abstract for a life scientist. Despite this, the basic principles underlying deep learning have many easy-to-grasp commonalities with human learning and universal logic. This primer presents these basic principles in what we feel is an intuitive manner, without relying on prerequisite knowledge of informatics and computer science, and with the aim of improving the reader's general literacy in artificial intelligence and deep learning. Here, biomineralization case studies are presented to illustrate the application of deep learning for solving segmentation and analysis problems of 3D images ridden by various artifacts, and/or which are plainly difficult to interpret. The presented portfolio of case studies includes three examples of imaging using micro-computed tomography (µCT), and three examples using focused-ion beam scanning electron microscopy (FIB-SEM), all on mineralized tissues. We believe this primer will expand the circle of users of deep learning amongst biomineralization researchers and other life scientists involved with 3D imaging, and will encourage incorporation of this powerful tool into their professional skillsets and to explore it further.

Altered topological blueprint of trabecular bone associates with skeletal pathology in humans.

Bone is a hierarchically organized biological material, and its strength is usually attributed to overt factors such as mass, density, and composition. Here we investigate a covert factor - the topological blueprint, or the network organization pattern of trabecular bone. This generally conserved metric of an edge-and-node simplified presentation of trabecular bone relates to the average coordination/valence of nodes and the equiangular 3D offset of trabeculae emanating from these nodes. We compare the topological blueprint of trabecular bone in presumably normal, fractured osteoporotic, and osteoarthritic samples (all from human femoral head, cross-sectional study). We show that bone topology is altered similarly in both fragility fracture and in joint degeneration. Decoupled from the morphological descriptors, the topological blueprint subjected to simulated loading associates with an abnormal distribution of strain, local stress concentrations and lower resistance to the standardized load in pathological samples, in comparison with normal samples. These topological effects show no correlation with classic morphological descriptors of trabecular bone. The negative effect of the altered topological blueprint may, or may not, be partly compensated for by the morphological parameters. Thus, naturally occurring optimization of trabecular topology, or a lack thereof in skeletal disease, might be an additional, previously unaccounted for, contributor to the biomechanical performance of bone, and might be considered as a factor in the life-long pathophysiological trajectory of common bone ailments.

Application of Deep Learning Convolutional Neural Networks for Internal Tablet Defect Detection: High Accuracy, Throughput, and Adaptability.

Tablet defects encountered during the manufacturing of oral formulations can result in quality concerns, timeline delays, and elevated financial costs. Internal tablet cracking is not typically measured in routine inspections but can lead to batch failures such as tablet fracturing. X-ray computed tomography (XRCT) has become well-established to analyze internal cracks of oral tablets. However, XRCT normally generates very large quantities of image data (thousands of 2D slices per data set) which require a trained professional to analyze. A user-guided manual analysis is laborious, time-consuming, and subjective, which may result in a poor statistical representation and inconsistent results. In this study, we have developed an analysis program that incorporates deep learning convolutional neural networks to fully automate the XRCT image analysis of oral tablets for internal crack detection. The computer program achieves robust quantification of internal tablet cracks with an average accuracy of 94%. In addition, the deep learning tool is fully automated and achieves a throughput capable of analyzing hundreds of tablets. We have also explored the adaptability of the deep learning analysis program toward different products (e.g., different types of bottles and tablets). Finally, the deep learning tool is effectively implemented into the industrial pharmaceutical workflow.

Impact of contrast injection and stent-graft implantation on reproducibility of volume measurements in semiautomated segmentation of abdominal aortic aneurysm on computed tomography.

PURPOSE: To assess the impact of contrast injection and stent-graft implantation on feasibility, accuracy, and reproducibility of abdominal aortic aneurysm (AAA) volume and maximal diameter (D-max) measurements using segmentation software. MATERIALS AND METHODS: CT images of 80 subjects presenting AAA were divided into four equal groups: with or without contrast enhancement, and with or without stent-graft implantation. Semiautomated software was used to segment the aortic wall, once by an expert and twice by three readers. Volume and D-max reproducibility was estimated by intraclass correlation coefficients (ICC), and accuracy was estimated between the expert and the readers by mean relative errors. RESULTS: All segmentations were technically successful. The mean AAA volume was 167.0 ± 82.8 mL and the mean D-max 55.0 ± 10.6 mm. Inter- and intraobserver ICCs for volume and D-max measurements were greater than 0.99. Mean relative errors between readers varied between -1.8 ± 4.6 and 0.0 ± 3.6 mL. Mean relative errors in volume and D-max measurements between readers showed no significant difference between the four groups (P ≥ 0.2). CONCLUSION: The feasibility, accuracy, and reproducibility of AAA volume and D-max measurements using segmentation software were not affected by the absence of contrast injection or the presence of stent-graft. KEY POINTS: • AAA volumetry by semiautomated segmentation is accurate on CT following endovascular repair. • AAA volumetry by semiautomated segmentation is accurate on unenhanced CT. • Standardization of the segmentation technique maximizes the reproducibility of volume measurements.

Clinical validation of semi-automated software for volumetric and dynamic contrast enhancement analysis of soft tissue venous malformations on magnetic resonance imaging examination.

OBJECTIVES: To evaluate venous malformation (VM) volume and contrast-enhancement analysis on magnetic resonance imaging (MRI) compared with diameter evaluation. METHODS: Baseline MRI was undertaken in 44 patients, 20 of whom were followed by MRI after sclerotherapy. All patients underwent short-tau inversion recovery (STIR) acquisitions and dynamic contrast assessment. VM diameters in three orthogonal directions were measured to obtain the largest and mean diameters. Volumetric reconstruction of VM was generated from two orthogonal STIR sequences and fused with acquisitions after contrast medium injection. Reproducibility (interclass correlation coefficients [ICCs]) of diameter and volume measurements was estimated. VM size variations in diameter and volume after sclerotherapy and contrast enhancement before sclerotherapy were compared in patients with clinical success or failure. RESULTS: Inter-observer ICCs were similar for diameter and volume measurements at baseline and follow-up (range 0.87-0.99). Higher percentages of size reduction after sclerotherapy were observed with volume (32.6 ± 30.7%) than with diameter measurements (14.4 ± 21.4%; P = 0.037). Contrast enhancement values were estimated at 65.3 ± 27.5% and 84 ± 13% in patients with clinical failure and success respectively (P = 0.056). CONCLUSIONS: Venous malformation volume was as reproducible as diameter measurement and more sensitive in detecting therapeutic responses. Patients with better clinical outcome tend to have stronger malformation enhancement. KEY POINTS: • Magnetic resonance imaging readily demonstrates diameters and volumes of venous malformations • MRI diameter calculations are reproducible in estimating the size of venous malformations • But volumetric models of malformations are more sensitive in detecting therapeutic response • Dynamic enhancement is also better assessed with automated volumetric software • Volumetric analysis of malformations offers promise to guide therapy and assess response.

Seeded ND medical image segmentation by cellular automaton on GPU.

PURPOSE: We present a GPU-based framework to perform organ segmentation in N-dimensional (ND) medical image datasets by computation of weighted distances using the Ford-Bellman algorithm (FBA). Our GPU implementation of FBA gives an alternative and optimized solution to other graph-based segmentation techniques. METHODS: Given a number of K labelled-seeds, the segmentation algorithm evolves and segments the ND image in K objects. Each region is guaranteed to be connected to seeds with the same label. The method uses a Cellular Automata (CA) to compute multiple shortest-path-trees based on the FBA. The segmentation result is obtained by K-cuts of the graph in order to separate it in K sets. A quantitative evaluation of the method was performed by measuring renal volumes of 20 patients based on magnetic resonance angiography (MRA) acquisitions. Inter-observer reproducibility, accuracy and validity were calculated and associated computing times were recorded. In a second step, the computational performances were evaluated with different graphics hardware and compared to a CPU implementation of the method using Dijkstra's algorithm. RESULTS: The ICC for inter-observer reproducibility of renal volume measurements was 0.998 (0.997-0.999) for two radiologists and the absolute mean difference between the two readers was lower than 1.2% of averaged renal volumes. The validity analysis shows an excellent agreement of our method with the results provided by a supervised segmentation method, used as reference. CONCLUSIONS: The formulation of the FBA in the form of a CA is simple, efficient and straightforward, and can be implemented in low cost vendor-independent graphics hardware. The method can efficiently be applied to perform organ segmentation and quantitative evaluation in clinical routine.