The research groups in computer vision, graphics, and machine learning have dedicated a substantial amount of attention to the areas of 3D object reconstruction, augmentation, and registration. Deep learning is the predominant method used in artificial intelligence for addressing computer vision challenges. However, deep learning on three-dimensional data presents distinct obstacles and is now in its nascent phase. There have been significant advancements in deep learning specifically for three-dimensional data, offering a range of ways to address these issues. This study offers a comprehensive examination of the latest advancements in deep learning methodologies. We examine many benchmark models for the tasks of 3D object registration, augmentation, and reconstruction. We thoroughly analyse their architectures, advantages, and constraints. In summary, this report provides a comprehensive overview of recent advancements in three-dimensional deep learning and highlights unresolved research areas that will need to be addressed in the future.

Generative Adversarial Networks (GANs) for 3D volume generation and reconstruction, such as shape generation, visualization, automated design, real-time simulation, and research applications, are receiving increased amounts of attention in various fields. However, challenges such as limited training data, high computational costs, and mode collapse issues persist. We propose combining a Variational Autoencoder (VAE) and a GAN to uncover enhanced 3D structures and introduce a stable and scalable progressive growth approach for generating and reconstructing intricate voxel-based 3D shapes. The cascade-structured network involves a generator and discriminator, starting with small voxel sizes and incrementally adding layers, while subsequently supervising the discriminator with ground-truth labels in each newly added layer to model a broader voxel space. Our method enhances the convergence speed and improves the quality of the generated 3D models through stable growth, thereby facilitating an accurate representation of intricate voxel-level details. Through comparative experiments with existing methods, we demonstrate the effectiveness of our approach in evaluating voxel quality, variations, and diversity. The generated models exhibit improved accuracy in 3D evaluation metrics and visual quality, making them valuable across various fields, including virtual reality, the metaverse, and gaming.

Alcohol dependence continues to be a major global burden despite significant research progress and treatment development. The aim of this study was to investigate whether neurofeedback training can alter resting state fMRI activity in brain regions that play a crucial role in addiction disorders in patients with alcohol dependence. For this purpose, a total of 52 patients were recruited for the present study, randomized, and divided into an active and a sham group. Patients in the active group received three sessions of neurofeedback training. We compared the resting state data in the active group as part of the NF training on six measurement days. When comparing the results of the active group from neurofeedback day 3 with baseline 1, a significant reduction in activated voxels in the ventral attention network area was seen. This suggests that reduced activity over the course of therapy in subjects may lead to greater independence from external stimuli. Overall, a global decrease in activated voxels within all three analysed networks compared to baseline was observed in the study. The use of resting-state data as potential biomarkers, as activity changes within these networks, may be to help restore cognitive processes and alcohol abuse-related craving and emotions.

Real-time three-dimensional echocardiography (RT3DE) is increasingly available in the veterinary field due to continuous reduction in costs and improvement of equipment. Much like its motion-mode and bi-dimensional counterparts, acquisition and analysis of RT3DE images and datasets is greatly improved by a thorough understanding of the technological aspects, basic physic principles, and knowledge of available modalities with their advantages and drawbacks. In this review, the authors aim to describe how the currently available RT3DE technology has evolved, explain technical aspects of the equipment, and illustrate the most commonly available modalities for image acquisition and visualization.

Obtaining a 3D feature description with high descriptiveness and robustness under complicated nuisances is a significant and challenging task in 3D feature matching. This paper proposes a novel feature description consisting of a stable local reference frame (LRF) and a feature descriptor based on local spatial voxels. First, an improved LRF was designed by incorporating distance weights into Z- and X-axis calculations. Subsequently, based on the LRF and voxel segmentation, a feature descriptor based on voxel homogenization was proposed. Moreover, uniform segmentation of cube voxels was performed, considering the eigenvalues of each voxel and its neighboring voxels, thereby enhancing the stability of the description. The performance of the descriptor was strictly tested and evaluated on three public datasets, which exhibited high descriptiveness, robustness, and superior performance compared with other current methods. Furthermore, the descriptor was applied to a 3D registration trial, and the results demonstrated the reliability of our approach.

OBJECTIVE: To compare unilateral extrapedicular vertebroplasty (UEV) and bilateral transpedicular vertebroplasty (BTV) by quantitatively calculating the structural changes of fractured vertebral body after percutaneous vertebroplasty (PVP) using 3-dimensional voxel-based morphometry (VBM).
METHODS: We calculated bone cement volume (BCV); vertebral body volume (VBV); leaked intradiscal BCV; and spatial, symmetric, and even bone cement distribution (BCD) in and out of 222 vertebral bodies treated with 2 different PVPs using VBM and evaluated the incidence of subsequent vertebral compression fracture (SVCF). Statistical analyses were conducted to compare values between the 2 different PVPs.
RESULTS: Relative BCV, which is a potential risk factor for SVCF, was higher in the BTV group based on the data using VBM (0.22±0.03 vs. 0.29±0.03; p<0.001, t-test); however, the SVCF incidence between the 2 surgeries was not significantly different (UEV, 24.7%; BTV, 31%; p=0.046, chi-square test). Spatial, even, and symmetric BCD along the 3 axes was not significantly different between UEV and BTV using VBM (x, y, z-axis, p=0.893, p= 0.590, p=0.908 respectively, chi-square test).
CONCLUSIONS: Contrary to intuitive concerns, UEV can inject a sufficient and more optimal BCV than BTV. Additionally, it can inject bone cement spatially, symmetrically, and evenly well-distributed without an increased rate of intradiscal leakage and SVCF compared with BTV based on VBM. Therefore, UEV could be a superior alternative surgical method with similar clinical effectiveness and safety, considering the above results and the consensus that UEV is less invasive.

The paper presents a new algorithm for reflection symmetry detection, which is specialized to detect maximal symmetric patterns in an Earth observation (EO) dataset. First, we stress the particularities that make symmetry detection in EO data different from detection in other geometric sets. The EO data acquisition cannot provide exact pairs of symmetric elements and, therefore, the approximate symmetry must be addressed, which is accomplished by voxelization. Besides this, the EO data symmetric patterns in the top view usually contain the most useful information for further processing and, thus, it suffices to detect symmetries with vertical symmetry planes. The algorithm first extracts the so-called interesting voxels and then finds symmetric pairs of line segments, separately for each horizontal voxel slice. The results with the same symmetry plane are then merged, first in individual slices and then through all the slices. The detected maximal symmetric patterns represent the so-called partial symmetries, which can be further processed to identify global and local symmetries. LiDAR datasets of six urban and natural attractions in Slovenia of different scales and in different voxel resolutions were analyzed in this paper, demonstrating high detection speed and quality of solutions.

Dose - volume histograms have been historically used to study the relationship between the planned radiation dose and healthy tissue damage. However, this approach considers neither spatial information nor heterogenous radiosensitivity within organs at risk, depending on the tissue. Recently, voxel-wise analyses have emerged in the literature as powerful tools to fully exploit three-dimensional information from the planned dose distribution. They allow to identify anatomical subregions of one or several organs in which the irradiation dose is associated with a given toxicity. These methods rely on an accurate anatomical alignment, usually obtained by means of a non-rigid registration. Once the different anatomies are spatially normalised, correlations between the three-dimensional dose and a given toxicity can be explored voxel-wise. Parametric or non-parametric statistical tests can be performed on every voxel to identify the voxels in which the dose is significantly different between patients presenting or not toxicity. Several anatomical subregions associated with genitourinary, gastrointestinal, cardiac, pulmonary or haematological toxicity have already been identified in the literature for prostate, head and neck or thorax irradiation. Voxel-wise analysis appears therefore first particularly interesting to increase toxicity prediction capability by identifying specific subregions in the organs at risk whose irradiation is highly predictive of specific toxicity. The second interest is potentially to decrease the radio-induced toxicity by limiting the dose in the predictive subregions, while not decreasing the dose in the target volume. Limitations of the approach have been pointed out.

Statistical lesion-symptom mapping is largely dominated by frequentist approaches with null hypothesis significance testing. They are popular for mapping functional brain anatomy but are accompanied by some challenges and limitations. The typical analysis design and the structure of clinical lesion data are linked to the multiple comparison problem, an association problem, limitations to statistical power, and a lack of insights into evidence for the null hypothesis. Bayesian lesion deficit inference (BLDI) could be an improvement as it collects evidence for the null hypothesis, i.e. the absence of effects, and does not accumulate α-errors with repeated testing. We implemented BLDI by Bayes factor mapping with Bayesian t-tests and general linear models and evaluated its performance in comparison to frequentist lesion-symptom mapping with a permutation-based family-wise error correction. We mapped the voxel-wise neural correlates of simulated deficits in an in-silico-study with 300 stroke patients, and the voxel-wise and disconnection-wise neural correlates of phonemic verbal fluency and constructive ability in 137 stroke patients. Both the performance of frequentist and Bayesian lesion-deficit inference varied largely across analyses. In general, BLDI could find areas with evidence for the null hypothesis and was statistically more liberal in providing evidence for the alternative hypothesis, i.e. the identification of lesion-deficit associations. BLDI performed better in situations in which the frequentist method is typically strongly limited, for example with on average small lesions and in situations with low power, where BLDI also provided unprecedented transparency in terms of the informative value of the data. On the other hand, BLDI suffered more from the association problem, which led to a pronounced overshoot of lesion-deficit associations in analyses with high statistical power. We further implemented a new approach to lesion size control, adaptive lesion size control, that, in many situations, was able to counter the limitations imposed by the association problem, and increased true evidence both for the null and the alternative hypothesis. In summary, our results suggest that BLDI is a valuable addition to the method portfolio of lesion-deficit inference with some specific and exclusive advantages: it deals better with smaller lesions and low statistical power (i.e. small samples and effect sizes) and identifies regions with absent lesion-deficit associations. However, it is not superior to established frequentist approaches in all respects and therefore not to be seen as a general replacement. To make Bayesian lesion-deficit inference widely accessible, we published an R toolkit for the analysis of voxel-wise and disconnection-wise data.

Lightweight bioinspired structures are extremely interesting in industrial applications for their known advantages, especially when Additive Manufacturing technologies are used. Lattices are composed of axial elements called ligaments: several unit cells are repeated in three directions to form bodies. However, their inherent structure complexity leads to several problems when lattices need to be designed or numerically simulated. The computational power needed to capture the overall component is extremely high. For this reason, some alternative methodologies called homogenization methods were developed in the literature. However, following these approaches, the designers do not have a local visual overview of the lattice behavior, especially at the ligament level. For this reason, an alternative mono-dimensional (1D) modeling approach, called lattice-to-1D is proposed in this work. This method approximates the ligament element with its beam axis, uses the real material characteristics, and gives the cross-sectional information directly to the solver. Several linear elastic simulations, involving both stretching and bending dominated unit cells, are performed to compare this approach with other alternatives in the literature. The results show a comparable agreement of the 1D simulations compared with homogenization methods for real tridimensional (3D) objects, with a dramatic decrease of computational power needed for a 3D analysis of the whole body.