Thanh-Hung Nguyen

Cranial reconstruction is a critical task in computer-assisted surgery, requiring both high geometric accuracy and computational efficiency for patient-specific implant design. While recent deep learning approaches, particularly 3D UNet-based models, have demonstrated promising performance, most studies primarily focus on architectural modifications, with limited attention to the systematic impact of data preparation and training strategies on reconstruction quality. In this study, we present a comprehensive data-centric investigation of key factors influencing the performance of a baseline 3D UNet for cranial defect reconstruction. Specifically, we analyze the effects of data preprocessing (denoising), dataset organization (ordered versus randomized training), defect morphology diversity, convolutional kernel size, and loss function design under controlled experimental conditions. Experiments were conducted on 250 complete skulls (NRRD format) from the MUG500+ dataset, with synthetically generated defects across multiple anatomical regions. From these volumes, a total of 3750 training samples were generated, including: (i) 1250 noisy samples with diverse defect morphologies, (ii) 1250 denoised samples with ellipsoidal defects, and (iii) 1250 denoised samples with multiple defect types. The results demonstrate that data-centric and training-related factors have a substantial impact on model performance, in several cases exceeding the influence of architectural design. In particular, denoising significantly improves boundary stability and reduces geometric error, while incorporating diverse defect morphologies enhances generalization to unseen shapes. Additionally, ordered training contributes to more stable convergence, and an optimal kernel size of (3 × 3 × 3) achieves the best trade-off between accuracy and computational efficiency. A hybrid Dice and boundary loss further improves boundary precision compared to conventional Dice loss. The optimized configuration achieves a Dice Similarity Coefficient of 0.94 and a Hausdorff Distance of 3.8 mm, with an average inference time of 0.004 s per case. These results demonstrate that data-centric optimization can be as influential as, or even more impactful than, architectural design in cranial defect reconstruction. The findings provide practical and reproducible guidelines for developing efficient, robust, and clinically applicable deep learning-based systems for patient-specific cranial implant design.

Compass needle oscillations reduce navigation accuracy and increase stabilization time. Existing estimation methods lack automation, robustness, and real‐time capability. This paper presents a real‐time, noncontact vision system for estimating the oscillation parameters of a magnetic compass needle using deep learning. The system employs a camera to continuously capture images of the needle. A lightweight convolutional neural network (YOLO11s) is used to detect the pivot and tip positions in real time. From these positions, the angular displacement is computed and tracked over time. Two mathematical models—a standard damped sinusoidal model and a modified oscillation model—are applied to fit the angle‐time data and estimate key oscillation parameters, including amplitude, angular frequency, and damping coefficient. The proposed system is evaluated against traditional image processing and object tracking methods under various lighting conditions, camera resolutions, and frame rates. The results demonstrate high detection accuracy (mAP 99.5% at IoU 0.5), strong tracking robustness (MOTA 95.7%, MODA 99.9%), and real‐time processing at 18 FPS. The modified model reduces fitting residuals and achieves average parameter estimation errors below 4%. The proposed framework provides an automated, high‐precision alternative to traditional methods and shows strong potential for embedded navigation systems, robotics stabilization, and physics education tools.

Cranioplasty is a standard neurosurgical procedure performed on patients following decompressive craniectomy. This study investigates the influence of implant thickness and material on biomechanical performance through finite element analysis. Cranial implants composed of Polyether Ether Ketone (PEEK), Polymethyl Methacrylate (PMMA), and Ti6Al4V with thicknesses in the range of 1 mm to 5 mm were systematically evaluated under three loading conditions. The findings indicate that increased implant thickness significantly reduces both deformation and von Mises stress. Implants with thicknesses of 1 mm and 2 mm exhibited the greatest deformation, potentially surpassing the material’s yield strength. Maximum stress was predominantly localized at the interface between the implant fixation points and the skull, with implant thickness exerting a more pronounced effect than material type. Notably, implants with a 4 mm thickness in PEEK and PMMA demonstrated optimal mechanical performance, maintaining strain values at screw locations within the safe range of 200 to 2500 μ ε , thereby promoting bone formation and growth. Additionally, implant micromotion (IMM) was assessed to determine the biological durability of the implants. This analysis provides critical insights into cranial implants’ biocompatibility and long-term mechanical stability.

The A* algorithm is a cornerstone in mobile robot navigation. However, the traditional A* suffers from key limitations such as poor path smoothness, lack of adaptability to dynamic environments, and high computational costs in large-scale maps. This review presents a comprehensive analysis of 20 recent studies (2020–2025) on improved A* variants and their hybrid integrations with complementary algorithms. The improvements are categorized into two core strategies: (i) geometric and structural optimization, heuristic weighting and adaptive search schemes in A* algorithm, and (ii) hybrid models combining A* with local planners such as Dynamic Window Approach (DWA), Artificial Potential Field (APF), and Particle Swarm Optimization (PSO). For each group, the mathematical formulations of evaluation functions, smoothing techniques, and constraint handling mechanisms are detailed. Notably, hybrid frameworks demonstrate improved robustness in dynamic or partially known environments by leveraging A* for global optimality and local planners for real-time adaptability. Case studies with simulated grid maps and benchmark scenarios show that even marginal improvements in path length can coincide with substantial gains in safety and directional stability. This review not only synthesizes the state of the art in A*-based planning but also outlines design principles for building intelligent, adaptive, and computationally efficient navigation systems.

The rocker arm plays a crucial role in an internal combustion engine’s valve train system. Issues often arise when there are errors in the production of the rocker arm, such as the following: (1) misalignment of the rocker arm spindle hole and thread hole; and (2) bevel errors in the rocker arm spindle hole. This paper focuses on the development of a rocker arm chamfering and coaxial defect inspection system, which utilizes the RANSAC algorithm. The algorithm consists of five steps: (i) representing a 2D point cloud as a matrix with X and Y coordinates; (ii) detecting the number of circles present in the input point cloud; (iii) setting a tolerance threshold for the distance between the selected circle and other points, crucial for achieving high accuracy; (iv) specifying the maximum number of iterations to find the best model from the input points; (v) defining the maximum number of points that can belong to the same circle, considering point density and the anticipated circle’s surface. The real-time inspection system is designed with a camera mounted on top of the box, and two LED arrays positioned around the box at suitable distances. Test results using a collected data set demonstrate the effectiveness and accuracy of the proposed method in detecting each type of error. In the experiments, the proposed method successfully detected all coaxial defect samples and achieved a detection rate of 95.1% for chamfer defect samples.