This study investigates the effect of screw diameter on the pull-out strength of a bone screw with a loosening-resistant auxetic structure manufactured using additive manufacturing. The variation in mechanical performance depending on screw diameter was examined experimentally and through finite element analysis in cancellous and cortical bone structures. First, screw designs with different diameters incorporating an auxetic cell structure were created in SolidWorks. The designed screws were analyzed using ANSYS Workbench finite element software, confirming that the structures exhibit a negative Poisson's ratio. The screws were manufactured from Ti6Al4V alloy using the Direct Metal Laser Sintering (DMLS) additive manufacturing process. The fabricated auxetic screws were inserted into synthetic cancellous and cortical bone (L4-L5 vertebral bone). The pull-out tests were performed under conditions conforming to the ASTM F543 standard. Force-displacement curves were obtained from the experiments, and tensile tests were also conducted to determine the mechanical strength of the screws. To evaluate the performance of the auxetic screws, identical pull-out tests were carried out on conventional bone screws. After the tests, SEM images were obtained to examine the damage mechanisms in detail. The results showed that auxetic bone screws exhibited approximately 10% higher pull-out strength compared to conventional screws. This improvement is attributed to increased surface roughness and radial expansion under tensile loading, which enhance mechanical interlocking at the bone-screw interface. Finite element analyses further indicated that the magnitude of the negative Poisson's ratio increased with increasing screw diameter, suggesting more effective mechanical behavior at larger diameters. In addition, the porous structure of the auxetic screw is expected to facilitate bone ingrowth, thereby supporting bone healing and reducing the risk of loosening. Finally, the developed auxetic bone screw has the potential to increase resistance to loosening. In addition, it may enhance biological integration with bone, which could contribute to improved surgical outcomes and greater patient comfort.

During emergency transfusion, stored whole blood at 1-6°C must be rapidly warmed to 36-38°C using blood warming devices. However, device validation commonly relies on Newtonian crystalloid intravenous fluids with a viscosity of 1 mPa·s, whereas whole blood exhibits temperature-dependent, shear-thinning rheology with viscosities of 3.5-4.5 mPa·s at 37°C. This discrepancy introduces systematic errors in performance evaluation. To address this gap, this research study characterized three blood-mimicking fluids (BMFs), including glycerol-water solutions of 22-50% w/w, egg whites, and TrueClot across temperatures of 5-40°C and shear rates of 5-100 s[-1]. Among the tested fluids, TrueClot most closely resembled blood in overall rheological properties. However, TrueClot demonstrated abnormal phase separation at temperatures below 10°C, with a 28% drop in viscosity from 55 ± 4 to 39.6 ± 0.5 mPa·s over just 2°C, limiting its use at refrigerated storage temperatures. At shear rates ≥ 100 s[-1], 50% glycerol-water solutions produced viscosity values closest to reported whole blood values. Based on the rheological profiles obtained, it is inferred that at shear rates < 100 s[-1], diluted TrueClot at temperatures higher than 10°C or modified egg whites with xanthan gum at 5-40°C may provide suitable shear-thinning properties, though experimental validation of these modified solutions is required. Two mathematical models, including the Casson model with Apostolidis and Beris parametrization and a combined shear-temperature power law model, were developed to predict viscosity values that correlate well with the experimental results.

Three-dimensional ground reaction forces (GRFs) are important parameters in human motion analysis, yet their measurement is confined to laboratory settings. GRF estimation from wearable inertial measurement unit (IMU) data using machine learning and deep learning has emerged as a promising avenue for kinetic analysis outside the laboratory, but no consensus on standardized methods or best practices has emerged. This review synthesizes existing methods and provides a framework of current practices and recommendations for future work. The review followed PRISMA guidelines (PROSPERO CRD42024578576) and searched PubMed, Scopus, IEEE Xplore, and ScienceDirect from 2011 to 2026. Studies were included if they used IMUs and machine learning or deep learning to estimate GRFs and reported quantitative performance metrics. Data were extracted on participant characteristics, IMU specifications and placement, preprocessing, models, validation protocols, and error metrics. Twenty-six studies met the inclusion criteria, with sample sizes ranging from 5 to 47 (median = 15) and IMU counts ranging from 1 to 16. Most studies used triaxial accelerometers and gyroscopes sampled at 100-1,000 Hz. Vertical GRF was generally estimated more accurately than the anteroposterior and mediolateral components within the same study. A single cross-study accuracy ranking was not defensible because studies used different normalization denominators, evaluation windows, tasks, sensor configurations, and validation strategies. Heterogeneity in validation protocols and demographic samples limited direct cross-study comparisons. This review synthesizes how implementation choices, including model architecture, feature-engineering strategy, sensor configuration, and validation protocol, shape GRF prediction across axes. The variability observed underscores the need for standardized methodologies, larger and more diverse cohorts, consistent performance metrics, and benchmarking on public datasets to enable fair cross-study comparisons and accelerate clinical translation.

The Clock Drawing Test is a core tool for the early screening of cognitive impairment. Traditional manual methods are time-consuming and prone to significant subjective bias, while most existing automated methods are "black-box" approaches that directly map images to scores or category labels. Not only do these methods lack interpretability in their decision-making processes, but they also fail to provide information beyond the raw scores. Furthermore, most of these methods employ a centralized training paradigm. However, due to strict privacy regulations that restrict medical data sharing, achieving multi-center collaboration remains difficult, which hinders clinical deployment. Therefore, this paper proposes the Federated Simplex-Constrained Network (Fed-SCN), which enables collaborative training through federated learning without raw data leaving the local clients; simultaneously, by embedding a simplex-constrained module in the model, a module that serves both as a latent-feature transformation layer and as a cognitive-pattern modeling module, it represents individual cognitive patterns of samples as convex combinations of multiple cognitive prototypes, thereby constructing a task-relevant library of cognitive prototypes. This approach simultaneously achieves fine-grained scoring, interpretability of the decision-making process, and cognitive pattern analysis. Performance experiments were conducted on a dataset of 1,370 cross-sectional clock-drawing images collected from a single institution, using the Dirichlet partitioning method to simulate multi-federated clients with highly skewed label distributions. The main experimental results show that this model achieved a macro-average accuracy of 0.847 (95% CI: 0.841-0.854) and a macro-average F1 score of 0.470 (95% CI: 0.405-0.536) on the scoring task. .

Wearable inertial sensors enable repeated movement recording during knee osteoarthritis rehabilitation, but methods for summarizing multivariable changes across routine sessions remain exploratory. This retrospective technical evaluation examined a normalized transport-derived deviation (NTD) constructed from paired shank-mounted inertial measurement units. Seven participants contributed 46 paired-sensor sessions over 45 days. An approximately 30 s active movement segment from each session was divided into 13 windows and represented using 58 time-domain, frequency-domain and cross-sensor features. Session distributions were compared with each participant's first usable sensor session using an archived entropically regularized transport implementation. The output was normalized by the baseline-to-final-session value. Mean NTD values at the scheduled assessment days were 0.00, 0.02, 0.23, 0.63 and 1.00. Fourteen of the 46 displayed values were structurally constrained baseline or final values; no monotonic constraint was applied to the 32 interior observations, which formed non-decreasing trajectories. At interior scheduled days 5, 15 and 30, absolute differences between mean NTD and participant-adjusted normalized elapsed time were 0.057, 0.076 and 0.019. Twenty-one of the 46 archived calculations reached the 500-iteration limit without meeting the 1 × 10-6 residual criterion. A log-domain stabilized recomputation of all 46 primary comparisons satisfied a 1 × 10-8 criterion, with a mean absolute archived-to-stabilized NTD difference of 0.004 (maximum 0.018) across the 32 interior observations. Across support-point counts of 100 to 1000, interior archived NTD values changed by 0.009 to 0.026 on average. Routine paired-shank recordings were organized into ordered longitudinal trajectories and the resulting trajectory structure was numerically reproducible under the evaluated implementation. NTD should presently be interpreted as an implementation-specific retrospective coordinate rather than a validated rehabilitation outcome. By defining the requirements for standardized movement tasks, prospective normalization and reproducible sensor placement, these findings provide a practical foundation for prospective validation of transport-based rehabilitation assessment.

Robotic endoscopy offers a potential route to autonomous gastrointestinal navigation. However, autonomous intestinal navigation is hindered by non-rigid lumen deformation and visual degradation from specular reflections, which compromise accurate perception and stable control. We propose a threshold-triggered, vision-driven framework incorporating Navier-Stokes-based highlight suppression. Algorithmically, the system is governed by a threshold-triggered mechanism that acts as a conditional perception gate, dynamically alternating between a lightweight "dark-area guidance" algorithm for approximately centered navigation and a more detailed "contour-based guidance" pipeline for corrective steering. The switching policy avoids executing highlight suppression and morphological contour processing in every frame. In phantom-based trials, the system achieved an average speed of 3.10 cm/s on the triangular path. Highlight suppression increased the proportion of frames with normalized center error below 0.1 from 65% to 71%, below 0.2 from 95% to 97%, and below 0.3 from 98% to 100%. The adaptive mechanism decreased runtime by approximately 2% compared to continuous contour processing. Although this absolute reduction was modest, it recovered approximately 89% of the runtime difference between the contour-only and dark-area-only baselines. These phantom-based results support conditional perception allocation while keeping average processing time close to that of lightweight navigation.

Rigid limb surrogates transfer full mechanical loads directly to the prosthetic socket during structural testing, causing premature distal failure and underestimating of actual load capacity. In contrast, soft-compliant surrogates buffer applied loads and enables realistic structural deformation. This study investigates: (i) the influence of rigid Plaster of Paris (PoP) versus biofidelic hyperelastic silicone surrogates on the structural load capacity of 3D-printed transfemoral prosthetic sockets; and (ii) orientation-dependent structural deformation and failure behaviors arising from asymmetric transfemoral anatomy. Sockets encapsulated with PoP and silicone surrogates were evaluated under ISO 10328:2016 static loading (Condition II) at two loading alignments: toe-off and toe-towards-lateral (90° laterally rotated) configurations. Initial testing with 3D-printed Bespoke Commercial (BC) connectors induced premature distal connector failure; however, under toe-towards-lateral loading, socket/silicone assemblies exhibited a 4.47-fold greater displacement and a 72.46% higher ultimate force than socket/PoP assemblies. Sockets with an enlarged BC connector encapsulated in the silicone surrogate showed distinct force-displacement hysteresis loops under toe-towards-lateral loading, indicating energy dissipation. Crucially, enlarging the BC connector shifted the failure site from interlayer delamination at the distal end under toe-off to trans-layer fracture across the lateral socket wall. Conversely, socket/PoP assemblies consistently failed at the enlarged BC distal connector regardless of loading orientation. Finite element analysis (FEA) confirmed non-compliant, uniform contact pressure in socket/PoP assemblies, whereas the silicone surrogate produced compliant pressure gradients. Loading orientation significantly affected the structural deformation (p < 0.05) of socket/silicone assemblies: toe-towards-lateral loading induced two-fold greater deformation and an 18.95% higher ultimate force than under toe-off loading due to the thicker lateral silicone resulting from asymmetric limb geometry. These findings demonstrate that a biofidelic hyperelastic surrogate buffers stress and redistributes contact pressure similar to biological soft tissue, establishing a more accurate pre-clinical framework for evaluating 3D-printed prosthetic socket strength.

The accurate assessment of midfoot motion during gait presents a considerable methodological challenge in biomedical measurement and modelling. This study developed and evaluated the intra-session repeatability of the Oita foot model, a novel multi-segment kinematic foot model that incorporates a dedicated midfoot segment defined by three markers on the dorsal cuneiform and cuboid to minimize soft tissue artifacts. Foot kinematics were acquired from 20 young adult participants (20.75 ± 1.34 years) using a 10-camera three-dimensional motion-capture system. The joint angles of five foot segments were calculated across three planes during the stance phase. The repeatability of the measurements was assessed using the intraclass correlation coefficient (ICC) and coefficient of multiple correlation (CMC). Waveform similarity was compared with the Rizzoli and modified DuPont foot models. Under single-examiner conditions, this model exhibited excellent intra-session repeatability (ICC: 0.928-0.985; CMC: 0.852-0.981) and substantial waveform agreement with the Rizzoli foot model in the sagittal and frontal planes (CMC: 0.920-0.999). Under single-examiner conditions, these findings indicate that the Oita foot model may provide a useful framework for the detailed assessment of foot kinematics during gait. However, its repeatability across different examiners and clinical populations remains to be established.

This study aimed to evaluate the feasibility of estimating patient positional errors during head-and-neck fixed-field intensity-modulated radiation therapy (IMRT) using a machine-learning (ML) regression models based on radiomics features of patient-transmitted fluence maps acquired by an electronic portal imaging device (EPID), as a proof-of-concept toward quantitative intrafractional monitoring. A virtual target, spinal cord, and parotid glands were contoured on an Radiosurgery Verification Phantom simulating nine fixed-field IMRT plans. Treatment plans were created with and without a 0.5 cm or 1.0 cm bolus placed on the phantom surface. For each field, phantom positional errors were simulated in the anterior-posterior (AP), left-right (LR), and superior-inferior (SI) directions at 1 mm intervals from -5 mm to 5 mm. Fluence maps were acquired using an EPID, and difference maps were generated between error and error-free cases. A total of 837 radiomics features were extracted. The dataset included 180 cases (with no bolus or 1.0 cm bolus) for training and 90 cases (with 0.5 cm bolus) for testing. Three ML regression models-random forest, AdaBoost and extra trees-were used. Positional error detection accuracy was evaluated using root mean square error (RMSE), mean absolute error (MAE) and Spearman's rank correlation coefficient. In the test data, RMSE values ranged from 0.550-0.779 mm (AP), 0.440-0.493 mm (LR), and 0.616-0.689 mm (SI). Corresponding MAE values ranged from 0.394-0.584 mm, 0.351-0.392 mm, and 0.492-0.573 mm, respectively. Spearman correlation coefficients ranged from 0.890-0.946 (AP), 0.949-0.953 (LR), and 0.860-0.898 (SI). This phantom-based study suggests that radiomics-based ML regression can quantitatively estimate translational positional-error magnitude from EPID-measured fluence differences under controlled conditions. These findings represent a proof-of-concept for the feasibility of intrafractional geometric monitoring rather than clinical validation. Further investigation using clinically realistic patient data, anatomical changes, and day-to-day machine variations is required before routine clinical implementation.

Random or central placement of respiratory sounds within padded windows can cause length-based masks to exclude recorded audio. We developed an exact support-propagation interface for the pretrained bidirectional encoder representation from audio transformers (BEATs), mapping sample support through filterbank frames and patch tokens to attention and pooling masks. We evaluated this interface alongside adaptation strategies, computational costs, probability calibration and class-conditional coverage. The study included 15 987 events from the International Conference on Biomedical and Health Informatics (ICBHI) 2017 and SPRSound databases, using patient-disjoint training, model-selection, calibration and primary test partitions at a fixed decision threshold of 0.5. The three-seed full-fine-tuning ensemble achieved average scores (AS; mean sensitivity and specificity) of 0.6619 on ICBHI (patient-cluster bootstrap 95% confidence interval, 0.6002-0.7152) and 0.9112 on SPRSound (0.8799-0.9362). Exact mapping corrected observed support exclusion and maintained finite outputs throughout placement tests; the length-based interface produced non-finite outputs in these tests and one of three matched training runs. Classification ablations showed no consistent advantage for exact mapping across the two completed seed pairs and no incremental gain from Jensen-Shannon consistency. Under matched update counts and learning-rate schedules, full fine-tuning exceeded rank-8 low-rank adaptation (LoRA) by 3.94 percentage points in AS averaged equally across databases (paired patient-cluster bootstrap 95% confidence interval, 2.54-5.40). LoRA reduced trainable parameters and recorded training time but did not reduce inference latency in the unmerged implementation. Temperature scaling improved the primary ensemble's held-out negative log-likelihood and Brier scores without uniformly improving calibration error. Entropy-based abstention achieved low accepted-set error but failed the prespecified validation requirement for adventitious-event coverage and was not applied. The integrated evaluation quantifies performance-resource trade-offs and shows that improved probability scores do not ensure adequate class-specific coverage for selective prediction. Clinical utility requires prospective validation.

Reliable evaluation of machine learning models for neonatal EEG seizure detection is a prerequisite for accurate performance assessment and comparison of methods. Model evaluation, however, is complicated by severe class imbalance and uncertain ground truth. To address these limitations, we propose evidence-based reporting recommendations tailored to the specific challenges of neonatal seizure detection. Using real and synthetic seizure annotations, we assessed standard performance metrics, consensus strategies, and human-expert level equivalence tests under varying class imbalance, inter-rater agreement, and number of raters. Matthews and Pearson's correlation coefficients outperformed the area under the receiver operating characteristic curve in reflecting performance under class imbalance. Consensus types are sensitive to the level of agreement and number of raters. Among the evaluated human-expert level equivalence tests, the multi-rater statistical Turing test using Fleiss'demonstrated the most consistent performance and robustness under the tested experimental conditions. Based on our findings, we recommend reporting: (1) at least one balanced metric, (2) sensitivity, specificity, PPV and NPV, (3) seizure burden agreement, (4) multi-rater agreement using the statistical Turing test with Fleiss', and (5) all of the above on a held-out validation set. This proposed framework provides an important prerequisite to clinical validation by enabling a thorough and honest appraisal of machine-learning or AI methods for neonatal seizure detection.

Osteogenesis imperfecta (OI) is a rare genetic disorder but is the most common disorder of bone etiology with little known about the impact of physiologic factors on fracture risk. Finite element analysis (FE) models of OI bone have been previously developed to examine fracture risk during ambulation and various daily activities of the femur and tibia. This study aims to further investigate the impact of long bone deformity and muscle activation forces in children and adolescents with OI. Pediatric femoral FE models of normal, OI type I, and OI type III bone material properties assessed effects of lateral bowing versus increased gluteus medius and gluteus maximus force production on bone injury risk. Models incorporating increased lateral bowing and muscle forces showed up to 7.56% higher stress levels than the standard model with no bowing and normal muscle forces, with muscle forces having a greater impact than lateral bowing. This study provides quantitative information to help establish methodology for FE analysis to assess femoral fracture risk in individuals with OI.

Rigid-wall left atrium (LA) and left atrial appendage (LAA) computational fluid dynamics models are practical but suppress deformation-driven transport, whereas image-derived wall motion is data- and processing-intensive. We evaluated whether a reduced-order, volume-constrained wall-motion prescription changes computed LA/LAA hemodynamic descriptors relative to a static-wall approximation. A patient-specific LA/LAA geometry was combined with literature-informed phasic volumes and boundary conditions to simulate healthy dynamic, grade-II diastolic dysfunction (DD II) dynamic, and DD II static configurations. The healthy-DD II comparison represents a composite disease-state contrast; the DD II dynamic-static pair isolates wall-motion sensitivity within the same DD II setup. Numerical verification supported the intermediate mesh,s, five simulated cycles, and the laminar model for the monitored WSS waveform. Suppressing DD II wall motion reduced mean LAA time-averaged wall shear stress (TAWSS) from 0.05 to 0.01 Pa and increased the analyzed atrial surface fraction with TAWSSPa from 14.4% to 22.4%. Complete-wall, area-weighted medians changed consistently: TAWSS decreased from 0.570 to 0.349 Pa, whereas oscillatory shear index, relative residence time, and endothelial cell activation potential increased from 0.121 to 0.190, 2.52 to 5.02 Pa, and 0.217 to 0.539 Pa, respectively. Atm s, mean persistent volume stasis was similar in the two DD II cases, but the volume with low velocity during at least 90% of the cycle increased from 28.5% to 33.8%. Thus, suppressing prescribed deformation shifted the computed LA/LAA environment toward lower shear, greater shear-directional oscillation, and more spatially persistent low velocity. Volume-constrained motion is therefore useful as an engineering sensitivity tier when image-derived kinematics are unavailable; it is not a validated surrogate for regional atrial motion or a patient-specific predictor of thrombosis.

Portable quantification of postural sway remains challenging in remote and resource-limited environments where laboratory force platforms and motion-capture systems are impractical. We evaluated an autonomous RGB-D stabilometry workflow during an Antarctic expedition by comparing camera-derived center-of-mass (CoM) sway with a portable center-of-pressure (CoP) reference and testing leave-one-subject-out calibration. Thirteen participants performed modified Romberg trials while an Intel RealSense D455 camera and a Wii Balance Board (WBB) recorded postural data. Full-body CoM trajectories were reconstructed from three-dimensional landmarks; facial-landmark head tracking was evaluated as a secondary fallback modality. In total, 192 quality-controlled WBB-camera pairs were analyzed. Temporal-alignment diagnostics showed strong anteroposterior signal overlap with the CoP reference (median post-lag Pearson= 0.847-0.929 for full-body tracking;= 0.825-0.905 for head tracking). Raw full-body metrics underestimated root-mean-square anteroposterior sway (bias =5.62 mm) and overestimated mean velocity (bias = 8.67 mm s). Passing-Bablok calibration reduced RMS AP bias relative to the WBB reference to 0.01 mm and narrowed the width of the limits of agreement from 12.15 to 7.15 mm. After empirical mapping, the inter-method reliability of Rambling RMS AP sway relative to the WBB reference was excellent (ICC(2,1) = 0.956), whereas mediolateral outcomes were less reliable and should be interpreted as exploratory in this camera configuration, likely reflecting lower ML amplitude and three-quarter camera geometry. Clock-based fixed windows yielded lower correlations, whereas a single global acquisition-latency correction applied identically to all trials, without per-trial window optimization, recovered correlations close to those of the optimized analysis. All estimates are relative to the portable WBB field reference rather than a laboratory-grade standard, and the calibration is a task- and setup-specific empirical mapping, not a universal biomechanical transformation.

Background.Accurate three-dimensional dynamic assessment of the thumb carpometacarpal (CMC) joint is important for understanding the pathomechanics of CMC osteoarthritis and evaluating treatment outcomes. Intensity-based biplane 2D-3D registration is a promising approach for dynamic bone motion analysis; however, conventional orthogonal biplane configurations are often difficult to implement for small-joint imaging in clinical settings.Methods.We validated the accuracy of intensity-based biplane 2D-3D registration for thumb CMC joint kinematic analysis using a phantom model under a low-angle, non-orthogonal biplane fluoroscopic configuration (53°). Registration-derived bone poses were compared with radiostereometric analysis-derived ground truth across 252 biplane fluoroscopic image pairs during palmar and radial abduction tasks. Translational and rotational errors were calculated for the thumb proximal phalanx, first metacarpal, trapezium, and second metacarpal.Results.The method achieved submillimeter mean overall component-wise translational errors for all evaluated bones (range: 0.11-0.31 mm). Mean overall component-wise rotational errors were below 3° for all bones (range: 0.44-2.64°). Larger axis-specific rotational root mean square errors were observed for the trapezium and thumb proximal phalanx, reflecting bone- and axis-dependent error patterns.Conclusion.Intensity-based biplane 2D-3D registration can provide sufficient accuracy for dynamic analysis of the thumb CMC joint using a low-angle, non-orthogonal biplane configuration. This approach may serve as a clinically feasible platform for assessing small-joint kinematics in settings where conventional orthogonal biplane systems are difficult to implement, supporting futurein vivostudies of dynamic thumb CMC joint motion.