BACKGROUND AND OBJECTIVES:The Finite Element Method (FEM) is a cornerstone of biomechanical analysis in dental implantology. However, the accuracy of FEM predictions is highly dependent on the underlying model, particularly the resolution of the input computed tomography (CT) images and the assigned material properties. A lack of standardized guidelines creates uncertainty about how these modeling choices affect outcomes, hindering the comparison of results across studies. This study aims to systematically quantify the impact of CT image resolution and material model selection on the predicted biomechanics of a bone-implant system. METHODS:Twenty-three computational models of a dental implant within a human mandible segment were developed and analyzed. The models were based on three micro-CT scan resolutions and featured two different geometric representations of cancellous bone. A range of material properties was assigned, including both homogeneous and non-homogeneous (density-based) Young's moduli. Peri-implant bone strain distribution, maximum implant stress, and displacement were evaluated. RESULTS:Principal Component Analysis demonstrated a clear segregation of models into distinct clusters, primarily driven by the geometric representation (trabecular vs. non-trabecular) and secondarily by image resolution. Low-resolution (150 µm) trabecular models predicted substantially higher strains and displacements than high-resolution (30 and 60 µm) models. Axial displacement values of the dental implant ranged from 8 to 22 µm in the 30 µm models and from 19 to 55 µm in the 150 µm models, while stresses ranged from 161 to 164 MPa and from 196 to 226 MPa, respectively. Notably, simplified non-trabecular models with non-homogeneous material properties could approximate the strain distributions of some complex trabecular models, but failed to accurately predict implant stress. CONCLUSION:There is no single "best" model; the optimal choice must be tailored to the specific research question. While high-resolution trabecular models provide the most comprehensive data, simplified non-trabecular models can be an efficient alternative for bone strain analysis. However, they are inadequate for predicting implant stress. These findings provide a crucial framework for developing standardized, application-specific modeling protocols in dental implantology.
BACKGROUND:Mobility impairment is influenced by intrinsic Parkinson's disease (PD) factors but also by extrinsic/environmental factors, such as indoor vs. outdoor locations as well as terrain type. Most existing studies address human activity recognition (HAR) and terrain classification separately and often focus on healthy cohorts. There remains a need for an integrated framework that enables contextual mobility assessment in people with PD (PwPD) using wearable sensors. METHODS:In this exploratory pilot study, we developed a unified multimodal wearable framework based on a one-dimensional convolutional neural network (1D-CNN) to perform HAR and terrain recognition in PwPD. A local dataset was collected from ten PwPD using synchronised inertial measurement units (IMUs) and surface electromyography (sEMG) sensors positioned on the lower back and lower limbs. The model was evaluated under three sensor configurations (M1, lower-back IMU; M2, individual IMU windows pooled across four lower-limb sensor locations; M3, corresponding IMU+sEMG windows pooled across the same locations). External benchmark evaluation was conducted using the UCI-HAR, WISDM and Uneven Walking Surface IMU datasets to assess the applicability of the same architecture across independent datasets. RESULTS:On the local dataset, HAR accuracy increased from 0.786 with M1 to 0.889 with M3, while terrain classification accuracy increased from 0.821 to 0.881. Participant-level analysis showed significant differences across configurations for activity accuracy (p = 0.032, W = 0.383), precision (p = 0.001, W = 0.753) and F1-score (p = 0.008, W = 0.531). For terrain classification, significant differences were observed for recall and F1-score (both p = 0.025, W = 0.370). When independently retrained on the external datasets, the same architectural design achieved accuracies of 0.976 on UCI-HAR, 0.981 on WISDM and 0.875 on the Uneven Walking Surface dataset, demonstrating the applicability of the same architecture across independent datasets. CONCLUSION:The proposed framework supports both HAR and terrain classification in PwPD using a common architecture and provides a basis for combining activity and environmental context in wearable mobility assessment. The findings provide preliminary evidence of performance differences across sensor configurations.
BACKGROUND AND OBJECTIVE:Space-occupying lesions of the head and neck can obstruct the aerodigestive tract and involve critical neurovascular structures, potentially resulting in life-threatening complications such as sepsis or acute airway compromise. Despite their clinical importance, publicly available medical imaging benchmarks primarily focus on malignant tumors and largely overlook other clinically relevant lesion types. METHODS:To address this limitation, we present 3D-HNSeg, a curated dataset of 204 head and neck CT scans with expert manually annotated voxel-wise 3D ground-truth of three lesion categories, including tumors, cysts, and abscesses. To address the substantial anatomical variability and heterogeneous appearance of these lesions, we propose TD-Mamba, a synergistic and spatially adaptive segmentation architecture. This model integrates Tri-oriented Dilated Mamba blocks for efficient multi-scale 3D contextual representation and Soft Signal-Adaptive Memory modules for token-wise feature refinement via adaptive memory gating. RESULTS:Experiments on the 3D-HNSeg dataset demonstrate that TD-Mamba outperforms baseline methods, achieving a Dice Similarity Coefficient (DSC), mean Intersection over Union (mIoU), and 95th percentile Hausdorff Distance (HD95) of 36.31%, 30.10%, and 21.59 mm, respectively. Ablation studies further validate the contribution of each architectural component. Although TD-Mamba improves over the evaluated baselines, its absolute segmentation performance remains substantially lower than inter-observer agreement, particularly for abscesses. Therefore, the model should be interpreted as a research baseline rather than an autonomous clinical solution. CONCLUSION:These results demonstrate the effectiveness of TD-Mamba as a strong baseline on the proposed 3D-HNSeg benchmark, supporting future research on automated head and neck lesion segmentation and related computer-assisted clinical applications. The dataset and source code are available at https://github.com/drthaodao3101/3D-HNSeg.
Background Slowed cardiac conduction velocity (CV) is a substrate for arrhythmia. Reported CV values from optically mapped whole-heart preparations vary widely in the literature, suggesting that analytical variability may contribute to inconsistency. Objective To determine how variation in vector inclusion criteria affects conduction velocity measurements and their interpretation, and to establish a principled, constraint-based approach for evaluating parameter sensitivity in the absence of ground truth. Methods CV and AR were quantified from optically mapped, paced Langendorff-perfused guinea pig hearts under four interventions: time control (N=7), gap junctional uncoupling (carbenoxolone, N=5), sodium channel inhibition (flecainide, N=7), and ephaptic disruption (mannitol, N=5). A semi-automated algorithm evaluated 441 combinations of dilation and angle. Paired t-tests compare baseline and intervention values across parameter combinations. Inclusion criteria were applied to determine parameter combinations constrained by minimal processing errors and consistency with controls. Results : Manual analysis produced significant inter-analyst variability, whereas the semi-automated approach eliminated analyst dependence. Increasing dilation or angle increased transverse CV (CVT) and decreased longitudinal CV (CVL), often altering AR and affecting statistical significance of intervention effects. Parameter selection determined whether anisotropic conduction slowing was significant. Parameter combinations meeting inclusion criteria were identified which minimized processing errors, produced significant differences between intervention and control and eliminated analyst variability. Conclusion : Vector inclusion parameters substantially influence reported conduction velocity and anisotropy, and can alter statistical conclusions. A constraint-based, parsimonious approach to parameter selection improves robustness and transparency without invoking optimization, providing a framework adaptable to other analytical workflows lacking ground truth.
BACKGROUND AND OBJECTIVE:Electrocardiogram modeling supports signal synthesis, analysis, and data augmentation. Existing generative approaches primarily focus on waveform fidelity and diversity, with limited interpretability of the underlying generation mechanisms, whereas disentangled representation learning improves interpretability through semantically structured latent factors but has been less explored from a generative perspective. METHODS:We propose an electrocardiogram modeling framework that integrates disentangled representation learning with conditional generative modeling to support interpretable and controllable ECG synthesis. Disentangled latent representations are learned using a β-Total Correlation Variational Autoencoder, and a self-attention mechanism is incorporated to promote semantic alignment between latent dimensions and signal morphological characteristics. A conditional flow module is further introduced in the latent space to enable condition-aware prior reshaping, mitigating distributional collapse while preserving the learned disentangled structure. RESULTS:Experimental results demonstrate that the proposed framework mitigates the generation collapse induced by strong disentanglement learning. While preserving a morphology-oriented latent structure, as indicated by a proxy-based mutual information gap score of 0.53, the conditional flow module improves generative performance. In an out-of-domain evaluation where only the flow is fine-tuned on the target dataset, the framework achieves a fidelity of 0.66 and diversity of 0.80, compared with 0.33 and 0.32 under standard prior sampling without flow adaptation. CONCLUSIONS:In summary, the proposed framework supports interpretable and controllable electrocardiogram generation through disentangled latent representations and condition-aware distribution alignment. The results demonstrate partial semantic consistency under out-of-domain evaluation and show that latent-space flow adaptation can mitigate generation collapse to some extent under strong disentanglement constraints.
OBJECTIVE:To investigate the internal mechanical response of the tibial shaft region of interest (ROI) during the landing-to-take-off phase of drop jumps (DJs) performed from different drop heights using musculoskeletal modeling and dynamic finite element analysis (FEA), and to examine the relationship between external and internal loading. METHODS:Seventeen healthy strength-trained males performed DJs from seven heights (30-90 cm). Kinematics, vertical ground reaction force (vGRF), and surface electromyography were collected synchronously. OpenSim combined with CEINMS was used to estimate dynamic muscle forces, which were applied to subject-specific dynamic finite element models of the tibia and fibula. P95 MaxP-stress, P95 MaxP-strain, P5 MinP-stress, P5 MinP-strain, P95 vM-strain, high-strain volume (HSV) above 3000 με, and HSV Percent were extracted within the ROI. SPM1D, repeated-measures statistics, and correlation analyses were performed. RESULTS:From 30 to 90 cm, peak P95 vM-strain increased from 1558.68 ± 105.80 to 2915.54 ± 148.86 με, HSV increased from 1278.59 ± 375.81 to 6599.91 ± 1205.36 mm³, and peak P5 MinP-stress changed from -21.83 ± 3.84 to -44.39 ± 2.99 MPa (p < 0.001;ηp2 = 0.313-0.448). P95 MaxP-strain and P5 MinP-strain also differed significantly across heights (p < 0.001), whereas P95 MaxP-stress did not (p = 0.16). Time-series differences occurred mainly during the early-to-middle portion of the landing-to-take-off phase. Peak vGRF was correlated with peak P95 vM-strain at 30-70 cm (R = 0.728-0.959, p ≤ 0.011), but not at 80-90 cm (p ≥ 0.153). CONCLUSIONS:Experimentally measured peak vGRF and FE-derived tissue-level variables provide complementary information regarding tibial mechanical loading during DJs. The findings support the interpretation of relative tibial loading differences among drop-height conditions but should not be regarded as direct evidence for an optimal DJ height.
Background and Objective Quantum machine learning is emerging as a promising extension of artificial intelligence, with potential advantages over classical approaches in handling complex biomedical data. This review aims to evaluate quantum machine learning applications in the detection, prediction, and personalized management of metabolic syndrome and type 2 diabetes mellitus. Methods We reviewed literature published between 1994 and 6 July 2026, with peer-reviewed journal articles and conference proceedings as the principal evidence base, complemented by selected preprints and technical sources. Quantum machine learning approaches, including quantum support vector machines, quantum neural networks, and quantum echo state networks, were classified and compared with classical counterparts across obesity and early metabolic dysregulation, diabetes diagnosis and glycemic management, and chronic complications. Results Quantum machine learning and hybrid quantum-classical systems demonstrated potential benefits in small-sample and noisy environments typical of wearable and biomedical sensor data. Reported performance gains included improvements in accuracy, robustness, and scalability, though interpretability and reproducibility remain challenges. Hardware limitations associated with noisy intermediate-scale quantum devices, data encoding, and privacy considerations emerged as key barriers to clinical translation. Conclusions Preliminary studies highlight the promise of quantum machine learning for predictive and personalized management of metabolic syndrome and type 2 diabetes mellitus. However, successful clinical adoption will require robust validation pipelines, regulatory sandboxes, and harmonized compliance frameworks. Domain-specific evaluation metrics and transparent conformity assessments are essential to ensure trustworthy, scalable, and equitable deployment. A staged roadmap is proposed to bridge experimental progress with ethical and regulatory readiness.
BACKGROUND AND OBJECTIVE:Epilepsy is a common neurological disorder characterized by recurrent seizures that substantially affect patients' quality of life. Electroencephalography (EEG) is crucial for monitoring brain activity, but traditional spectrogram-based methods often fail to highlight critical frequency bands or capture long-term temporal dynamics. To tackle these challenges, we propose the Epileptic Spectrogram to State-Space (Epi-Spec2State), a convolutional state space model via spectrogram image sequences for prediction-oriented seizure state classification. METHODS:This study employs a private clinical stereo-electroencephalography (SEEG) dataset collected from four patients with drug-refractory epilepsy, and three public EEG datasets including Bonn, CHB-MIT, and Siena. Epi-Spec2State transforms EEG signals into enhanced time-frequency image sequences using short-time Fourier transform and nonlinear frequency mapping to highlight critical bands. It then integrates convolutional layers and pool operation with state space models to capture spatiotemporal dependencies effectively while preserving temporal dynamics through sliding windows. RESULTS:Extensive experiments with respect to patient-specific, mixed-subject, and cross-subject evaluations show the superiority of Epi-Spec2State. Across four patient-specific tasks on the clinical SEEG dataset, the accuracy and specificity reach 96.40%-96.60% and 98.20%-98.30%, respectively. Mixed-subject and cross-subject validation on public datasets further demonstrates its superiority over nine state-of-the-art methods across multiple metrics. CONCLUSION:The proposed Epi-Spec2State can effectively capture the complex spatiotemporal dependencies in EEG signals. Its ability to handle diverse EEG types highlights its potential to support seizure state analysis and pre-seizure warning in clinical practice.
BACKGROUND AND OBJECTIVE:Explainable artificial intelligence is essential for clinical adoption of deep learning models in prostate magnetic resonance imaging. Although ensemble learning can improve robustness, its impact on explanation stability, spatial consistency, and clinical interpretability remains insufficiently quantified. This study aims to evaluate post-hoc interpretability methods across both single-model and ensemble configurations, and to examine whether ensemble-based explanations provide more reliable and clinically meaningful insights than single-model explanations. Critically, this work treats interpretability as a measurable property rather than a purely qualitative visualization. METHODS:Convolutional neural networks and bagging-based ensemble models (five VGG16-based classifiers trained on bootstrap samples with replacement, aggregated by soft averaging) were trained on the public PROSTATEx dataset using T2-weighted and apparent diffusion coefficient images. Visual explanations were generated using Gradient-weighted Class Activation Mapping (Grad-CAM) and saliency maps. Lesion localization was evaluated using centroid distance and Dice similarity coefficient with expert-annotated lesion masks. An agreement metric was introduced to quantify spatial consistency between attribution methods and its relationship with prediction reliability. RESULTS:The baseline classifier achieved an area under the curve of 0.84, with sensitivity of 0.81 and specificity of 0.86. Grad-CAM localized lesion centroids with higher precision on T2-weighted images (mean error 6.93 pixels) than apparent diffusion coefficient images (mean error 16.3 pixels). Combining saliency maps and Grad-CAM improved the mean Dice score from 0.42/0.45 (individual methods) to 0.52. Ensemble-based explanations were significantly smoother and less variable than individual classifier explanations (Mann-Whitney U, Levene test, all p<0.001). The agreement metric strongly separated correctly and incorrectly classified cases (Mann-Whitney U=84.5, p<0.001; point-biserial r=0.749; ROC-AUC =0.960). CONCLUSIONS:These findings suggest that interpretability quality can be quantitatively assessed and improved through multi-method and ensemble-based analysis. The proposed agreement-driven framework enhances explanation robustness and supports reliable and transparent clinical decision support for prostate magnetic resonance imaging.
Background and objective Diffuse optical tomography (DOT) is a non-invasive imaging technique with promising biomedical applications. However, its reconstruction quality is severely limited by the ill-posed inverse problem, leading to low spatial resolution and quantitative accuracy. This study aims to improve DOT image reconstruction performance by enhancing multi-scale feature extraction and feature reuse through an advanced deep learning framework. Methods A deep neural network-based DOT reconstruction framework integrating dilated convolution and densely connected networks (DenseNet) is proposed. To systematically evaluate the contributions of dense connectivity and dilated convolution, four models involving ResNet, ResNet with dilation convolution (DResNet), DenseNet, and DenseNet with dilation convolution (DDenseNet) are constructed. Numerical simulations and physical phantom experiments are conducted under varying target sizes and absorption contrasts. Reconstruction performance was quantitatively assessed using Mean Absolute Error (MAE), Quantitativeness Ratio (QR), Contrast-to-Noise Ratio (CNR), and Structural Similarity Index Measure (SSIM). Results Both simulation and phantom results demonstrate that the proposed DDenseNet consistently outperforms the other methods. It achieves the lowest Mean Absolute Error, the highest Contrast-to-Noise Ratio, and values of Quantitativeness Ratio and Structural Similarity Index Measure closest to 1.0 across different experimental conditions. Conclusions By combining dense connections with dilated convolution, the proposed DDenseNet effectively enhances multi-scale feature learning and reconstruction accuracy. This framework provides a robust and accurate solution for high-quality DOT imaging and shows strong potential for future clinical applications.
Background and Objective: Model-informed precision dosing (MIPD) relies on drug concentration measurements to individualize dosage regimens, yet emerging point-of-care (POC) technologies may introduce substantial measurement uncertainty compared with conventional laboratory assays. Existing frameworks lack systematic tools to evaluate how such inaccuracies propagate to dosage adaptation decisions and affect therapeutic outcomes across different sampling times. We propose ETODA (error tolerance of dosage adaptation), a computational framework that constructs automatic three-dimensional error-tolerance grids to quantify the robustness of dosage decisions under measurement uncertainty. Methods: By integrating population pharmacokinetic (popPK) models, patient-specific covariates, Bayesian posterior estimation, and Monte Carlo simulations, ETODA maps the relationship between measured and true drug concentrations, sampling time, and resulting therapeutic risk. The framework was applied to imatinib and vancomycin as model drugs with distinct therapeutic targets and dosing strategies, using 50 × 50 grids from simulated steady-state concentration ranges. Results: The grids revealed drug-specific responses to discrepancies between measured and true concentrations, and highlighted the influence of sampling time on therapeutic classification. For imatinib, peak sampling at 4 h yielded the highest distribution-weighted therapeutic-target classification percentage (52.24%), while the distribution-weighted percentage of grid points in the inefficacy alarm range increased to 53.47% at 24 h. For vancomycin, AUC/MIC-based monitoring showed higher distribution-weighted therapeutic-target classification percentages and lower distribution-weighted alarm-level classification percentages than trough-based monitoring across the evaluated sampling times. Conclusions: ETODA provides a robustness-evaluation layer that operates on top of existing popPK and MIPD workflows, mapping measurement error onto downstream dose decisions and therapeutic-risk classifications. It thereby supports more robust MIPD and informs the design of POC monitoring technologies for safer, more effective precision dosing.
Background and Objective: Three-dimensional tumor spheroids are widely used in vitro models for studying tumor growth, invasion, and treatment response. Quantitative analysis remains challenging because imaging conditions vary across experiments, annotation policies differ between datasets, and large expert-corrected segmentation datasets are limited. This study introduces SpheroSeg, an open-source platform for bright-field spheroid segmentation that combines a large annotated dataset, standardized model benchmarking, and an accessible web-based analysis workflow. Methods: We introduced SpheroHQ, a dataset comprising 22,683 bright-field images with expert-corrected annotations across seven cancer cell lines, and SpheroMix, a 32,367-image corpus integrating SpheroHQ with external spheroid datasets. Eight deep-learning segmentation architectures, including convolutional, attention-based, and state-space designs, were evaluated under a unified two-stage protocol and assessed on three stratified test sets: within-distribution SpheroHQ images, externally annotated DTS images, and the held-out out-of-distribution HTS-Seg dataset. Performance was assessed using standard segmentation metrics, with bootstrap 95% confidence intervals reported in the full benchmark to separate within-dataset performance, annotation-policy effects, and out-of-distribution generalization. A complementary SegFormer analysis was performed under the same stratified evaluation protocol to assess transformer-based segmentation models. Results: Within the main eight-architecture benchmark, the best-performing models differed across test settings. CBAM-ResUNet achieved the highest within-distribution SpheroHQ IoU (0.9478), whereas MambaBot-UNet achieved the highest externally annotated DTS IoU (0.9351) and the strongest held-out HTS-Seg IoU (0.5867) within the main benchmark. The stratified analysis revealed a substantial gap between within-distribution and held-out out-of-distribution performance, indicating that the released models should be interpreted within the documented imaging and annotation conditions. In the complementary transformer analysis, SegFormer-B0 provided a fast transformer option with strong externally annotated DTS performance (IoU = 0.9449) and competitive HTS-Seg performance (IoU = 0.5139). The deployed web application provides three complementary model options: CBAM-ResUNet, SegFormer-B0, and MambaBot-UNet. Conclusions: SpheroSeg contributes a large expert-corrected spheroid dataset, a protocol-aware benchmark, and a hosted and Dockerized web application for GPU-accelerated spheroid segmentation, polygon-based correction, morphometric measurement, and export for downstream analysis. Rather than serving as a universal spheroid detector, SpheroSeg provides a reproducible reference benchmark and deployable analysis platform for the documented bright-field imaging and annotation settings.
BACKGROUND AND OBJECTIVE:Thyroid function tests (TFT) - notably free thyroxine (FT4) and thyrotropin (TSH) - are fundamental to the diagnosis and monitoring of thyroid disorders. However, different frames of reference from variations in laboratory assays calibrations for TFT result in inconsistent normative ranges that hinder direct comparison and trending, leading to clinical misinterpretations and inefficient longitudinal follow-up. METHODS:Using differential geometry and tensor calculus, we formulate inter-laboratory harmonization as a coordinate transformation between laboratory coupled FT4-TSH measurements spaces defined on a common differentiable physiological manifold endowed with metric tensor describing its reference range geometry, with polynomial mappings serving as local approximations of the transformation. Predictive performance of the transformation was evaluated using leave-one-out cross-validation and Bland-Altman agreement analysis. RESULTS:Via a training TFT dataset, the coefficients of the mapping polynomial were computed. Both affine and quadratic mappings demonstrated excellent goodness-of-fit (R2 > 0.998). However, LOOCV indicated that the affine model yielded lower prediction error than the quadratic model for the small training dataset sample, suggesting greater stability of affine mapping when paired datasets are limited. Bland-Altman analysis showed that predicted and observed values were generally in agreement. CONCLUSIONS:In practice, affine mappings may offer greater robustness when calibration datasets are small in clinics, whereas higher-order models may be advantageous for larger datasets in hospital laboratories. This geometric framework provides a principled approach for reference interval harmonization across analytical platforms while preserving physiological structure of the FT4-TSH relationship, facilitating automated assay alignment in healthcare establishments.
Background and Objectives In transcranial sonography (TCS), Parkinson’s disease (PD) patients often show substantia nigra hyperechogenicity (SN+), which is currently the most reliable imaging biomarker for early PD diagnosis through TCS. The quantitative assessment of both SN+ and midbrain is important for early PD detection. However, current segmentation methods for SN+ and midbrain depend on manual delineation and small, non-public datasets, limiting reproducibility and fair comparison of automatic segmentation techniques. Methods TCS images were retrospectively collected and underwent screening, standardization, and de-identification. Two experienced neurosonologists independently annotated midbrain and SN+ regions under a double-blind protocol. Inter-observer agreement was evaluated, and target-specific procedures were used to construct gold standard labels. A third neurosonologist independently annotated the SN+ subset, and probabilistic soft labels were generated from the three annotations. Fourteen representative models from six segmentation paradigms were evaluated through five fold cross validation at the subject level using unified region and boundary metrics. Results The resulting MSN-TCSeg dataset comprised 700 TCS images with pixel-level midbrain annotations and 370 TCS images with pixel-level SN+ annotations, with probabilistic soft labels additionally provided for SN+ subset. Inter-observer agreement was substantially higher for the midbrain than for SN+, confirming greater annotation uncertainty of SN+. Across the 14 benchmark models, midbrain segmentation achieved consistently higher accuracy and lower boundary errors, whereas SN+ segmentation remained considerably more challenging. Conclusions MSN-TCSeg provide a standardized reference platform for automated TCS analysis. They are expected to promote the development and objective evaluation of algorithms for large-scale quantification of PD-related imaging biomarkers.
BACKGROUND AND OBJECTIVE:Hepatitis B virus (HBV) remains a major global health burden, affecting ∼296 million people worldwide. Chronic infection can progress to liver cirrhosis and hepatocellular carcinoma (HCC), partly due to viral persistence mechanisms that current antivirals do not fully address, including the failure to eliminate covalently closed circular DNA (cccDNA) and the emergence of drug resistance. Hepatitis B X-interacting protein (HBXIP) is a key host factor that supports HBV persistence and promotes oncogenic processes, making it an attractive therapeutic target. METHODS:This study investigated natural phytochemicals as potential inhibitors of HBXIP using an integrated computational pipeline. A library of 343 ethnobotanically sourced phytochemicals was curated and filtered using ADMET profiling to prioritize compounds with favorable pharmacokinetic and toxicity profiles. To validate complex stability, 200 ns molecular dynamics simulations were performed. RESULTS:The shortlisted compounds were docked against HBXIP (PDB ID: 3MSH), identifying glycyrrhisoflavone, diosmetin, lignans, and luteolin as leading candidates. Glycyrrhisoflavone showed the strongest binding affinity and formed stable interactions within the HBXIP active site. Glycyrrhisoflavone displayed minimal protein-ligand fluctuations, indicating a stable complex, whereas Luteolin showed greater flexibility that may enable adaptive binding but could reduce interaction stability. RMSD, RMSF, and SASA analyses supported these observations, with glycyrrhisoflavone exhibiting the most favorable overall dynamics. CONCLUSIONS:Glycyrrhisoflavone and lignans emerge as promising natural leads for HBXIP-targeted therapies, potentially complementing existing antivirals with lower side-effect potential. The integration of multi-level computational analyses highlights the novelty of this study and supports glycyrrhisoflavone as a promising HBXIP-targeting phytochemical lead for future experimental validation.
BACKGROUND AND OBJECTIVE:Reliable probability estimates and robust performance across heterogeneous acquisition settings are important for artificial intelligence-assisted screening mammography. In federated learning, acquisition-driven non-identically distributed data can impair both discrimination and calibration across clients. This study aimed to develop a federated multi-view mammography framework that improves worst-client robustness and probability calibration while avoiding calibration procedures that require sharing instance-level prediction-label pairs. METHODS:We proposed EquiFL-X, a federated framework combining client-local masked autoencoder pretraining, worst-client-oriented aggregation, and post-hoc temperature scaling from label-aggregated histogram statistics. Evaluation followed a public-data-based protocol with explicitly documented experimental settings using the Newfoundland and Labrador Breast Screening (NLBS) Dataset (NL-Breast-Screen), in which two native-resolution acquisition domains were treated as federated clients. Examination-level predictions were obtained through permutation-invariant pooling across available views. External testing was performed on the RSNA Screening Mammography dataset using the frozen model without fine-tuning, threshold re-selection, or recalibration. Uncertainty was quantified using 95% bootstrap confidence intervals and paired bootstrap comparisons. RESULTS:Compared with standard federated averaging on pooled held-out examinations from NLBS, EquiFL-X improved the area under the receiver operating characteristic curve from 0.880 to 0.915 and improved worst-client area under the receiver operating characteristic curve from 0.860 to 0.910, while reducing expected calibration error from 7.0% to 2.5%. At the predefined operating points, sensitivity reached 0.60 at 95% specificity and specificity reached 0.80 at 90% sensitivity. External testing on the RSNA Screening Mammography dataset was restricted to threshold-free metrics. CONCLUSIONS:EquiFL-X improved pooled discrimination, worst-client robustness, and probability calibration under acquisition-driven heterogeneity while limiting calibration-time information exchange to aggregated statistics. The framework may support more reliable federated decision support for screening mammography, although broader multi-institutional validation remains necessary.
BACKGROUND AND OBJECTIVE:Adrenocortical carcinoma (ACC) is a rare but highly aggressive malignancy, and alternative splicing (AS) serves as a key post-transcriptional regulatory mechanism involved in its malignant progression. A prominent feature of ACC is metabolic reprogramming, which is closely associated with poor prognosis. However, the mechanism by which AS mediates metabolic abnormalities to promote ACC invasion and metastasis remains unclear. METHODS:We identified three splicing subtypes based on metabolism-related splicing events, which were derived by correlating metabolic flux scores (inferred via the METAFlux algorithm) with AS PSI values using multi-omics data. A 9-gene AS-based metastatic prognostic signature was constructed using integrated computational models, and subtype-specific crosstalk between AS and metabolism was further explored to clarify the underlying mechanism. RESULTS:Three splicing subtypes were identified, capturing the dynamic metabolic shift from catabolism to anabolism during ACC progression and a continuous trajectory from an immuno-metabolically active state to a proliferative and metastatic state. The 9-gene signature exhibits robust independent predictive value for ACC metastasis and prognosis. Mechanistically, activated acylglyceride metabolism in the metastasis-like ProgDiv subtype may up-regulate IGF2BP2 to modulate RPS11 exon skipping, suggesting a potential role in promoting ACC metastasis. CONCLUSIONS:This study establishes an AS-centered framework for understanding metabolic reprogramming in ACC. The 9-gene signature is a promising tool for individualized prognostic stratification of ACC patients, while the delineated AS-metabolism crosstalk furnishes valuable insights for developing targeted therapies against aggressive ACC.
Background and Objective In vitro culture systems are widely used in cell-based research, yet they predominantly rely on flat substrates that fail to capture the curved mechanical landscapes encountered by cells in vivo. While curved substrates offer enhanced physiological relevance, their adoption is often limited by fabrication complexity and cost. Here, we investigate whether flat, soft substrates can reproduce the nuclear stress profiles induced by rigid curved surfaces during cell adhesion. Methods Using an axisymmetric finite element model of a human mesenchymal stem cell (hMSC), we show that the nuclear stress states elicited by adhesion to concave geometries can be mimicked by adhesion to flat substrates, provided that substrate stiffness is properly tuned. In contrast, stress patterns associated with convex topographies cannot be reproduced through stiffness modulation, highlighting the dominant mechanobiological role of surface curvature. Results We derive a nonlinear relationship between the Young’s modulus of soft flat substrates and the curvature radius of rigid geometries, providing a predictive framework for the rational design of planar systems that replicate essential aspects of curved mechanotransductive environments. Conclusions These findings demonstrate that substrate stiffness can only partially replicate curvature-driven nuclear mechanics, identifying surface curvature as a fundamental and, in some cases, non-replaceable regulator of mechanotransduction, while providing practical guidelines for engineering cost-effective yet physiologically relevant culture systems.
BACKGROUND AND OBJECTIVE:The recognition of motor imagery electroencephalogram (EEG) signals, which non-invasively capture the macroscopic electrical activity of the brain, is critical for medical rehabilitation and intelligent control. In these applications, reliable prediction is essential due to the safety risks associated with misclassification. However, existing methods often suffer from limited generalization in cross-subject scenarios caused by substantial inter-subject variability. To address this challenge, this work develops adaptive modeling strategies to improve robust cross-subject recognition performance. METHODS:We propose a Hybrid Adaptive Domain Graph Convolutional Network (HAD-GCN) to enhance decoding performance through multi-level adaptability. At the spatial level, an adaptive generator synthesizes electrocardiogram (ECG) signals, which record the electrical activity of the heart, from EEG signals and concatenates them within a connected graph structure, thereby mitigating the limitations of non-invasive data acquisition. At the temporal level, an adaptive splitter selects the most suitable time-frequency domain processing method for each subject's signal and routes the data into the corresponding branches for feature extraction. RESULTS:Accuracy and the Kappa coefficient, which are widely adopted in motor imagery research, are used as evaluation metrics. Cross-subject experiments conducted on the Mixed dataset and the BCI Competition IV-2a dataset achieve accuracies of 83.10% ± 6.54% and 74.81% ± 8.97%, respectively, with corresponding Kappa values of 0.778 ± 0.06 and 0.655 ± 0.09. CONCLUSIONS:Experimental results demonstrate that HAD-GCN significantly improves cross-subject classification performance and prediction reliability while maintaining strong generalization capabilities. The proposed multi-level adaptive approach consistently enhances classification accuracy for individual subjects, highlighting its potential for practical applications in EEG-based technologies. Our code is available at https://github.com/chuanlaiair/HAD-GCN.
