Abstract Breast ultrasound is widely used for lesion characterization, yet reported deep-learning performance varies substantially with dataset composition, preprocessing, and validation design. A major source of bias arises when patient-level separation is not enforced, allowing correlated images from the same subject to inflate performance estimates. This study presents a leakage-aware benchmark of convolutional neural networks (CNNs), a Vision Transformer (ViT), and a CNN–transformer late-fusion configuration for benign-versus-malignant breast ultrasound classification on the BUS-UCLM dataset. After exclusion of normal-category images, the final cohort comprised 264 images from 36 patients, including 174 benign and 90 malignant images. Seven CNN-family models, one ViT baseline, and one ResNet18–ViT probability-level late-fusion configuration were evaluated using strict patient-level fivefold cross-validation. Additional analyses included fusion ablation, paired Wilcoxon signed-rank testing, gradient-weighted class activation mapping (Grad-CAM) visualization, and an image-level leakage demonstration. Under strict patient-level evaluation, performance was moderate across all models. GoogLeNet achieved the highest mean accuracy (61.48%), InceptionV3 achieved the highest mean macro-F1 (59.24%), and ResNet50 achieved the highest mean area under the receiver operating characteristic curve (AUC) (0.6699), whereas the standalone ViT showed weaker overall discrimination. The late-fusion configuration remained competitive in threshold-dependent metrics but did not surpass the strongest CNN baselines in AUC. Overall, no architecture demonstrated a clear advantage across both threshold-dependent and threshold-independent metrics. By contrast, image-level splitting substantially inflated apparent performance, underscoring the importance of rigorous patient-level separation for credible benchmarking in breast ultrasound classification.
Abstract The integration of biomechanics into nursing practice is vital for enhancing patient care, especially in surgical contexts. Surgical procedures significantly impact a patient's physical capabilities and recovery, necessitating evidence-based nursing strategies grounded in biomechanical principles to optimize outcomes. A narrative review was conducted by searching PubMed, Web of Science, and CINAHL for articles published between 2010 and 2023. Keywords included “biomechanics,” “nursing interventions,” “pre-operative care,” “postoperative care,” and “human movement.” Studies were included if they focused on biomechanical applications in adult surgical nursing. Data were synthesized thematically to identify key intervention strategies and their outcomes. The review indicates that biomechanical principles directly inform personalized nursing interventions. Pre-operatively, tailored exercises improved lumbar range of motion (ROM) by 15% and quadriceps strength by 20% (p < 0.05). Postoperatively, biomechanics-guided pain management, posture optimization, and range of motion control reduced analgesic use by 40% and pressure sore incidence by 20%, and improved joint flexion by 15% with fewer movement-related overextension incidents. Furthermore, the application of biomechanical knowledge facilitated the development of structured postoperative mobility standards and informed equipment use (e.g., support belts, walkers) to enhance patient safety and nursing efficiency. Optimizing nursing interventions through biomechanical principles significantly improves postoperative recovery, reduces pain, and minimizes complications. To successfully implement this approach, specific recommendations are provided for clinical practice, nursing management, policy development, and professional education.
Abstract Neck musculature plays a critical role in attenuating head motion during dynamic impacts, yet structure-specific contributions remain poorly quantified. This study used a detailed Global Human Body Model Consortium (GHBMC) finite element head-neck model to evaluate how muscles, ligaments, and vertebrae share internal energy during low-severity head impacts. Seven impact directions were analyzed under three neuromuscular conditions: passive (unanticipated), active fast (power-based), and active slow (endurance-based) at 3 m/s velocity. In passive neck models, deep muscles contributed more during lateral and oblique impacts, whereas superficial muscles dominated in the sagittal plane. Passive and active slow cases shifted internal energy share toward the cervical spine and ligaments, implying greater mechanical demand on these structures. Individual muscle-level analysis identified direction-specific dominant muscles, with sternocleidomastoid consistently the largest single contributor and additional deep and superficial muscles entering the dominant group depending on direction. Active muscle neck models increased muscular energy absorption, particularly in sagittal impacts, accompanied by reductions in linear acceleration of the head. Overall, this study quantifies the roles of deep and superficial neck muscles in stabilizing the head and identifies priority muscles, informing targeted neuromuscular training and personalized protective strategies.
Abstract Pedicle screw fixation systems are essential for spinal stabilization, yet the transient g-forces generated during manual set screw torquing, a critical phase with implications for implant stability, remain poorly understood. This study employs a multimodal experimental approach to quantify these dynamic forces and validate theoretical torque failure models. A sawbone spinal construct was instrumented with accelerometers at biomechanically strategic locations (screw head, spinal center, contralateral pedicle, and surrounding media) to capture transient accelerations during screw fracture. High-speed imaging (40,000 fps) and motion tracking complemented accelerometer data, while distortion energy theory (DET) and fully plastic torque (FPT) models predicted break-off torque. Results revealed extreme g-forces (up to 832 g) localized at the screw head, attenuating rapidly (20-fold reduction at the spinal center). Theoretical predictions (DET: 11.08 N·m; FPT: 11.1 N·m) aligned closely with experimental torque wrench measurements (11.3 N·m, <1.3% error), validating analytical models. Digital image analysis confirmed screw geometry precision (<1.3% error). While the rigid sawbone model limited physiological fidelity, findings emphasize the localized stress propagation and energy dissipation during screw failure, critical for optimizing implant designs, particularly in osteoporotic bone. This integrated methodology bridges biomechanical theory and experimental validation, offering actionable insights to mitigate screw loosening risks and enhance spinal construct durability. Future work will focus on advanced synthetic bone analogs and clinical correlation to refine translational relevance.
Abstract Lung cancer, particularly nonsmall cell lung cancer (NSCLC), remains a leading cause of cancer-related deaths. NSCLC's complexity, due to its heterogeneous nature and resistance to conventional therapies, is influenced by the tumor microenvironment and mechanical properties of cancer cells. Recent studies highlight the importance of these mechanical signaling pathways in tumor progression and treatment response. Qiyu Sanlong Decoction, a traditional Chinese medicine (TCM), has shown potential as an anticancer agent, though its mechanisms, especially regarding mechanical signaling pathways, are not well understood. This study investigates how Qiyu Sanlong Decoction modulates these pathways in NSCLC cells. Through clinical data and in vitro experiments, we examine the decoction's effects on cell proliferation, migration, and the mechanical properties of cancer cells. Our findings suggest that Qiyu Sanlong Decoction significantly alters mechanical signaling pathways, reducing tumor growth and enhancing apoptosis in NSCLC cells. It also affects key molecules involved in cellular mechanics, offering a novel approach to cancer treatment. This research not only advances the integration of traditional medicine with modern oncology but also presents a promising strategy for targeting the mechanical aspects of tumor biology in cancer therapy. In conclusion, Qiyu Sanlong Decoction demonstrates potential as a therapeutic agent for NSCLC by regulating mechanical signaling pathways, warranting further research to explore its molecular mechanisms and clinical application.
Abstract To address the difficulty in dynamically coupling student competencies with medical job requirements in clinical medical education, this paper applies a career planning model based on a dynamic coupling mechanism. First, the analytic hierarchy process (AHP)-entropy weight method is used to quantify student clinical competencies. This method, combined with the LSTM (long short-term memory) algorithm, predicts future departmental talent needs. A weighted Euclidean distance and cosine similarity fusion algorithm is designed to dynamically match competencies with positions. Collaborative filtering and knowledge graph techniques are further incorporated to generate personalized career path recommendations. Competency assessment and recommended paths are dynamically updated through an online learning mechanism. Finally, SHAP (Shapley additive explanations) interpretability analysis is integrated to visualize the contribution of each competency dimension to the recommended results. Experimental results demonstrate that the proposed model achieves high competency assessment accuracy (average 0.855) and job prediction accuracy (average 7.47%). The overall adoption intention for the top-1 recommended path is as high as 74.2%, effectively improving the scientificity, precision, and practicality of medical students' career planning.
Abstract Traditional fixed strategies struggle to balance efficacy and safety due to the nonlinear coupling of control variables like pump and flow rates, and poor personalized control from patient variability. This paper constructs a digital twin-reinforcement learning (RL) soft actor-critic (SAC) framework for continuous extracorporeal blood circulation devices. The twin layer employs a physical-data hybrid model with a residual network for online error correction and Ensemble Kalman Filtering for real-time parameter assimilation. The control layer uses a constrained SAC algorithm, integrating a Lagrange cost, action change rate constraints, and a safety projection operator. Training involves offline pretraining followed by online refinement on the digital twin, with prioritized experience replay and domain randomization. Systematic validation includes simulation, benchtop, and real-world testing. Results show the framework achieves average steady-state errors of 0.37%, 0.45%, and 0.73% in low/medium/high-viscosity patient groups. The comprehensive assessment reports oxygenation efficiency of 95 ± 2 mL O2/min, response time of 1.2 ± 0.1 s, and average severity of 1.8, improving personalized regulation accuracy, real-time response, and operational safety.
Abstract Narrowing of the internal carotid artery (ICA), known as stenosis, can significantly alter normal blood flow and increase the likelihood of cerebrovascular complications such as ischemic stroke. This study presents an in vitro investigation of flow dynamics in a stenosed ICA. A blood-mimicking fluid (BMF) was circulated through the model, allowing for flow behavior similar to physiological conditions. Flow visualization was achieved using particle streak velocimetry (PSV), a noninvasive optical measurement technique capable of capturing the detailed velocity fields. The axial velocity revealed an interesting phenomenon: the velocity profile became smoother and tended toward a more parabolic shape compared to the nonstenotic case, where the peak typically shifts toward the inner curvature region. However, this asymmetry became less pronounced within the stenotic section. The presence of stenosis introduces an opposing curvature, effectively neutralizing the asymmetric flow distribution. Furthermore, the transverse velocity component (v) was plotted along the axial direction (x) to assess secondary flow behavior. The resulting v versus x plot exhibited clear indications of flow disturbances and recirculation zones downstream of the constriction. These findings demonstrate how stenosis modifies the flow patterns within the ICA, potentially contributing to adverse clinical outcomes. The experimental approach using a blood-mimicking fluid and PSV enables a detailed understanding of hemodynamic behavior in stenosed arteries. Such insights are valuable for advancing diagnostic methods and improving treatment planning in vascular diseases.
Abstract Traditional massage teaching lacks quantifiable data and real-time feedback, making it hard for trainees to master proper technique. Existing virtual systems focus on visual simulation but neglect mechanical and physiological feedback. This paper integrates multimodal biofeedback systems, including electromyography (sEMG), pressure, and posture, with a virtual interactive platform to create a “visible-touchable-evaluable” closed-loop teaching system. The system uses hand training gloves equipped with force sensing resistor (FSR), inertial measurement unit (IMU), and sEMG sensors to collect real-time data on force and muscle group activation. Long short-term memory (LSTM) identifies temporal movement patterns and detects deviations. A teacher-side interface provides visual feedback on force, trajectory, and rhythm. Experimental results show that the average trajectory deviation reduces from 18.31% to 15.22%, with force uniformity improving and muscle activation increasing from 0.636 to 0.844. This method significantly enhances massage teaching quality and trainee skill mastery.
Abstract Magnetic fluid hyperthermia (MFH) is an emerging, minimally invasive cancer therapy that induces localized tumor heating using magnetic nanoparticles (MNPs) exposed to an alternating magnetic field. The present study explores the performance of two magnetic-coil arrangements, the Helmholtz coil and a solenoidal coil, through numerical analysis to evaluate their effectiveness in inducing therapeutic hyperthermia in a liver tumor infused with a nanofluid. The computational model incorporates intratumoral injection and diffusion of magnetic nanoparticle-based nanofluid, where the transport of diluted species governs nanoparticle distribution. Temperature increase due to nanoparticle power dissipation is modeled using the Pennes bioheat equation. Results show that the Helmholtz coil offers better magnetic field uniformity, leading to more consistent tumor heating even at off-center positions. In contrast, the solenoidal coil exhibits reduced effectiveness near the tumor edges due to field nonuniformity. The study also underscores the limitation of single-point nanoparticle injection, where limited diffusion results in poor edge heating. These findings suggest that both coil design and nanoparticle delivery strategy play critical roles in enhancing MFH efficacy.
Abstract Kinetic impact projectiles (KIPs) are widely utilized in law enforcement as a nonlethal means for crowd control, yet they remain capable of causing severe or fatal injuries. This study utilizes the total human model for safety (THUMS) finite element model developed by Toyota Motor Corporation and Toyota Central R&D Labs, Inc., to evaluate thoracic injury risk across various impact locations and incident angles for two types of KIPs: the flash-ball and the 40 mm sponge round projectiles. The viscous criterion (VCmax) was used to assess the injury risk considering a threshold of 0.8 m/s, which represents a 50% probability of sustaining a thoracic injury of abbreviated injury scale (AIS) 2 or 3. Simulations demonstrated that impact location and incident angle influence injury severity, with near-perpendicular impacts yielding the highest VCmax values. Results indicated that due to its more concentrated frontal profile, the sponge round projectile transfers approximately 77% of its initial kinetic energy to the body, whereas the flash-ball projectile distributes force more widely, transferring 60% of its initial kinetic energy. Critically, both projectiles approached or exceeded the 0.8 m/s VCmax injury threshold at the manufacturer's recommended minimum firing distances, especially when impacting regions directly over the heart and lungs. In particular, the sponge round projectile presented up to an 80% probability of AIS 2–3 injury under standard operational conditions. These findings suggest that current KIP designs and thoracic targeting protocols may pose a substantial risk of severe injury, highlighting the urgent need for safer projectile designs and improved training to minimize unintended fatalities during deployment.
Abstract Traditional epileptic seizure detection methods suffer from poor interpretability, excessive parameters, and fixed parameters in time-domain dimensionality expansion. To address these, this study proposes a convolutional self-attention adaptive dimensionality expansion network (CSADI-Net), integrating convolutional self-attention and adaptive dimensionality expansion. Convolutional self-attention uses convolutional layers to generate Q (query, representing task-related attention cues), K (key, representing inherent signal characteristics), and V (value, representing input data), reducing trainable parameters (TP). Adaptive dimensionality expansion combines with network training for parameter adjustment. Class activation heatmaps enable visual interpretability. Validated on children's hospital Boston and the Massachusetts institute of technology (CHB-MIT) (Accuracy:98.87%, F1:98.49%) and temple university hospital (TUH) (Accuracy:98.26%, F1:98.13%) datasets, it outperforms CNN, CNN-LSTM, and linear self-attention Transformer. With high accuracy, antinoise ability, and interpretability, it provides a new perspective for seizure detection.
Gestational diabetes mellitus (GDM) involves maternal hyperglycemia, dyslipidemia, and inflammation, impairing placental function through overexpression of nutrient transporters and altered vascular responses. These changes contribute to fetal overgrowth, hyperinsulinemia, and long-term metabolic risks. Therapies like insulin and glucagon-like peptide-1 (GLP-1) receptor agonists (e.g., liraglutide) aim to restore metabolic and vascular balance in GDM pregnancies. To compare insulin and liraglutide in GDM management, focusing on placental molecular transport regulation, endothelial shear stress adaptation, and efficacy in restoring placental function and fetal nutrient balance. A comparative literature review was conducted using peer-reviewed studies on insulin and liraglutide therapy in GDM. Key molecular pathways were analyzed, including glucose transporter type 1 (GLUT1)-mediated glucose transport, mechanistic (mammalian) target of rapamycin (mTOR)-regulated lipid and amino acid transport, and amp-activated protein kinase (AMPK)/endothelial nitric oxide synthase (eNOS)-related endothelial function. Human clinical data and experimental model findings were assessed to evaluate changes in placental transporter expression, maternal lipid profiles, inflammatory markers, and uteroplacental blood flow. Both therapies reduce maternal hyperglycemia and downregulate GLUT1, limiting fetal glucose excess. Insulin improves placental lipid handling by reducing maternal lipotoxicity. Liraglutide, though not approved for GDM, enhances insulin sensitivity, lowers triglycerides and free fatty acids, and reduces inflammation in experimental settings. It activates AMPK/eNOS signaling, improving endothelial function and placental perfusion, and alleviates endoplasmic reticulum (ER) stress, helping restore transporter regulation under hyperglycemia. Insulin remains the standard GDM treatment by correcting insulin deficiency. Liraglutide shows promise due to its broader metabolic and vascular benefits, but requires further research to confirm its safety and efficacy in pregnancy.
Remote photoplethysmography (rPPG) enables contactless estimation of physiological signals from skin videos, offering a promising solution for unobtrusive cardiovascular monitoring. While most rPPG studies have focused on facial regions, privacy concerns limit their broader applicability. In this study, we explore the neck area as an alternative region of interest (ROI) for rPPG-based heart rate (HR) estimation, leveraging its proximity to major blood vessels. Video recordings of the neck of 15 adult subjects were recorded and processed using six rPPG methods: GREEN, CHROM, POS, OMIT, ICA, and LGI. HR estimates derived from each method were compared against those calculated from a gold-standard fingertip PPG using Bland–Altman and correlation analysis. Among all methods, GREEN method demonstrated the best agreement, with a bias of −0.82 bpm and limits of agreement (LoA) of −8.86–7.22 bpm, followed closely by POS (bias = −0.10 bpm; LoA = −8.50–8.30 bpm). Both methods also exhibited strong linear correlation with reference HR (r = 0.95), indicating excellent consistency. Spatial analysis of the neck region identified localized artifacts due to swallowing, light illumination, and vascular asymmetry as potential sources of signal degradation in certain cases. These findings demonstrate the viability of neck-based rPPG for HR monitoring while highlighting the importance of method selection and region-specific optimization. The study provides valuable insights for developing robust, privacy-conscious rPPG systems in clinical and remote healthcare applications.
To address misjudgment of endothelial toxicity caused by chemical heterogeneity of impurities in traditional Chinese medicine injections, this paper established a systematic research framework for chemical and biological effects. A three-dimensional fingerprint was constructed by combining microscopy, Raman spectroscopy, and dynamic light scattering; an energy dispersive spectroscopy (EDS)-based source apportionment model using energy dispersive spectroscopy and pyrolysis-gas chromatography/mass spectrometry (Py-GC/MS) accurately distinguishes medicinal material residues, excipient leaching, and packaging-introduced particles. A particle–cell interaction kinetics simulation algorithm was designed to quantify the bloodstream deposition behavior of particles with varying zeta potentials. Transcriptome and phosphoprotein micro-array data were integrated to identify core mitochondrial stress pathways. Finally, a chemical–biological scoring system was constructed for interpretable toxicity prediction. Experimental results showed polysaccharide/protein aggregates are the key particle type inducing damage to human umbilical vein endothelial cells, increasing reactive oxygen species (ROS) production, disrupting the transendothelial electrical resistance (TEER) barrier, promoting inflammatory factor release, and inducing late apoptosis. Model validation demonstrated a Kappa coefficient of 0.78 with high predictive accuracy (73.3%). This study reveals how the chemical nature of microparticles dominates endothelial toxicity and establishes an integrated “multidimensional characterization-biological response-risk prediction” paradigm. The proposed “characterization-mechanism” strategy can be extended to safety evaluation of microparticles in other complex injections, providing a universal technical path for improving quality control of high-risk injections.
Approximately one-third of individuals with spinal cord injury (SCI) experience a pressure injury and complications contribute to a leading cause of death. Currently, there is a lack of adoption and evidence demonstrating the effectiveness of technologies aimed at preventing pressure injuries. This study presents and evaluates a new monitoring and feedback system to facilitate pressure relief activity performance in 22 manual wheelchair users with SCI. After a two-week baseline, participants received pressure injury education, set goals, and were instructed on how to use the system. Data were then collected using the system for 8 months, and a Technology Acceptance Model (TAM) questionnaire was used to assess usability, with 10 participants identified as Responders. Immediately following goal setting and system biofeedback activation, Responders showed a four-fold increase in pressure relief activity (4.48±4.24 vs. 23.37±14.85 s/h, p=0.004) but activity eventually declined to Baseline levels by Month 8 (4.26±5.78 s/h). Recorded sitting time also dropped significantly (5.62±3.30 h/day at Baseline vs. 1.81±2.08 h/day at Months 6-8, p=0.013). Thirty-six out of 40 questionnaire items scored above the acceptability threshold (>5/7). The system demonstrated good acceptability and effectively promoted short-term increases in pressure relief activity. Its greatest value may be as a short-term training tool during rehabilitation or re-training. Future iterations of the system must address limitations associated with promoting long-term adherence.
Experimental characterization of brain white matter (BWM) using Magnetic Resonance Elastography (MRE), Diffusion Tensor Imaging (DTI), and numerical modeling is expensive, time-consuming, and constrained by computational limitations and model approximations. To address the scarcity of high-fidelity data, this study develops a machine learning (ML) workflow to predict single-frequency viscoelastic properties, specifically the homogenized storage modulus, of BWM. The dataset originates from a sensitivity study conducted in house where BWM was modeled as a 2D triphasic composite of axons, myelin, and glial matrix. The triphasic unidirectional composite only considers 2D mechanics and diffusion in the transverse plane (perpendicular to axonal direction). Microstructural properties such as fiber volume fraction, intrinsic phase moduli, and axonal geometry were used as features for the ML model. Ensemble of regression and decision tree-based models, coupled with hyperparameter optimization, were explored, with model interpretation performed using SHAP analysis. Decision trees yielded the best predictive performance, with SHAP highlighting the importance of glial moduli and fiber volume fraction. This ML framework offers a surrogate to expensive in vivo characterization, provides insight into BWM mechanical dependencies, and can serve as a foundation for future inverse models aimed at understanding aging, dementia, and traumatic brain injury mechanisms in neuroimaging studies
Robotics and machine learning algorithms can potentially enhance upper limb rehabilitation, addressing the limitations of traditional therapy methods. This study presents a novel Human–Robot Interaction (HRI) platform with human brain activities assessment capability aimed at enhancing upper limb rehabilitation by addressing the limitations of conventional therapy. Utilizing a 7DOF Franka Emika robotic arm, the system supports patients in performing lifting, grasping, and reaching tasks structured based on Wolf Motor Function Test (WMFT). Functional near-infrared spectroscopy (fNIRS) concurrently monitors cortical activation and functional connectivity to evaluate neural engagement and recovery. Visual feedback guides participants, while forearm electromyography (EMG) and brain activity from the moving limb are recorded to train deep learning models that classify physiological movement and cognitive load in real-time. Quantitative performance metrics, including average trajectory deviation and nondimensional squared jerk, assess movement accuracy and smoothness, correlating with task complexity. The platform also incorporates a robot impedance control scheme and an interactive interface to adapt assistance dynamically based on predicted movement. By integrating biomechanical performance data with neural indicators, this approach enables a personalized, data-driven rehabilitation framework.
