Cortical folds encode the architecture of human cognition, yet the mechanisms that transform the smooth fetal cortex into its convoluted geometry remain elusive. Biophysical modeling enables mechanistic insight into cortical morphogenesis, but existing models often lack anatomical realism and fail to capture key hallmarks and morphometrics of dynamic cortical folding in the developing human brain. Here, we introduce a whole-brain developmental framework that integrates region-specific, data-driven growth laws with anatomically realistic cortical geometry to enable biologically interpretable modeling of cortical morphogenesis during gestation. Growth fields derived from large-scale prenatal magnetic resonance imaging data capture spatiotemporal variations in cortical expansion and thickness across parcellated regions. Incorporating heterogeneous growth yields folding patterns that match key anatomical landmarks and quantitative morphometrics from human imaging. Systematic perturbations of geometry and growth attributes delineate control parameters that produce realistic morphological variability and replicate clinically atypical brain phenotypes consistent with lissencephaly, pachygyria, and polymicrogyria. This framework provides a quantitative foundation for elucidating the mechanisms of typical and atypical fetal brain development.
Large Language Models (LLMs) offer new avenues to simulate online communities and social media. Potential applications range from testing the design of content recommendation algorithms to estimating the effects of content policies and interventions. However, the validity of using LLMs to simulate conversations between various users remains largely untested. We evaluated whether LLMs can convincingly mimic human group conversations on social media. We collected authentic human conversations from Reddit and generated artificial conversations on the same topic with two LLMs: Llama 3 70B and GPT-4o. When presented side-by-side to study participants, LLM-generated conversations were mistaken for human-created content 39% of the time. In particular, when evaluating conversations generated by Llama 3, participants correctly identified them as AI-generated only 56% of the time, barely better than random chance. Our study demonstrates that LLMs can generate social media conversations sufficiently realistic to deceive humans when reading them, highlighting both a promising potential for social simulation and a warning message about the potential misuse of LLMs to generate new inauthentic social media content.
Advances in cell design have improved lithium-metal battery (LMB) cycle life, but few studies assess performance under discharge profiles representative of real-world use. These profiles, which are often overlooked due to the complexity and risk of misinterpretation, can hinder accurate analysis or even prevent publication. Realistic discharge profiles include high currents during acceleration, current reversal during regenerative braking, and low steady currents at cruising speed. This work examines LMB performance using localized high-concentration electrolytes (LHCEs) under dynamic cycling, focusing on acceleration and regeneration pulses. These profiles bridge practical usage and controlled conditions for reproducible trends. Single-layer pouch cells are tested with LHCEs of lithium bis(fluorosulfonyl)imide (LiFSI), 1,2-dimethoxyethane (DME), with either 1,1,2,2-tetrafluoroethyl 2,2,3,3-tetrafluoropropyl ether (TTE) or bis(2,2,2-trifluoroethyl) ether (BTFE). The inclusion of pulsing dramatically alters the failure of the cells and increases cell-to-cell variability. Increasing the ionic conductivity and electrolyte volume decreases cell-to-cell performance variability. Cells with LHCE-BTFE exhibit more consistent cycling capacity behavior and fewer performance metric fluctuations, such as a rise of polarization or peak cell pressure, under non-uniform cycling compared to LHCE-TTE. These findings suggest that rapid transition from benchtop testing to real-world deployment for LMBs will require the inclusion of more realistic cycling conditions.
Viaduct structures, which are girders supported on piers, are essential for high speed railway (HSR) lines to ensure uninterrupted flow in densely populated regions. Owing to dynamic amplification of deformations and forces, which can be severe under resonant conditions, and the need to limit deck acceleration for structural safety, passenger comfort and rolling stock performance, dynamic analysis is imperative in the design process. Current design standards provide little guidance regarding dynamic analysis approaches, and the research literature indicates that the options are Euler-Bernoulli beam modal analysis, Kirchhoff-Love plate theory-based Generalised Beam Theory (GBT) modal solutions, and finite element (FE) based time domain approaches. This paper investigates the applicability of the simple beam modal solutions on four representative viaduct sections under three typical axle load sets, operating between speeds of 200 to 350 km per hour. First, eigenanalysis solutions using GBT and a solid FE model are compared, indicating excellent agreement under an idealised simply supported boundary condition with torsional restraint, but significant deviations when the FE boundary is modified to represent realistic bearing-type restraints. These observations propagate to dynamic responses under high speed train loading, which generally show good agreement for the analytical boundary condition and differences under realistic boundary condition modelling. Although analytical predictions are typically conservative, the FE predictions are at times significantly higher due to a shift in resonant speed. The results indicate that simplified analytical solutions can be utilised with caution and should be conducted at all applicable speeds to capture the maximum plausible resonant response. The paper concludes with a discussion on its limitations and future research needs.
The illegal trade of rhinoceros horns continues to threaten wildlife populations and undermine conservation efforts worldwide. This study explores the feasibility of using existing radiation portal monitors to detect rhinoceros horns implanted with radioactive sources during international sea transportation. By leveraging the Gamma Detector Response and Analysis Software, a series of simulated pass-bys were conducted under controlled variables, including fill capacities, material densities, and 60Co source strengths. The models incorporated a 6.10-m (20-ft) shipping container with a keratin horn configuration, simulating realistic concealment scenarios. Background environments were modeled after Denver, CO, to reflect elevated natural radiation conditions. Detection performance was assessed using theoretical probability of detection, incorporating International Atomic Energy Agency Nuclear Security Series No. 1 standards for operational consistency. Results show that increased cargo capacity and denser shielding materials, such as aluminum and stainless steel, reduce probability of detection, though such solid-block shielding is unlikely in real-world cargo. Shadow shielding from empty or lightly filled containers enhanced detection capability by reducing background noise. Furthermore, modeling scenarios with multiple horns improved probability of detection, aligning with seizure patterns observed in sea transport trafficking cases. While the study does not determine the minimum detectable activity due to law enforcement sensitivities, it demonstrates the plausibility of radiation portal monitor systems identifying radiologically tagged horns under realistic shipping conditions. The findings support the concept of radiation tagging as a cost-effective, scalable strategy to deter trafficking, increase seizure rates, and reinforce international conservation efforts without disrupting trade flow or requiring infrastructure overhauls.
Untrustworthy randomised controlled trials (RCTs) and other study types are an increasingly recognised problem in health and medical research. Assessing and responding to this problem is essential for ensuring the reliability of scientific evidence informing clinical practice. We consider theory, methods and tools for assessing research integrity and the trustworthiness of both aggregate and individual-level data from peer-reviewed, published RCTs. We review approaches to the assessment of trustworthiness based on published checklists, the statistical analysis of aggregate and individual participant data, and the use of AI tools. We found checklists that examine the trustworthiness of data considering questions about the timeframe, authors, governance, and plausibility of a published, peer-reviewed paper. These checks of authenticity are complemented by statistical techniques that identify unusual patterns in aggregate or individual data, including new procedures for simulating data to determine if any of the possible distributions are realistic. The unique character of RCTs provides additional opportunities for checks, specifically for the baseline data. The approaches presented offer a series of novel, quantitative tools for assessing research integrity and data trustworthiness across published RCTs. These techniques can detect instances of data duplication, questionable research practices, and check if any aspects of the study or results are unrealistic. They work best when employed beyond isolated, single-trial analyses and, when embedded in AI platforms, should scale to meet the volume of corrupted papers already published and the stream of fake research from paper mills.
Artificial intelligence (AI) has emerged as a tool to augment plastic and reconstructive surgery (PRS) education. AI-generated videos (AIVs) are entirely or partially created using machine learning to generate frames, scenes, or sequences depicting either fictional or realistic footage. Deepfakes are AI-generated audiovisual content in which a real person's identity or likeness, such as their face, voice, or expressions are altered or synthesized to convincingly make it appear as though they said or did something they did not.While legitimate concerns exist regarding their authenticity, AIVs and deepfakes can disseminate information using a realistic human avatar with minimal time constraints. In this study, we demonstrate the ability of AI-simulated videos to deliver educational content using a board-certified PRS surgeon who trained the AI as the avatar. By leveraging AIV technology, this study highlights a new approach to patient education with implications for virtual consultation and content creation in PRS. This method also addresses the evolving role of information sharing and marketing in PRS while focusing on maintaining professional standards and ethical integrity.
Parkinson's disease is a rapidly growing neurodegenerative disorder with various motor and non-motor symptoms, affecting millions of people worldwide. These symptoms demonstrate significant medication-related fluctuations and inter-patient variability, highlighting the need for personalized management. Objective longitudinal symptom monitoring through wearable sensors and machine learning can support individualized care. However, to date, most approaches have been tested in lab-constrained environments. This study aims to develop a modular pipeline to automatically detect three cardinal Parkinson's disease motor symptoms, tremor, bradykinesia, and levodopa-induced dyskinesia in more realistic scenarios. The proposed approach was evaluated on three datasets: the Levodopa Response Study and two newly introduced ALAMEDA datasets, containing tri-axial wrist accelerometer data collected with commercial wearable devices during clinical assessments and activities of daily living. For each symptom, separate context-agnostic models were developed using 92 hand-crafted features. Multiple segmentation window lengths and preprocessing techniques, including resampling and dimensionality reduction, alongside various machine learning models, including logistic regression, k-nearest neighbor, multilayer perceptron, support vector machine, decision tree, AdaBoost, and random forest, were explored. Statistical significance between configurations was assessed with the Wilcoxon signed-rank test. Model interpretability was investigated using Shapley additive explanations to identify highly influential predictors and assess their physiological relevance. In the Levodopa Response Study dataset, tremor, bradykinesia, and dyskinesia detection reached 0.664, 0.636, and 0.443 area under the precision-recall curve, respectively, demonstrating scalability in high-complexity settings and revealing physiologically meaningful patterns. When evaluated on the ALAMEDA datasets, tremor and dyskinesia detection achieved 0.879 and 0.648 area under the precision-recall curve, highlighting strong model and feature generalizability. Across symptoms, longer segmentation windows and random forest classifiers performed better, while synthetic oversampling and principal component analysis showed limited impact. Automated Parkinson's disease symptom detection is feasible in more realistic, free-living conditions, with only a slight performance decrease despite substantially increased complexity. With carefully selected features and pipeline components, the objective, unobtrusive monitoring of motor symptoms can support personalized, evidence-based treatment suggestions, eventually improving patients' quality of life. Not applicable.
Emotional experience and the regulation thereof are typically studied using picture inventories or short films to induce and modify affective states. These approaches, however, lack ecological validity due to their passive and receptive nature. Recent innovations in virtual reality and mobile neurophysiological technologies have enabled researchers to study the behavioral and neural correlates of more ecologically valid emotional responses. In this preregistered study, 58 healthy participants were randomly assigned to either use cognitive reappraisal (intervention) or to immerse themselves in their senses and surroundings (control) while walking across a wooden plank suspended 80 stories above the ground in virtual reality. We measured subjective fear ratings, salivary alpha amylase and cortisol levels, as well as frontal brain asymmetries, captured using mobile electroencephalography (EEG). Across both conditions, we found decisive evidence of increased subjective fear and salivary alpha amylase, a marker of sympathetic activation. However, we found no increase in cortisol levels following the task suggesting that subjective fear alone is not sufficient to trigger a cortisol response. In contrast to our hypotheses, the reappraisal group did not show any difference compared to the control group for neither emotional, endocrine nor neural measures. On the one hand, our findings may suggest that reappraisal might not be a suitable strategy to regulate realistic and intensely frightening situations. On the other hand, further analyses also indicated that the control group may have also regulated their emotions due to increased mindfulness of their inner states and their environment. Future studies are needed to confirm these observations and ascertain the efficacy of cognitive reappraisal on fear in realistic settings.
Objective evaluation of quantitative imaging (QI) methods with patient data is often hindered by the lack of gold standards. To address this challenge, a class of regression-without-truth (RWT) techniques have been developed. These techniques assume that the true and measured values are linearly related and estimate the linear-relationship parameters without access to true values. However, reliable estimation of these parameters typically requires many patient samples, which can be expensive and time consuming to obtain, and even impossible in settings such as studies with rare diseases or with new clinical imaging procedures. Thus, there is an important need for strategies to perform evaluation of quantitative imaging methods with a small number of patient samples. In this context, we note that datasets with known ground truth, such as physical phantom studies, could be available. In this manuscript, we propose an approach that integrates information from both patient data without ground truth and known-ground-truth datasets to perform objective evaluation of QI methods. We validated the proposed approach using numerical studies, which showed that the proposed approach yielded improved performance in ranking QI methods compared with RWT technique. The results demonstrate the potential of the proposed approach for evaluating QI methods when patient data are limited and motivate further validation with clinically realistic simulation studies and clinical data.
This study presents classification models trained to diagnose and grade prostate cancer using fresh prostate biopsies. We compare the performance of classification models with optimised sensitivity and specificity (standard models) with application-specific models designed to maximise sensitivity and negative predictive value (NPV). Standard models achieve 80% sensitivity and 81% specificity. Application-specific models, calibrated to 90% sensitivity and 95% NPV, are intended to provide clinicians with a tool they can use with confidence to support intraoperative decisions, specifically to improve tissue retention during biopsy procedures and to ensure clear surgical margins. To this end, we introduce a 5-layer algorithm that combines 5 application-specific models chosen for overall best performance. This algorithm can reduce the number of biopsy samples required for diagnosis by 47% while maintaining 90% sensitivity, 95% NPV, and 62% specificity. All models are independently validated using two large patient cohorts. These results support the targeted use of Raman spectroscopy for real-time tissue analysis in diagnostic and intraoperative settings. The technology's clinical value as a decision-support tool aligns with the shared goal of pathologists and urologists to reduce the number of prostate biopsy cores while maintaining high sensitivity for clinically significant cancer. Prior studies have improved biopsy efficiency, but their performance has been variable, and concerns remain regarding underdetection of significant disease, revealing the need for approaches that improve biopsy efficiency without increasing diagnostic risk. The technology described here provides a realistic solution for targeted biopsy guidance to support more precise and evidence-based clinical decisions.
Modern interconnected power systems with high penetration of renewable energy sources (RES) and large‑scale use of electric vehicles (EVs) experience recurring frequency and voltage disturbances. Accordingly, the main objective in this work is to overcome this limitation in a power system under coordinated load frequency control (LFC) and automatic voltage regulation (AVR), which requires highly accurate control strategies. Fuzzy logic‑based controllers are among the most promising approaches for disturbance rejection, control precision, system stability, and robust performance. However, their performance is often limited because the selection of crisp ranges (i.e., the universes of discourse of the fuzzy variables) is usually set heuristically. In this paper, the crisp output range of a Fuzzy Proportional Integral Derivative Double Derivative (FPIDD2) controller, which is based on a previous study, is reconfigured while maintaining the original rule base and membership function structure unchanged. This reconfiguration is presented to change the controller's response by recentering (shifting) the zero output membership function rightward, thereby improving control sensitivity and dynamic performance. The effectiveness of the proposed approach is validated via MATLAB simulation on a multi‑area interconnected power system under realistic operating conditions, including stochastic input fluctuations due to renewable energy source penetration and electric vehicle participation, as well as nonlinear constraints such as generation ramp‑rate limits and governor dead zones. Several metaheuristic optimizers, including Particle Swarm Optimization (PSO), Gorilla Troops Optimizer (GTO), and Marine Predators Algorithm (MPA), are used to further evaluate the robustness of the proposed method. The results demonstrate that optimized FLC configuration with modified crisp ranges significantly improves controller sensitivity, damping characteristics, and robustness. Consequently, the Integral of Time- Absolute Error (ITAE) is reduced by up to 69% compared to the original controller configuration.
Paratuberculosis, caused by Mycobacterium avium subsp. paratuberculosis (MAP), is a chronic, incurable enteritis of ruminants in which late-onset clinical signs preclude timely intervention, making early genomic stratification a key control lever. We investigated the genetic architecture underlying the host antibody response to MAP - measured as anti-MAP ELISA serostatus - using genome-wide SNP data from 474 animals representing seven indigenous Turkish goat breeds sampled across 11 provinces (237 seropositive, 237 seronegative; estimated true seroprevalence 14.1%). Fourteen architecturally diverse machine learning frameworks - encompassing regularized regression, GBLUP, kernel machines, tree ensembles, deep neural networks, and a meta-ensemble - were evaluated under repeated stratified cross-validation using both discriminative and calibration metrics. Mutual information-based preselection served a dual function: reducing computational dimensionality while prioritizing markers with detectable phenotypic association. Six regularized and kernel-based models converged at a statistically indistinguishable discrimination ceiling (mean AUC ≈ 0.982; range 0.002), a convergence pattern consistent with an additive component captured by the available SNP data. Critically, whereas ROC-AUC varied only 1.14-fold across all models, Brier score varied 4.85-fold (0.046-0.223), with tree ensembles exhibiting systematic miscalibration despite competitive discrimination - indicating that predicted probabilities vary substantially across model architectures and cannot be reduced to discrimination metrics alone when informing breeding or biosecurity decisions. Post-hoc offset recalibration across realistic field seroprevalence conditions preserved tier-level performance rankings across the tested prevalence scenarios. These findings indicate that genome-wide SNP variation encodes sufficient signal to support accurate and calibrated classification of anti-MAP ELISA serostatus in indigenous goat populations, and provide a calibration-aware framework potentially extensible to other complex disease traits in livestock.
This review summarizes current evidence on the biological, diagnostic, prognostic, and therapeutic significance of microRNAs (miRNAs/miRs) in prostate cancer (PCa). Dysregulated miRNA networks contribute to PCa initiation, progression, metastasis, and therapy resistance by regulating androgen receptor signaling, proliferation, apoptosis, epithelial-mesenchymal transition, angiogenesis, hypoxia signaling, bone tropism, and castration-resistant evolution. Oncogenic miRNAs, including miR-21, miR-93, miR-9, miR-181a, and miR-182, promote malignant phenotypes through survival, TGF-β, PI3K/AKT, MAPK, HIF-1α, and EMT-associated pathways, whereas tumor-suppressor miRNAs, including the miR-34 family, miR-145, miR-122, and miR-382, restrict proliferation, stem-like traits, invasion, metastasis, and treatment resistance. Circulating, urinary, and exosomal miRNAs have potential as minimally invasive biomarkers for PCa detection, risk stratification, recurrence monitoring, metastatic risk prediction, and assessment of castration-resistant disease. Multi-miRNA panels, such as miR-375-3p/miR-182-5p, miR-34b-3p/miR-361-5p/miR-200c-3p, and miR-200c/miR-605/miR-135a/miR-433/miR-106a, may outperform individual markers by capturing multiple biological pathways and reducing single-marker variability; however, most remain at the discovery or validation stage rather than routine clinical implementation. Translation requires standardized sample handling, hemolysis control, reproducible RNA extraction, validated normalization strategies, assay harmonization, locked thresholds, and multicentre prospective validation against PSA, imaging, histopathology, and established risk models. High-throughput platforms, including qRT-PCR, microarray, next-generation sequencing, and nCounter digital counting, support miRNA discovery and validation but differ in sensitivity, specificity, cost, throughput, bioinformatic complexity, and clinical deployability. Therapeutically, anti-miRs, miRNA mimics, sponges, masks, CRISPR-based approaches, and nanocarrier-assisted delivery systems provide experimental strategies for inhibiting oncogenic miRNAs or restoring tumor-suppressor miRNAs. Preclinical evidence supports miR-21 inhibition and replacement of miR-34a, miR-145, miR-15a/miR-16-1, miR-124, and miR-205. Still, clinical translation is limited by off-target effects, immune activation, delivery efficiency, endosomal escape, tumor heterogeneity, pharmacokinetics, toxicity, scalability, and manufacturing reproducibility. Overall, miRNAs provide mechanistic insight into PCa heterogeneity and offer promising opportunities for precision diagnosis, prognosis, and therapy, with the most realistic near-term application being integration of validated miRNA panels with PSA, multiparametric MRI, pathology, and multi-omic or AI-assisted risk models.
Plastic pollution resulting from the continued dominance of fossil-derived polymers is a major global environmental challenge. Although biodegradable plastics, such as polylactic acid (PLA), polyhydroxyalkanoates (PHAs), polybutylene succinate (PBS), polybutylene adipate terephthalate (PBAT), and related materials, are increasingly being deployed as alternatives, their environmental performance is frequently constrained by infrastructure gaps, uncontrolled carbon loss, and incomplete degradation under realistic conditions. Therefore, biodegradability alone does not guarantee circularity of the material. To address this, intentional rerouting of plastics and their monomers into upcycling streams offers a widely applicable solution. This review advances the circular bioeconomy framework built on engineered depolymerization and metabolic bio-funneling of biodegradable and selected synthetic plastics. We present recent progress in enzyme-mediated polyester breakdown, emphasizing hydrolases and oxidoreductases, the kinetic and structural determinants of activity, and protein engineering strategies that broaden substrate scope and enhance operational stability. We then organize bio-upcycling strategies according to key metabolic entry nodes: pyruvate, acetyl-CoA/β-oxidation, and aromatic/dicarboxylate pathways, to demonstrate how plastic-derived monomers can be systematically redirected toward platform chemicals, fuels, specialty monomers, and next-generation biopolymers through pathway rewiring, flux control, and redox balance. In addition to biological conversion, we evaluate chemo-biological hybrid systems and integrated techno-economic and life cycle considerations, including process efficiency, enzyme cost, toxicity mitigation, and infrastructure compatibility. We further highlight emerging tools, such as systems biology, adaptive laboratory evolution, synthetic consortia design, and machine-learning-guided protein optimization, which accelerate the design-build-test-learn cycle for scalable microbial platforms for plastic upcycling. Collectively, this study reframes biodegradable plastics not as materials designed merely to disappear but as programmable carbon reservoirs that can be captured, upgraded, and reintegrated into biomanufacturing value chains. Actively closing the loop through engineered bio-upcycling, rather than relying on passive environmental degradation, offers a practical pathway to align plastic utility with environmental sustainability and achieve a truly circular bioeconomy.
This study aims to investigate the potential impact of the Plasmodium falciparum sporozoite (PfSPZ) vaccine and indoor residual spraying (IRS) on malaria transmission in Keerom, Papua. An optimal control model is introduced in this article for malaria transmission using a deterministic compartmental host-vector model. The model considers the potential impact of the new PfSPZ vaccine and IRS through a novel host-vector framework. Treatment failure is also accounted for in the model to provide a more realistic representation of the disease dynamics. Mathematical analysis regarding the existence and stability of equilibria is conducted rigorously, and the basic reproduction number is derived using the next-generation matrix approach. We found that the malaria-free equilibrium is always locally asymptotically stable when the basic reproduction number is less than 1. Conversely, using a center manifold approach, we showed that the malaria-endemic equilibrium is asymptotically stable when the basic reproduction number is greater than 1 but close to 1. Model parameter values are estimated using incidence data from Keerom, Papua, an area in Indonesia with one of the highest malaria incidences at the national scale. A global sensitivity analysis is conducted using Partial Rank Correlation Coefficients, which reveal the importance of vaccination and IRS strategies in reducing the basic reproduction number. Through cost-effectiveness analysis based on the optimal control simulation results, we find that although the IRS-only intervention appears to be the most cost-effective strategy, the combination of PfSPZ vaccination and IRS yields the greatest overall impact on reducing malaria transmission. In particular, the combined strategy produces the largest reduction in the number of infected individuals, while the IRS-only strategy gives the most favorable cost-effectiveness outcome. These results suggest that PfSPZ vaccination and IRS can provide important benefits for malaria control in high-endemic areas such as Keerom, Papua.
Our study demonstrates that a compact, low-cost, phantom-calibrated diffuse reflectance spectroscopy system can provide realistic estimates of tissue optical properties, supporting its progression toward in situ cartilage diagnostics. We developed and validated a fiber-based DRS system capable of quantitative estimation of absorption coefficient (µa) and reduced scattering coefficient (µs') in cartilage using a phantom-derived calibration model at wavelengths of 660, 780, and 850 nm. Diffuse reflectance spectra were acquired using a custom-built multi-distance probe, corrected using wavelength-specific calibration coefficients derived from optical phantoms, and fitted using diffusion theory with extrapolated boundary correction to recover µa and µs'. Our system was validated using phantoms with known optical properties, calibrated to correct system response, and applied to articular cartilage. Theoretical and measured reflectance showed an excellent agreement (bias ∼0, 95% limits of agreement ±0.0114). After calibration, µa and µs' were extracted from bovine patellar cartilage at the same wavelengths. The estimated µa values (0.07-0.24 cm-1) were characteristic of weakly absorbing, hydrated soft tissue, while µs' values (9.7-15.8 cm-1) showed the expected consistent decrease with wavelength. We found our estimated optical properties were consistent with literature trends, especially µs' values, which were between the range reported for µs' values from integrating-sphere and Monte Carlo-based analysis for bovine cartilage. Phantom-calibrated DRS enables accurate, reproducible estimation of cartilage optical properties, providing a validated framework for future translation toward arthroscopic optical assessment of joint health.
Futsal requires high-intensity explosive actions, including vertical jumps. Force-time analysis from dual force plates provides multidimensional biomechanical data that traditional statistical methods struggle to analyze effectively. To develop and evaluate a supervised classification framework capable of reproducing internally-derived vertical jump performance categories in futsal athletes using independent force-time metrics from dual force plates. Fifty-one male athletes performed countermovement jumps (CMJ), squat jumps (SJ), and drop jumps (DJ) on VALD ForceDecks dual force plates (1,000 Hz), yielding 148 valid observations after exclusion of records with missing PCA-input variables. Four internally derived performance categories were constructed using Principal Component Analysis of six biomechanical variables. The first two retained components explained 75.6% of the total variance. We compared four machine learning algorithms using an approximate 75/25 athlete-level stratified group split and stratified 5-fold cross-validation with group constraints to prevent leakage from repeated jump records from the same athlete. Robustness was assessed through reduced-predictor sensitivity analysis, jump-test ablation, learning curves, calibration, and nested cross-validation. Logistic Regression achieved strong performance on the independent test set (F1-Score = 0.830, AUC-ROC = 0.977). Grouped 5-fold cross-validation yielded more conservative estimates (F1 = 0.770 ± 0.069, AUC = 0.941 ± 0.039), reflecting a more realistic estimate of internal generalization to unseen athletes. Coefficient-based and permutation importance analyses consistently identified CMJ peak power as the most influential predictor, while DJ-derived variables, particularly eccentric mean force, concentric impulse, and peak power, also showed substantial discriminative relevance. Ablation analysis confirmed the critical role of drop-jump variables: removing them reduced F1-score from 0.830 to 0.604, corresponding to an absolute decrease of 0.226 and a relative reduction of approximately 27%. Machine learning algorithms, particularly Logistic Regression, supported accurate classification of internally derived PCA-based vertical jump performance categories in futsal athletes. The combination of PCA-based category construction and supervised classification, with rigorous athlete-level validation and comprehensive robustness diagnostics, provides a reproducible methodological framework that may support talent identification, individualized training, and longitudinal monitoring after external validation.
Microbial keratitis requires rapid pathogen identification to guide treatment, but culture- and PCR-based diagnostics are slow and resource-intensive. We developed a triple-phase multimodal framework for bacterial-versus-fungal keratitis classification using slit-lamp photographs acquired under blue-light, sclerotic-scatter, and white-light illumination, together with clinical metadata. The model combines cross-modality contrastive learning, modality-specific fine-tuning, and feature-level multimodal ensemble learning for patient-level prediction. We evaluated the framework on a multicenter dataset of 1,645 patients and 17,158 images from India and the United States. The model achieved 85.84% accuracy, 84.46% average F1-score, and 0.885 AUC. Site-specific evaluation showed that pooled results were overly optimistic, whereas resampling- and balance-based re-evaluation provided a more realistic assessment of cross-site generalization. Under all settings, our framework remained the top-performing approach. The code is available at https://github.com/yqwang01/TPMKA and dataset access will be provided subject to University of Michigan data-sharing clearance.
Somalia faces a substantial mental health treatment gap driven by severe workforce shortages, limited specialist services, and weak health-system infrastructure. Artificial intelligence (AI) is increasingly promoted as a tool to strengthen health systems, yet its application in low-resource and conflict-affected settings requires careful consideration of feasibility, governance, and equity. This commentary examines realistic opportunities for integrating AI into mental health research and service delivery in Somalia. It requires AI to be viewed as a system-support tool rather than a substitute for clinicians, with potential to strengthen task-shared primary care, supervision, referral pathways, and routine data use within the District Health Information Software 2 (DHIS2) platform. Examples of potential applications include WHO Mental Health Gap Action Programme (mhGAP)-aligned decision-support tools, automated data-quality checks, supervision dashboards, and mobile-based appointment reminders with privacy safeguards. Key constraints include uneven digital connectivity, affordability barriers, limited regulatory capacity, and evolving data protection frameworks under Somalia's 2023 Data Protection Act. Ethical considerations such as confidentiality, stigma, algorithmic bias, and digital exclusion are highlighted, alongside the need for human in the-loop oversight and monitoring of performance across vulnerable groups. In low-resource settings such as Somalia, responsible AI integration requires feasibility, equity, and accountability. Successful implementation also requires alignment with existing health-system platforms, strong governance safeguards, measurable outcomes, and sustained investment in workforce capacity. When carefully deployed, AI can support but not replace mental health system strengthening in Somalia.