Physiological jaundice is present in the initial week of life in neonates owing to the rise in level of bilirubin thereby resulting in yellowish coloring of sclera and skin. Severe jaundice and lethal bilirubin levels may result from brain damage as bilirubin is located in the central nervous system. Present diagnostic techniques consist of time-consuming and a painful invasive blood test and non-invasive tests using expensive transcutaneous bilirubin meters. Then regular monitoring is important, numerous efforts are conducted to progress non-invasive devices for testing utilizing a smartphone camera. Various efforts have been deployed to automate the neonatal jaundice diagnosis applying dissimilar machine learning, image processing, and CV methods. With the rise of smartphone-based and computer vision applications in clinical environments, especially for early detection of diseases like neonatal jaundice, reliability becomes more critical. This paper develops an Advancing Early Jaundice Detection of Neonatal with a Smartphone-based Computer Vision System and Golden Jackal Optimizer algorithm (AEJDN-SCVGJO) for Clinical Decision-Making. The image pre-processing applies an adaptive median filter (AMF) to enhance image quality by removing the noise. For the feature extractor, the SE-DenseNet has been deployed. Moreover, the proposed AEJDN-SCVGJO model executes the temporal convolutional network (TCN) model for the classification process. Finally, the Golden Jackal Optimizer (GJO) adjusts the parameter value of the TCN model optimally and outcomes in higher solution of classification. To exhibit the enhanced execution of the presented AEJDN-SCVGJO methodology, a wide-ranging experimental investigation is made. The comparative outcomes reported the improvised characteristics of the AEJDN-SCVGJO model.
BackgroundBallistocardiography (BCG)-based heart rate (HR) monitoring faces accuracy degradation due to motion artifacts, limiting its practical deployment.ObjectiveThis study aims to enhance HR estimation reliability under motion-contaminated conditions while ensuring real-time performance.MethodsA hybrid system integrating adaptive filtering and enhanced continuous wavelet transform (CWT) is developed. The framework localizes motion segments (95.1% accuracy) and employs spectral reconstruction via magnitude-frequency nullification to restore HR from contaminated windows. Computational latency was evaluated on an embedded ARM platform to verify real-time feasibility.ResultsValidation using 6000 min of data demonstrated that the proposed method achieved an MAE of 2.94 BPM, comparable to the CNN-LSTM baseline (2.85 BPM), while reducing the average processing latency from 450.2 ms to 86.4 ms. Compared with conventional methods, the proposed framework reduced the MAE by 53.8% and improved monitoring stability by 35.6%. Bland-Altman analysis confirmed limits of agreement within [-4.77, 5.23] BPM, validating clinical reliability.ConclusionsThe proposed hybrid framework provides a computationally efficient and accurate solution for non-contact HR monitoring in motion-prone and bed-based clinical environments.
BackgroundAugmented reality (AR)-based exercise systems can automatically quantify task performance; however, evidence linking AR-derived metrics with conventional clinical measures remains limited.ObjectiveTo explore associations between AR-derived functional task performance and conventional clinical measures in healthy young adults.MethodsThirty healthy young adults completed four AR-based tasks: shuttle run (SR), on-the-spot running (OSR), side jump (SJ), and standing long jump (SLJ). Performance was captured using a LiDAR-based sensing system. Clinical assessments included sit-to-stand tests, pulmonary function, respiratory muscle strength, hand grip strength, body composition, and isometric quadriceps strength. Correlation and regression analyses examined associations between AR-derived outcomes and clinical measures. Additional multivariable analyses for SLJ adjusted for sex and height.ResultsSLJ showed the broadest pattern of associations with clinical measures. After adjustment, SLJ remained significantly associated with 1-min sit-to-stand performance, percent-predicted forced vital capacity, hand grip strength, skeletal muscle mass index, skeletal muscle mass, lower limb muscle mass, and Biodex peak torque. Associations with absolute pulmonary function measures, respiratory muscle strength, and FEV1/FVC were attenuated after adjustment. SR and OSR showed limited associations with selected measures, whereas SJ showed no significant associations.ConclusionIn healthy young adults, AR-derived SLJ performance showed the most consistent cross-sectional associations, primarily with muscle-related measures. Its association with percent-predicted forced vital capacity also remained significant after adjustment. These findings are hypothesis-generating and should not be interpreted as evidence of technical or clinical validation of the AR-based platform.Trial Registration: This study was approved by the Institutional Review Board of Pusan National University Hospital (IRB No. 2505-024-151) and registered with the Clinical Research Information Service (CRIS; KCT0010729).
ObjectiveSustained postural loading is a common contributor to neck-shoulder pain, yet its effects on muscle microvascular function and cervical segmental motion are not fully understood. This study investigated changes in muscle microcirculation and cervical intervertebral mobility in individuals with postural neck-shoulder pain and examined the effects of therapeutic stretching.MethodsThirty asymptomatic participants and thirty with neck-shoulder pain were assessed using laser-Doppler flowmetry and videofluoroscopy to measure blood flow in the sternocleidomastoid (SCM), upper trapezius (UT), and masseter (MA) muscles and spinal motion, respectively. The patient group participated in an 8-week supervised stretching intervention.ResultsCompared to asymptomatic controls, patients with neck-shoulder pain showed 7.3%-11.2% lower blood volume and 5.8%-13.6% lower blood speed in target muscles. Post-intervention, significant improvements were observed in blood volume (SCM: 71.26 ± 9.00, UT: 75.80 ± 27.22, MA: 83.78 ± 19.26 au) and speed (SCM: 16.09 ± 4.75, UT: 17.65 ± 6.83, MA: 24.55 ± 10.58 au; all p < 0.05). Intervertebral flexion/extension mobility from C2/3 to C6/7 also significantly improved.ConclusionsAlterations in muscle microcirculation may be associated with impaired cervical biomechanics in postural neck-shoulder pain. Combined assessment of local blood perfusion and intervertebral mobility may provide valuable insights into the underlying mechanisms and support the development of targeted rehabilitation strategies to patients experiencing postural neck-shoulder pain.
Background: EEG responses to violence-related visual stimuli are relevant to neuroscience and digital forensics. Yet most EEG classification models emphasize predictive performance and provide limited evidence about the signal patterns that support each decision.Method: In this research, Tensor Pattern (TensorPat) was proposed as a three-dimensional feature extractor for EEG-based violence-stimulus classification. TensorPat was combined with CWINCA for feature selection, tkNN for classification, and Directed Lobish (DLob) for explainable result generation. In this way, a lightweight and traceable explainable feature engineering framework was developed.Results: A new 32-channel EEG dataset was collected from 34 participants, yielding 527 violence and 1285 control segments. TensorPat generated 10,240 features per segment, of which 131 were retained by CWINCA. LOSO CV was the primary validation protocol. The model achieved 96.14% accuracy and 94.37% balanced accuracy. Violence and control sensitivities were 90.13% and 98.60%, respectively. The DLob sentence had a complexity ratio of 90.92% and identified recurrent frontal, occipital, and parietal channel-symbol patterns. These outputs are model-derived explanations rather than direct activation maps.Conclusions: TensorPat provides a lightweight and explainable framework for EEG-based violence-stimulus classification. DLob outputs should be interpreted as model-derived explanations, not as direct brain activation maps or clinically validated biomarkers. The between-subject acquisition design defines the present scope of inference; matched within-subject cohorts and external datasets are required for broader validation.
ObjectiveTo develop and evaluate a single-chamber dual-channel optical probe leveraging near-infrared spectroscopy (NIRS) for the objective assessment of pressure-modulated tissue response, with emphasis on subcutaneous microvascular-related optical changes.MethodsA single-chamber and dual-cavity optical probe was designed using a quantitative approach inspired by the blanching test principle. The probe simulated two conditions: pressure and non-pressure, using 740 nm and 850 nm near-infrared LEDs as light sources. The optical properties of skin tissues were evaluated in these conditions to analyze localized blood flow changes. The study recruited healthy participants and individuals with pressure injuries, targeting high-risk anatomical sites to compare reflectance intensity differences across the two wavelengths.ResultsData from 398 measurement points demonstrated significant differences in NIRS reflectance between healthy and pressure injury tissues. In healthy tissues, reflectance at 740 nm and 850 nm increased under pressure due to effective blood displacement, indicating normal vascular function. Conversely, pressure injury sites exhibited reduced reflectance changes and higher hemoglobin absorption, particularly at 850 nm, reflecting impaired blood perfusion. Variance analysis confirmed significantly lower reflectance variation in pressure injury tissues compared to healthy tissues (p < 0.001), highlighting restricted blood flow dynamics.ConclusionThis pilot study demonstrates the feasibility of NIRS-based optical probing for assessing pressure-modulated tissue response at high-risk anatomical sites. These preliminary findings support further validation in larger and more rigorously designed clinical studies.
BackgroundAging causes declines in cognitive and motor functions, often manifested in altered gait. Dual-task gait is a sensitive marker for early functional changes. This study investigated whether neural network (NN) models can differentiate gait patterns across age groups under dual-task conditions, and which joint features contribute most to age classification.MethodsThirty-six healthy were recruited for three groups: young (18-37 years), middle-aged (38-57), and older adults (≥58), with 12 in each. Gait data were collected under dual-task conditions (reverse-counting while walking) using a Vicon motion capture system and force plates. The outcomes from gait analysis included spatiotemporal parameters, full-body joint angles, and lower limb joint force, moment and power. Eleven kinds of NN models with different combinations of gait outcomes were constructed using NNs to classify age groups.ResultsThe NN model combining all kinematics, kinetics, and spatiotemporal parameters achieved the highest classification accuracy (95.70 ± 3.15%), with F1 scores (the harmonic mean of precision and recall) above 0.948 across all age groups. Among single-feature models, the full-body kinematic model performed best (94.17 ± 3.47%), while the spatiotemporal model showed the lowest accuracy. The middle-aged group achieved the highest accuracy in the full-body kinematic model. Ankle-related features were a main factor in contributing to the NN models.ConclusionNN models with all-parameters can accurately classify age-related dual-task gait differences. By capturing gait changes, NN offers a quantitative tool for early detection of motor and cognitive decline, with potential applications in fall risk prediction, dementia screening, and targeted rehabilitation.
ObjectiveInfusion and syringe pumps are essential for safe and continuous medication administration, but quantitative indicators for their allocation across hospitals remain unclear. This study examined whether pump ownership in hospitals employing clinical engineers was associated not only with hospital size but also with hospital functional and structural characteristics.MethodsThis nationwide cross-sectional study integrated questionnaire data on infusion and syringe pump ownership with publicly available hospital-level data in Japan. A total of 248 hospitals were analyzed. Associations with hospital characteristics were examined using Spearman's rank correlation, non-parametric group comparisons, and multiple linear regression analyses.ResultsPump ownership was strongly correlated with the number of beds for both infusion and syringe pumps. Pumps per bed ranged from 0.26 to 0.48 for infusion pumps and from 0.16 to 0.45 for syringe pumps. In multivariable analyses, infusion pump ownership was associated with bed count, chemotherapy cases, emergency transports, and advanced treatment hospital status. Syringe pump ownership was associated with bed count, emergency transports, surgeries under general anesthesia, secondary and tertiary emergency care status, and advanced treatment hospital status.ConclusionPump ownership was associated with hospital size and functional characteristics. These findings support function-based and data-driven medical device allocation indicators.
Real-time analysis of biomedical time-series on edge devices faces significant challenges due to signal non-stationarity, environmental noise, and the strict computational limitations of embedded hardware. While deep learning offers superior accuracy, standard architectures are often too computationally demanding for embedded deployment. In this work, we propose TinyBioNet, a highly compact convolutional neural network designed explicitly for the efficient affective state classification on devices with limited resources. Unlike traditional 1D CNNs that implicitly learn spectral features, or lightweight architectures that treat the STFT as an isolated preprocessing operation, our approach natively embeds a fixed-basis time-frequency transformation directly within the network execution graph using parallel 1D convolutional layers. This design eliminates separate preprocessing stages, reduces runtime, and allows a highly compact backend 2D CNN topology to operate immediately on structured complex time-frequency feature maps. By combining this structured input with depthwise convolutions and residual connections, TinyBioNet achieves state-of-the-art performance with only 5.6k parameters. We further optimize the model for embedded targets through aggressive low-bit quantization. Comprehensive evaluations on three public datasets demonstrate robust generalization across various biomedical signal modalities. The proposed framework achieves classification accuracies of up to 98.89% and 99.38% for PPG and ACC signals, respectively, while maintaining negligible performance degradation under 4-bit integer quantization.
Natural products from fungi are a significant source for drug discovery. This study investigates six previously uncharacterized metabolites isolated from the poisonous mushroom Tricholoma pardinum using an integrated in silico approach to evaluate their therapeutic potential. Pharmacokinetic (ADMET) profiling predicted varied drug-likeness and toxicity profiles, with several compounds showing potential to cross the blood-brain barrier. Molecular docking and MM-GBSA calculations identified compound 1 as a potent inhibitor of Poly [ADP-ribose] polymerase 1 (PARP1), with a strong binding affinity (XP GScore: -6.756; ΔGbind: -48.75 kcal/mol). This interaction is anchored by a robust network of hydrogen bonds with key residues, including ASP914, CYS908, and THR866. Similarly, compound 6 emerged as a strong binder to Phosphatidylinositol 5-phosphate 4-kinase type-2 gamma (PIP4K2γ) (XP GScore: -7.705; ΔGbind: -45.94 kcal/mol), with its binding stabilized by extensive hydrophobic interactions complemented by a critical hydrogen bond with the residue Methionine 206 (MET206). Subsequent 100 ns molecular dynamics simulations confirmed the high stability of both protein-ligand complexes, validating the persistence of these key interactions. These computational findings highlight the potential of metabolites from T. pardinum as novel scaffolds for developing anticancer agents targeting PARP1 and PIP4K2γ, warranting further experimental validation.
Background Conventional surgical instruments made from Stainless steel (SS), Titanium (Ti), and Tantalum (Ta) are widely utilized because of their excellent corrosion resistance and strength-to-weight ratio. However, these materials lack antibacterial properties, which could increase the risk of surgical site infections. Recent advancements in antimicrobial and biocompatible coatings, particularly Ag-Ta 2 O 5 (Silver-Tantalum Pentoxide), present better solutions for new surgical instrument design. Despite numerous studies published on this topic, a comprehensive review specifically addressing Ag-Ta 2 O 5 coatings remains absent, resulting in fragmented information across the literature. Systematizing this information and consolidating the developments in this field are beneficial, and this is the motivation behind this review. Objective For these reasons, this review examines the development and applications of these advanced coatings, with a focus on their antimicrobial and biocompatible properties. Methods We critically examine the technological challenges, innovative coating methodologies, and the comparative advantages of Ag-Ta 2 O 5 over traditional and hybrid coatings. Results Furthermore, this review identifies future research directions and proposes strategic collaborations among clinicians, engineers, policymakers, and materials scientists to expedite the clinical adoption of these coatings. Conclusion It is envisioned that this review paper will serve as a valuable source of information for engineers, researchers, and clinicians to stay current with the latest developments in this area and for new researchers to initiate their exploration of coating technology.
BackgroundAs an integral part of the comprehensive treatment strategy for hepatocellular carcinoma (HCC) in China, Traditional Chinese Medicine (TCM) offers unique therapeutic advantages. This study aims to elucidate the anti-HCC mechanism of ZhenWu Tang, a classic TCM formula, by employing an integrated approach of network pharmacology and molecular docking.MethodsThe chemical components of ZhenWu Tang were retrieved from the TCMSP database and published literature. Their corresponding potential targets were then predicted via the SwissTargetPrediction database. The names of target proteins were standardized using the UniProt database. Effective therapeutic targets for hepatocellular carcinoma (HCC) were obtained from databases including OMIM and GeneCards. A Venn diagram was generated to identify the overlapping targets between ZhenWu Tang and HCC. A protein-protein interaction (PPI) network for these potential therapeutic targets was constructed using the STRING database and visualized with Cytoscape 3.7.2. Finally, "drug-component-target" and "drug-component-anti-HCC-target" network diagrams were constructed. Using metascape platform, GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Gene and Genomes) enrichment analysis were carried out on the potential anti hepatoma targets of ZhenWu Tang. Autodock was used to predict the binding mode and affinity between the active compounds and the core targets.ResultsAccording to oral bioavailability and drug like conditions, 61 chemical components of ZhenWu Tang were collected, including the main active components α-Amyrin, Pachymic acid, Kaempferol, β-Sitosterol et al., 249 Zhenwu Tang anti HCC targets were obtained. Protein-Protein interaction (PPI) network calculated that there are mainly TP53, AKT1, GAPDH, TNF and other key targets. Enrichment analysis yielded 218 signal pathways, involving the role of MicroRNAs in cancer, cancer proteoglycan, PI3K/AKT signal pathway, MAPK signal pathway, FoxO signal pathway, etc. Molecular docking results showed that Zhenwu Tang had good affinity with liver cancer protein targets.ConclusionZhenWu Tang can treat HCC by inhibiting tumor cell proliferation and promoting tumor cell apoptosis through multiple components, multiple targets and multiple pathways.
Abstract Background Artificial intelligence (AI) has generated rapidly growing research in breast cancer diagnosis and treatment, yet its intellectual structure, collaboration patterns, and thematic evolution remain unmapped. Objective To analyze the global research landscape of AI in breast cancer from 2001 to 2025, identifying publication trends, contributors, collaboration networks, thematic clusters, and translational gaps. Methods A bibliometric analysis of 7673 Web of Science articles. VOSviewer was used for co-authorship networks and keyword co-occurrence with overlay visualization; CiteSpace for burst detection. Results The field grew exponentially (29.20% annual growth), with 88.41% of publications from 2020–2025. China (31.63%) and the United States (21.10%) dominated output, but co-authorship networks revealed limited international collaboration for China (22.4% non-Chinese co-authors) and structural exclusion of low- and middle-income countries. Seven keyword clusters showed persistent separation between technology-centric and clinically oriented terms. Temporal overlay revealed a shift from computer-aided detection (pre-2015) to deep learning (2015–2022) and explainable AI, federated learning, and vision transformers (2022–2025). Burst detection confirmed “feature selection” (strength=19.53) and “computer aided detection” (strength=20.24) as historical hotspots; limited recent bursts indicate emerging frontiers are still accumulating citation impact. Conclusions AI research in breast cancer has expanded rapidly with evolving themes, yet a translational gap persists between innovation and clinical integration. Future efforts should prioritize prospective validation, international collaboration with underrepresented regions, and standardized frameworks for integrating explainable and privacy-preserving AI into workflows.
Background: Osteoporosis and acute myocardial infarction (AMI) are major health challenges in the aging population. Osteoporosis increases bone fragility, while AMI, often due to atherosclerosis, causes myocardial ischemia and inflammation. Their co-morbidity and shared mechanisms remain unclear. Objective: To explore the causal relationship and shared molecular mechanisms between osteoporosis and AMI. Methods: GWAS data from the Risteys FinnGen R9 database (osteoporosis: 621 cases, 122,861 controls) and the IEU Open GWAS program (AMI: 20,917 cases, 461,823 controls) were analyzed using Mendelian Randomization (IVW, MR-Egger, weighted median). Gene expression datasets (GSE56815, GSE48060) were used to identify differentially expressed genes (DEGs). Functional enrichment, immune infiltration (MCPcounter), and weighted gene co-expression network analysis (WGCNA) were performed, and overlapping hub genes were identified. Results: MR analysis demonstrated a significant causal association between osteoporosis and AMI. Transcriptomic analysis revealed 2434 DEGs in osteoporosis and 2827 in AMI. Enrichment highlighted pathways including immune regulation, MAPK signaling, and cancer-related pathways. Immune infiltration showed altered monocytes and dendritic cells in osteoporosis, and cytotoxic lymphocytes and neutrophils in AMI. WGCNA identified 6 modules in osteoporosis and 11 in AMI, with 1423 common hub genes. Conclusion: Osteoporosis and AMI share genetic and molecular mechanisms, especially involving inflammation and calcium signaling. These findings provide new insights into their co-morbidity and suggest that targeting shared therapeutic pathways may support integrated strategies for improving bone and cardiovascular health.
BackgroundThe comet assay is a sensitive and widely used technique for assessing DNA damage at the single-cell level. Despite its advantages, traditional manual scoring methods remain time-consuming, subjective and limited in scalability, posing challenges for high-throughput and standardized analysis.ObjectiveThis study aims to develop and evaluate a deep learning-based system for automated comet assay image classification, addressing limitations of manual and semi-automated approaches while enhancing accuracy, reproducibility and processing efficiency.MethodA YOLOv5-based object detection model was trained on a dataset of 875 annotated comet assay images, curated through a three-step expert-reviewed process. Various hyperparameters and data augmentation techniques were optimized to improve performance. The dataset was split into training, validation and test sets, and model performance was evaluated using mAP, precision, recall and confusion matrix analysis.ResultsThe model achieved strong performance, with mAP@0.5 reaching 0.98 and recall exceeding 0.8. Detailed analyses revealed robust learning behavior and generalization capacity. Visual outputs, including precision-recall curves and class-wise confusion matrices, confirmed high classification accuracy, although overlapping comet structures and class imbalance posed challenges. The model demonstrated improved scalability and processing speed compared to traditional tools, supporting its integration into web-based applications.ConclusionThe proposed YOLOv5-based system offers a scalable and accurate solution for automating comet assay analysis. It significantly enhances throughput and reduces human error, supporting its application in genotoxicity testing, biomonitoring and molecular epidemiology. Future work will focus on handling overlapping structures, benchmarking against existing tools and optimizing deployment in real-world laboratory settings.
BackgroundRecognizing that walking is a critical marker of independence, fall risk, and overall health in adults with intellectual disability (ID) and that comparative evidence by Down syndrome (DS) status remains limited, this study aimed to quantify spatial, temporal, and kinematic gait differences across flat and compliant surfaces in adults with and without DS and to evaluate the classification performance of Kinect-derived gait data using artificial neural networks (ANNs).MethodsCross-sectional, exploratory study at a state special education center in eastern Türkiye. Sixty-nine participants aged 18-27 years (ID without DS: n = 46; ID with DS: n = 23; mild-moderate levels) completed three 3-m trials at preferred speed on flat (concrete) and compliant (foam) surfaces. Kinect V1 recorded 3D joint trajectories. Step count, step length, step duration, and walking speed were computed via a threshold-based method. ANN models were trained to classify by DS status, gender, and disability level/age.ResultsOn compliant surfaces, step count and walking time increased while walking speed decreased relative to flat surfaces. Performance decrements were more pronounced in participants with DS and those with moderate ID. Women took more steps and walked more slowly than men on both surfaces. ANN models showed high correlation values across tasks (R = 0.96-0.99); however, these findings should be interpreted as exploratory.ConclusionA Kinect-based approach may provide a practical and accessible method for characterizing gait in adults with ID with and without DS; however, the findings should be interpreted with caution due to methodological limitations and the need for further validation. Findings support integrating surface-specific balance and gait training, age- and gender-sensitive strategies, and individualized programs for DS. ANN-based analyses may provide preliminary insights for classification purposes; however, their clinical applicability requires further validation and should be interpreted cautiously.
BackgroundThe use of social media platforms like YouTube has surged among patients and their families seeking medical information. Despite the widespread use of video content for health education, no previous study has systematically evaluated the quality of YouTube videos on dynamic spinal stabilization.MethodsA YouTube search using the keyword "dynamic stabilization" was conducted in December 2024. Thirty eligible videos were assessed independently by two neurosurgery specialists using the DISCERN scale. Video characteristics and content quality were analyzed using descriptive statistics, correlation analysis, and regression models.ResultsThe overall quality of the videos was low, with a mean DISCERN score of 39.15. A negative correlation was found between the number of views and DISCERN score (r = -0.28), while positive correlations were observed between DISCERN scores and video duration (r = 0.19), and between DISCERN-1 and DISCERN-2 scores (r = 0.957). Videos including: a medically trained speaker,explanation of dynamic stabilization differences,surgical procedure details, andpre/post-operative information scored significantly higher, with each factor contributing to an average increase of 7.09 points (p < 0.05). ConclusionMost YouTube videos on dynamic stabilization offer low-quality and potentially misleading information. Medically trained contributors consistently produce more reliable and informative content, yet such videos often receive less viewer engagement. Increasing the visibility and clarity of evidence-based content is essential to improve patient education through social media.
BackgroundIn the digital era, iGaming has become a rapidly expanding phenomenon that generates complex behavioral patterns and increases the risk of problematic gambling. The integration of technologies such as Artificial Intelligence of Things and the Internet of Behavior enables a shift from reactive to preventive approaches in public health protection.ObjectiveThis study combines clinical practice, AI, machine learning and the Internet of Behavior concept to analyze player behavior patterns and develop machine learning models for early detection of addiction risk through behavioral markers.MethodsBehavioral data were collected from an online game supplier operating through 52 operators in Republika Srpska, Croatia, Romania, Brazil, Somalia, and Mali. Psychiatric expertise and clinical experience were applied to identify harmful behavioral markers, which served as inputs for training MLP neural networks. Models trained per country classified player behavior into recreational, risky, and problematic categories.ResultsThe analysis included 109,418 players across three continents, aggregating 5,135,179,510 online slot game bets. Results revealed significant cross-country variation in risky and problematic gambling, shaped by socio-economic, cultural, and regulatory factors.ConclusionThe integration of AI-driven behavioral analysis with psychiatric insight provides a robust framework for early risk detection and personalized interventions supporting responsible gaming.
BackgroundCardiovascular disease, especially heart failure, is a substantial global health issue. By integrating PHR with machine learning, early disease detection could be made possible.ObjectiveIn this study, an attempt was made to develop and fine-tune an AI model that would forecast the likelihood of heart failure based on patient data in PHRs.MethodsData from 1025 patients and 12 clinical/demographic criteria were used. An untuned multilayer perceptron (MLP) with two hidden layers (10 and 5 neurons) was first trained (1000 epochs). Then, using the same dataset, we performed systematic hyperparameter tuning (grid search with 5-fold cross-validation) for Logistic Regression, Random Forest, SVM, and an enhanced MLP. Performance metrics included accuracy, precision, recall, F1-score, MCC, and ROC-AUC with 95% confidence intervals.ResultsThe original untuned MLP gave a mean accuracy of 0.7244 (±0.0245) and mean ROC-AUC of 0.724 (±0.038). After tuning, Random Forest achieved the highest performance (AUC = 0.959, 95% CI 0.924-0.986; accuracy = 0.890). The tuned MLP reached AUC = 0.830 (CI 0.763-0.893), outperforming the untuned version and showing comparable performance to Logistic Regression (AUC = 0.824) and SVM (AUC = 0.842).ConclusionThese results suggest potential use of AI models to anticipate the risk of heart failure from a subject's medical history and provide an avenue toward scalable and personal medicine, resulting in improved early prevention and treatment of cardiovascular disease.
PurposeArtificial intelligence (AI) shows considerable potential for sports injury prediction, yet a comprehensive methodological review of its empirical applications remains limited. This study aimed to systematically review the empirical literature on the use of AI and machine learning (ML) for sports injury prediction.MethodsFollowing the PRISMA 2020 guidelines, a systematic search was conducted across PubMed, IEEE Xplore, SPORTDiscus, Web of Science, and Scopus for literature published between January 2015 and March 2026. In addition to the structured database search, a small number of database-recommended articles identified through platform recommendation functions were also screened for eligibility. After a multi-stage screening process, 18 empirical studies were included in the final qualitative synthesis.ResultsRisk of bias assessment using PROBAST indicated that only one study (5.6%) had a low overall risk of bias, whereas most studies were judged to be high risk, primarily because of weaknesses in the analysis domain and reliance on internal validation. Across the included studies, AI-based models demonstrated potential for handling multidimensional training load, physiological, biomechanical, and psychological data; however, most prediction models relied exclusively on internal validation, limiting confidence in their generalizability.ConclusionAI demonstrates clear promise for sports injury prediction, but the current evidence base remains constrained by limited external validation, inconsistent reporting practices, and the continued overrepresentation of male cohorts. Future research should prioritize methodological rigor, broader geographic and demographic representation, standardized reporting, and explainable AI (XAI) approaches to enhance the trustworthiness and practical utility of these models for coaches and clinicians.
