Search Results (41)

Speech alterations are early markers of neurodegenerative diseases. Transformer-based speech models such as Whisper have advanced automated speech assessment, but most systems rely on cloud-based computation, raising privacy concerns. On-device processing could offer a scalable and privacy-preserving alternative. This research’s objective was to evaluate whether a fully on-device speech analysis pipeline can achieve competitive accuracy in detecting Alzheimer’s disease and to quantify the contributions of acoustic, linguistic, and embedding features. Therefore, we developed an iOS application running all components, including acoustic analysis, two transformer-based speech-to-text modules (WhisperBase and quantized CrisperWhisper), linguistic feature extraction, and embedding generation, directly on the device. Using the ADReSS Challenge 2020 dataset (N = 156), we trained classical machine-learning classifiers across 20 configurations and evaluated them via a stratified 10-fold cross-validation. Area under the receiver operating curve (AUC), accuracy, precision, recall, and F1 scores were used as performance metrics. An ablation study examined the relevance of each feature group. The best-performing setup (Random Forest with CrisperWhisper transcription and Apple embeddings) achieved an accuracy of 85.4% and an AUC of 0.85. Performance was 5–7% below benchmark models relying on manual transcripts or server-based processing. Embedding features provided the strongest individual contribution, but the highest accuracy required combining acoustic, linguistic, and embedding information. A fully on-device pipeline for Alzheimer’s disease detection from speech is feasible and achieves competitive accuracy while maintaining strict data privacy. These findings highlight the potential of on-device transformer architectures for scalable, privacy-preserving digital screening. Future work should validate the approach in larger and more diverse cohorts. Full article

Background: Pneumonia of unknown etiology (PUE), defined as pneumonia cases without an identified pathogen at the time of clinical presentation, represents a critical clinical warning signal for emerging infectious disease (EID) outbreaks with pandemic potential. Yet, conventional pathogen-centric surveillance systems suffer from an inherent blind spot: they cannot detect early clustering signals before the causative agent is identified, creating a window of vulnerability during novel pathogen emergence. To address this gap, this study aims to develop a deep learning model that leverages unstructured chest imaging text—a routinely available clinical data stream—to enable real-time, automated screening of PUE cases and early warning of EID clusters, independent of prior pathogen knowledge, within an integrated multi-pathogen surveillance framework. Methods: We retrospectively collected data from 8860 patients with respiratory illnesses at a tertiary hospital in Beijing, China, including 980 PUE cases (11.1%) and 7880 known-etiology pneumonia cases. A deep learning model (RoBERTa with attention enhancement) was developed using unstructured chest imaging reports. The Matthews correlation coefficient (MCC) curve was employed to determine the optimal decision threshold. Model performance was assessed for PUE case identification and clustering signal detection on a test set. Results: The model achieved an area under the receiver operating characteristic curve of 0.986 (95% CI: 0.981–0.991). At the optimal threshold of 0.08, selected by maximizing the Matthews correlation coefficient (MCC)—a balanced metric that accounts for all four confusion matrix outcomes—sensitivity was 89.8%, and specificity was 97.0% for identifying PUE cases. In a simulated surveillance exercise, the model showed a high correlation between the predicted and actual case counts (Pearson’s r = 0.901), suggesting its potential to detect abnormal clustering signals prior to pathogen identification. Conclusions: The developed model demonstrates potential to detect clustering signals of PUE caused by unknown pathogens and can be integrated with hospital information systems, providing a feasible, low-cost tool for integrated surveillance of pathogens with pandemic potential. This approach enables earlier outbreak detection and supports public health decision-making during the critical window before pathogen identification. Full article

We present FingerType, a one-handed text input method based on thumb-to-finger gestures. FingerType detects tap events from 3D hand data using a Temporal Convolutional Network (TCN) and decodes the tap sequence into words with an n-gram language model. To inform the design, we examined thumb-to-finger interactions and collected comfort ratings of finger regions. We used these results to design an improved T9-style key layout. Our system runs at 72 frames per second and reaches 94.97% accuracy for tap detection. We conducted a six-block user study with 24 participants and compared FingerType with controller input and touch input. Entry speed increased from 5.88 WPM in the first practice block to 10.63 WPM in the final block. FingerType also supported more eyes-free typing: attention on the display panel within ±15° of head-gaze was 84.41%, higher than touch input (69.47%). Finally, we report error patterns and WPM learning curves, and a model-based analysis suggests improving gesture recognition accuracy could further increase speed and narrow the gap to traditional VR input methods. Full article

Background and Objectives: Hip joint disorders exhibit diverse and overlapping radiological features, complicating early diagnosis and limiting the diagnostic value of single-modality imaging. Isolated imaging or clinical data may therefore inadequately represent disease-specific pathological characteristics. Materials and Methods: This retrospective study included 605 hip joints from Center A (2018–2024), comprising normal hips, osteoarthritis, osteonecrosis of the femoral head (ONFH), and femoroacetabular impingement (FAI). An independent cohort of 24 hips from Center B (2024–2025) was used for external validation. A multimodal deep learning framework was developed to jointly analyze radiographs, CT volumes, and clinical texts. Features were extracted using ResNet50, 3D-ResNet50, and a pretrained BERT model, followed by attention-based fusion for four-class classification. Results: The combined Clinical+X-ray+CT model achieved an AUC of 0.949 on the internal test set, outperforming all single-modality models. Improvements were consistently observed in accuracy, sensitivity, specificity, and decision curve analysis. Grad-CAM visualizations confirmed that the model attended to clinically relevant anatomical regions. Conclusions: Attention-based multimodal feature fusion substantially improves diagnostic performance for hip joint diseases, providing an interpretable and clinically applicable framework for early detection and precise classification in orthopedic imaging. Full article

Introduction: Chronic obstructive pulmonary disease (COPD) is a leading cause of morbidity and mortality. Early identification and timely intervention for COPD exacerbations can reduce hospitalizations and complications, as well as improve patient outcomes. Methods: To develop and evaluate predictive models for COPD exacerbations using machine learning (ML), we performed a retrospective study using intensive care unit patient records. Records including 31,667 clinical notes and 10,489 vital signs were used to train and validate two machine learning models to predict COPD exacerbations in patients with known or suspected COPD. Predictive performance was evaluated for support vector machine, quadratic discriminant analysis, and adaptive boosting algorithms using area under the receiver operating characteristic curve (AUC). Results: The clinical note-based support vector machine model achieved an AUC of 0.81 and accuracy of 84.0% in predicting COPD exacerbations. Data from patient monitors and hospital information systems provided sufficient information for accurate prediction, demonstrating the utility of combining physiological signals with clinical text data. Discussion: Clinically available patient data and vital signs can effectively predict COPD exacerbations, potentially enabling earlier interventions, improved outcomes, and reduced healthcare burden. These findings suggest that integrating unstructured clinical notes with structured vital signs using ML frameworks may improve early detection of exacerbation risk, thus enabling appropriate patient counseling, triage, and treatment based on COPD severity. Full article

Deep anomaly detection in multilingual environments remains challenging due to limited labeled data, semantic inconsistency across languages, and the unstable distribution of rare abnormal patterns. These challenges are particularly severe in low-resource scenarios—characterized by scarce labeled anomaly data and non-standardized terminology—where conventional supervised or transfer-based models suffer from semantic drift and feature mismatch. To address these limitations, a data-driven cross-lingual anomaly detection framework, LR-SSAD, is proposed. Targeting paired text and behavioral data without requiring parallel translation corpora, the framework is built upon the joint optimization of complementary self-supervised objectives. A cross-lingual masked prediction module is designed to capture language-invariant semantic structures to align semantic spaces, while a Mamba-based sequence reconstruction module leverages its linear computational complexity ($\mathcal{O}\left(N\right)$) to efficiently model long-range dependencies in transaction histories, overcoming the computational bottlenecks of quadratic attention mechanisms. To further enhance robustness under noisy supervision, a noise-aware pseudo-label refinement mechanism is introduced. Evaluated on a newly constructed real-world financial dataset (spanning January–June 2023) comprising 1.2 million multilingual texts and 420,000 transaction sequences, experimental results demonstrate that LR-SSAD achieves substantial improvements over state-of-the-art baselines. The model achieves an accuracy of 0.932, a precision of 0.914, a recall of 0.891, and an F1-score of 0.902, with the Area Under the Curve (AUC) reaching 0.948. The proposed framework provides a scalable and data-efficient solution for anomaly detection in real-world multilingual environments. Full article

The growing scale of libraries necessitates intelligent management solutions, particularly for book inventory tasks. To address the challenge of book spine recognition in dense, text-heavy environments, this study proposes an integrated approach combining an enhanced YOLOv11 model with a hyperparameter-optimized PaddleOCR framework. The methodology involves augmenting the YOLOv11 object detector with a Channel-Spatial Dual Attention Mechanism (CBAM) to better extract spine texture features and suppress interference from adjacent books. For the text recognition stage, PaddleOCR’s hyperparameters were task-optimized by adopting the RecAug data augmentation strategy, adjusting the curved text detection loss weight, expanding the character dictionary, and modifying the input image size. Experimental results on a self-constructed Book Spine Dataset show that the improved YOLOv11 achieved a segmentation accuracy of 97.4%, a 2.1% increase over the baseline, while reducing computational load and parameters. The optimized PaddleOCR saw its character error rate drop from 8.6% to 3.2%. Consequently, the end-to-end system attained a 96.8% single-book recognition accuracy in real bookshelf scenarios, demonstrating that this targeted strategy significantly enhances performance for intelligent library management. Full article

Aims/Background: The growing integration of artificial intelligence (AI) into clinical medicine has opened new possibilities for enhancing diagnostic accuracy, therapeutic decision-making, and biomedical innovation across several domains. This review is aimed to evaluate the clinical applications of AI across five key domains of medicine: diagnostic imaging, clinical decision support systems (CDSS), surgery, pathology, and drug discovery, highlighting achievements, limitations, and future directions. Methods: A comprehensive PubMed search was performed without language or publication date restrictions, combining Medical Subject Headings (MeSH) and free-text keywords for AI with domain-specific terms. The search yielded 2047 records, of which 243 duplicates were removed, leaving 1804 unique studies. After screening titles and abstracts, 1482 records were excluded due to irrelevance, preclinical scope, or lack of patient-level outcomes. Full-text review of 322 articles led to the exclusion of 172 studies (no clinical validation or outcomes, n = 64; methodological studies, n = 43; preclinical and in vitro-only, n = 39; conference abstracts without peer-reviewed full text, n = 26). Ultimately, 150 studies met inclusion criteria and were analyzed qualitatively. Data extraction focused on study context, AI technique, dataset characteristics, comparator benchmarks, and reported outcomes, such as diagnostic accuracy, area under the curve (AUC), efficiency, and clinical improvements. Results: AI demonstrated strong performance in diagnostic imaging, achieving expert-level accuracy in tasks such as cancer detection (AUC up to 0.94). CDSS showed promise in predicting adverse events (sepsis, atrial fibrillation), though real-world outcome evidence was mixed. In surgery, AI enhanced intraoperative guidance and risk stratification. Pathology benefited from AI-assisted diagnosis and molecular inference from histology. AI also accelerated drug discovery through protein structure prediction and virtual screening. However, challenges included limited explainability, data bias, lack of prospective trials, and regulatory hurdles. Conclusions: AI is transforming clinical medicine, offering improved accuracy, efficiency, and discovery. Yet, its integration into routine care demands rigorous validation, ethical oversight, and human-AI collaboration. Continued interdisciplinary efforts will be essential to translate these innovations into safe and effective patient-centered care. Full article

Large language models (LLMs) pose significant challenges to content authentication, as their sophisticated generation capabilities make distinguishing AI-produced text from human writing increasingly difficult. Current detection methods suffer from limited information capture, poor rate–distortion trade-offs, and vulnerability to adversarial perturbations. We present CurveMark, a novel dual-channel detection framework that combines probability curvature analysis with dynamic semantic watermarking, grounded in information-theoretic principles to maximize mutual information between text sources and observable features. To address the limitation of requiring prior knowledge of source models, we incorporate a Bayesian multi-hypothesis detection framework for statistical inference without prior assumptions. Our approach embeds imperceptible watermarks during generation via entropy-aware, semantically informed token selection and extracts complementary features from probability curvature patterns and watermark-specific metrics. Evaluation across multiple datasets and LLM architectures demonstrates 95.4% detection accuracy with minimal quality degradation (perplexity increase < 1.3), achieving 85–89% channel capacity utilization and robust performance under adversarial perturbations (72–94% information retention). Full article

Introduction: The figure-of-eight walk test (F8WT) assesses gait on a curved path, reflecting everyday walking complexity. Despite recognized validity among elderly individuals, its application in neurological disorders remains inadequately explored. This scoping review summarizes evidence regarding F8WT use, validity, and clinical applicability among individuals with neurological disorders. Methods: A systematic literature search was conducted in the PubMed, Scopus, Embase, and Web of Science databases. After reading the full text of the selected studies and applying predefined inclusion criteria, seven studies, involving participants with multiple sclerosis (n = 3 studies), Parkinson’s disease (n = 2 studies), and stroke (n = 2 studies), were included based on pertinence and relevance to the topic. Results: F8WT demonstrated strong reliability and validity across various neurological populations and correlated significantly with established measures of gait, balance, and disease severity. Preliminary evidence supports its ability to discriminate individuals at increased fall risk and detect subtle motor performance changes. Discussion: The F8WT emerges as a valuable tool, capturing multifaceted gait impairments often missed by linear walking assessments. Sensitive to subtle functional changes, it is suitable for tracking disease progression and intervention efficacy. Conclusions: F8WT is reliable and clinically relevant, effectively identifying subtle, complex walking impairments in neurological disorders. Full article

Image-based spam poses a significant challenge for traditional text-based filters, as malicious content is often embedded within images to bypass keyword detection techniques. This study investigates and compares the performance of six machine learning models—ResNet50, XGBoost, Logistic Regression, LightGBM, Support Vector Machine (SVM), and VGG16—using a curated dataset containing 678 legitimate (ham) and 520 spam images. The novelty of this research lies in its comprehensive side-by-side evaluation of diverse models on the same dataset, using standardized dataset preprocessing, balanced data splits, and validation techniques. Model performance was assessed using evaluation metrics such as accuracy, receiver operating characteristic (ROC) curve, precision, recall, and area under the curve (AUC). The results indicate that ResNet50 achieved the highest classification performance, followed closely by XGBoost and Logistic Regression. This work provides practical insights into the strengths and limitations of traditional, ensemble-based, and deep learning models for image-based spam detection. The findings can support the development of more effective and generalizable spam filtering solutions in multimedia-rich communication platforms. Full article

Type 1 diabetes (T1D) is an autoimmune condition characterized by the destruction of insulin-producing pancreatic beta cells, leading to lifelong insulin dependence and significant complications. Early detection of T1D is essential to delay disease onset and improve outcomes. Recent advancements in artificial intelligence (AI) and machine learning (ML) have provided powerful tools for predicting and diagnosing T1D. This systematic review evaluates the current landscape of AI/ML-based approaches for early T1D detection. A comprehensive search across PubMed, EMBASE, Science Direct, and Scopus identified 1447 studies, of which 10 met the inclusion criteria for narrative synthesis after screening and full-text review. The studies utilized diverse ML models, including logistic regression, support vector machines, random forests, and artificial neural networks. The datasets encompassed clinical parameters, genetic risk markers, continuous glucose monitoring (CGM) data, and proteomic and metabolomic biomarkers. The included studies involved a total of 49,172 participants and employed case–control, retrospective cohort, and prospective cohort designs. Models integrating multimodal data achieved the highest predictive accuracy, with area under the curve (AUC) values reaching up to 0.993 in sex-specific models. CGM data and plasma biomarkers, such as CXCL10 and IL-1RA, also emerged as valuable tools for identifying at-risk individuals. While the results highlight the potential of AI/ML in revolutionizing T1D risk stratification and diagnosis, challenges remain. Data heterogeneity and limited model generalizability present barriers to widespread implementation. Future research should prioritize the development of universal frameworks and real-world validation to enhance the reliability and clinical integration of these tools. Ultimately, AI/ML technologies hold transformative potential for clinical practice by enabling earlier diagnosis, guiding targeted interventions, and improving long-term patient outcomes. These advancements could support clinicians in making more informed, timely decisions, thus reducing diagnostic delays and paving the way for personalized prevention strategies in both pediatric and adult populations. Full article

Background: Recent evidence has shown that insulin resistance (IR), a hallmark of nonalcoholic fatty liver disease, predicts bladder cancer (BC) presence. However, the best surrogate marker of IR in predicting BC is still unclear. This study examined the relationships among ten surrogate markers of IR and the presence of BC. Methods: Data from 209 patients admitted to two urology departments from September 2021 to October 2024 were retrospectively analyzed. Individuals (median age 70 years) were divided into two groups (123 and 86 patients, respectively) based on the presence/absence after cystoscopy/TURB of non-metastatic BC. Univariate logistic regression was used to determine the relationships between groups, and the following IR parameters: Triglyceride–Glucose (TyG) index, TyG-BMI, HOMA-IR HOMAB, MetS-IR, Single Point Insulin Sensitivity Estimator, Disposition Index, non-HDL/HDL, TG/HDL-C ratio and Lipoprotein Combine Index. Stepwise logistic regressions were carried out to evaluate the significant predictions and LASSO regression to confirm any significant variable(s). The predictive value of the index test for coexistent BC was evaluated using receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC). Results: The univariate analysis revealed that the TyG index and MetS-IR were associated with the BC presence. Specifically, the associations of the TyG index and MetS-IR were more significant in participants =/> 65 years old. In multivariate analysis, the stepwise logistic regression, evaluating the most representative variables at univariate analysis, revealed a prediction of BC by only TyG index (OR 2.51, p = 0.012), confirmed by LASSO regression, with an OR of 3.13, p = 0.004). Assessing the diagnostic reliability of TyG, it showed an interesting predictive value for the existence of BC (AUC = 0.60; 95% CI, 0.51–0.68, cut-off 8.50). Additionally, a restricted cubic spline model to fit the dose–response relationship between the values of the index text (TyG) and the BC evidenced the presence of a non-linear association, with a high predictive value of the first knot, corresponding to its 10th percentile. The decision curve analysis confirmed that the model (TyG) has utility in supporting clinical decisions. Conclusions: Compared to other surrogate markers of IR, the TyG index is effective in identifying individuals at risk for BC. A TyG threshold of 8.5 was highly sensitive for detecting BC subjects and may be suitable as an auxiliary diagnostic criterion for BC in adults, mainly if less than 65 years old. Full article

Introduction: The prompt prehospital identification of intracerebral haemorrhage (ICH) may allow very early delivery of treatments to limit bleeding. Current prehospital stroke assessment tools have limited accuracy for the detection of ICH as they were designed to recognise all strokes, not ICH specifically. This systematic review aims to evaluate the performance of prehospital models in distinguishing ICH from other causes of suspected stroke. Methods: We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Following a predefined strategy, we searched three electronic databases via Ovid (MEDLINE, EMBASE, and CENTRAL) in July 2023 for studies published in English, without date restrictions. Subsequently, data extraction was performed, and methodological quality was assessed using the Prediction Model Risk of Bias Assessment Tool. Results: After eliminating duplicates, 6194 records were screened for titles and abstracts. After a full-text review of 137 studies, 9 prediction studies were included. Five of these described prediction models were designed to differentiate between stroke subtypes, three distinguished between ICH and ischaemic stroke, and one model was developed specifically to identify ICH. All studies were assessed as having a high risk of bias, particularly in the analysis domain. The performance of the models varied, with the area under the receiver operating characteristic curve ranging from 0.73 to 0.91. The models commonly included the following as predictors of ICH: impaired consciousness, headache, speech or language impairment, high systolic blood pressure, nausea or vomiting, and weakness or paralysis of limbs. Conclusions: Prediction models may support the prehospital diagnosis of ICH, but existing models have methodological limitations, making them unreliable for informing practice. Future studies should aim to address these identified limitations and include a broader range of suspected strokes to develop a practical model for identifying ICH. Combining prediction models with point-of-care tests might further improve the detection accuracy of ICH. Full article

Visually Impaired Persons (VIPs) have difficulty in recognizing vehicles used for navigation. Additionally, they may not be able to identify the bus to their desired destination. However, the bus bay in which the designated bus stops has not been analyzed in the existing literature. Thus, a guidance system for VIPs that identifies the correct bus for transportation is presented in this paper. Initially, speech data indicating the VIP’s destination are pre-processed and converted to text. Next, utilizing the Arctan Gradient-activated Recurrent Neural Network (ArcGRNN) model, the number of bays at the location is detected with the help of a Global Positioning System (GPS), input text, and bay location details. Then, the optimal bay is chosen from the detected bays by utilizing the Experienced Perturbed Bacteria Foraging Triangular Optimization Algorithm (EPBFTOA), and an image of the selected bay is captured and pre-processed. Next, the bus is identified utilizing a You Only Look Once (YOLO) series model. Utilizing the Sub-pixel Shuffling Convoluted Encoder–ArcGRNN Decoder (SSCEAD) framework, the text is detected and segmented for the buses identified in the image. From the segmented output, the text is extracted, based on the destination and route of the bus. Finally, regarding the similarity value with respect to the VIP’s destination, a decision is made utilizing the Multi-characteristic Non-linear S-Curve-Fuzzy Rule (MNC-FR). This decision informs the bus conductor about the VIP, such that the bus can be stopped appropriately to pick them up. During testing, the proposed system selected the optimal bay in 247,891 ms, which led to deciding the bus stop for the VIP with a fuzzification time of 34,197 ms. Thus, the proposed model exhibits superior performance over those utilized in prevailing works. Full article