Multimodal learning systems typically assume strong alignment between data modalities, an assumption frequently violated in real-world scenarios such as ecological monitoring, where visual and acoustic observations are weakly or inconsistently paired. This paper proposes a hybrid quantum-classical fusion framework that integrates independently learned modality-specific representations through a Parameterized Quantum Circuit (PQC). The proposed Quantum Fusion Layer (QFL) is designed to capture high-order cross-modal interactions under weak alignment conditions. Experimental evaluation on a multimodal bird species classification task shows that the proposed framework outperforms classical fusion baselines, achieving Accuracy and Macro F1-score of 0.9875 on a held-out test set. Additional analyses comprising stratified cross-validation, ablation studies, robustness evaluation under controlled noise conditions, a parameter-matched classical MLP control, and cross-domain generalization on a geographically out-of-distribution benchmark (Bird-SEA10) of 238 multimodal pairs from Singapore and Thailand indicate stable performance across data partitions, resilience to moderate input degradation, and meaningful generalization under geographic domain shift (cross-domain Accuracy =0.8445, Macro F1 =0.8382, ΔF1=−0.1493 relative to in-domain performance). The parameter-matched MLP control achieves Accuracy of 0.2125 and Macro F1 of 0.0897, suggesting that the observed performance advantage originates from the structural properties of the quantum circuit rather than from reduced parameter capacity. These findings indicate that quantum-enhanced feature transformations may provide an effective mechanism for multimodal fusion under weak alignment, with potential applicability to broader multimodal learning problems involving heterogeneous and imperfectly aligned data sources.

Accurate segmentation of retinal vessels is critical for the early detection of vision-threatening diseases. Although U-Net-based methods have shown strong performance, they often fail to capture thin vessels and preserve boundary details due to repeated downsampling. To overcome these limitations, we propose an enhanced U-shaped network that incorporates a multi-scale attention guided filtering module, allowing the model to retain edge details and suppress noise more effectively. Experiments conducted on the DRIVE, STARE, CHASE_DB1, and HRF datasets demonstrate that the proposed method consistently achieves the best results across multiple metrics. The improvements in F1 score and sensitivity confirm its capability to recover fine vascular structures and its potential for clinical application.

Retinal vessel segmentation plays a crucial role in the early diagnosis and treatment of ophthalmic diseases, yet manual segmentation methods are time-consuming and prone to human error. The increasing demand for automated solutions has led to the exploration of deep learning-based approaches that improve segmentation accuracy and efficiency. However, existing methods often struggle with poor image resolution, a lack of robustness across datasets, and challenges in detecting fine vessel structures, limiting their applicability in large-scale clinical settings. To address these limitations, this study proposes a deep learning-based framework using high-resolution network (HRNet) combined with an advanced preprocessing pipeline, including grayscale conversion, gamma correction, CLAHE, and normalization, to enhance image quality for segmentation. The proposed HRNet-based framework was evaluated on two publicly available datasets, DRIVE and CHASE_DB1. It achieved the highest sensitivity of 0.8193 on DRIVE and the highest specificity of 0.9807 on CHASE_DB1, demonstrating its capability to balance fine vessel detection and false positive reduction. These findings indicate that the proposed framework effectively improves the segmentation accuracy of ophthalmoscope images, enhancing its potential for clinical decision-making in ophthalmology.

Most voice-based Parkinson’s disease studies evaluated a narrow set of models on fixed feature spaces, which limited generalizability and obscured feature–model synergies. Many also neglected fold-wise fitting and runtime measurement, increasing the risk of data leakage and hindering telehealth deployment. Therefore, the study developed a leak-safe, runtime-aware factorial benchmark that crossed nine representation strategies with twelve classifiers, giving 108 pipelines under stratified 10-fold cross-validation on a 22-feature sustained-vowel dataset. All transformations were fit strictly within training folds, class imbalance was addressed with in-fold SMOTE, and four endpoints, accuracy, F1, AUC, and per-fold runtime, were jointly analyzed across classical machine learning, ensembles, and interaction-aware deep networks, explicitly profiling accuracy–latency trade-offs relevant to point-of-care screening. The results showed that simple, information-preserving filters paired with interaction-aware learners dominated the Pareto frontier. The top pipeline, Variance plus Deep Cross, achieved mean accuracy 94.37

Abstract Magnesium (Mg) alloys are promising candidates for biodegradable orthopedic implants due to their mechanical compatibility and biocompatibility; however, rapid corrosion and hydrogen gas evolution in physiological environments limit their clinical application. This study presents a green, biocompatible coating strategy using a polymer–herbal composite of polyvinyl alcohol and neem ( Azadirachta indica ) extract, deposited on AZ31 Mg alloy via electrophoretic deposition. Electrochemical analyses, including open circuit potential, electrochemical impedance spectroscopy, and potentiodynamic polarization, demonstrated enhanced corrosion resistance for the 5 % polyvinyl alcohol–neem coating, with increased charge transfer resistance and improved surface stability. Contact angle measurements ( n = 6) indicated slightly higher hydrophilicity for the 2 % coating; however, the difference was not statistically significant ( p > 0.05), suggesting comparable wettability. Scanning electron microscopy and energy-dispersive X-ray spectroscopy confirmed uniform, crack-free coatings with successful composite deposition. No antibacterial activity was observed, likely due to limited release of bioactive compounds from the polymer matrix. Overall, the coating effectively modifies corrosion behavior and surface properties, indicating potential for controlled degradation and improved performance of Mg-based biodegradable implants.

In year 2023, around 2.2 billion people globally have near or distance visual impairment. Previous systems lacked comprehensive functionality, focusing only on basic obstacle detection or standalone features like fall detection. Moreover, the studies struggled with bulky designs, poor low-light performance, and limited environmental awareness. To address these challenges, the study develops ObstaSense, a wearable Electronic-Travel-Aid (ETA) for obstacle detection and navigation assistance. The system employed TensorFlow Lite, a Raspberry Pi-5, and a Pi-Camera Module-V3 to detect objects (e.g., people, potholes, vehicles) and relay avoidance instructions via Bluetooth earbuds. Its Real-Time Navigation (RTN) feature combined Global Positioning System (GPS), a compass sensor, and Plus Codes for precise guidance, enhanced by Google's Speech-To-Text (STT) and Text-to-Speech (TTS). Operating at 4-10 Frames Per Second (FPS), ObstaSense further integrated the Gemini Application programming interface (API) for multilingual (50-languages) image-to-text conversion. The system achieved consistent results by leveraging precise RTN functionality, which uses compass sensor data and vibration feedback to guide users accurately. Offline dataset training and evaluation were conducted solely to support the deployment of a real-time embedded assistive system on Raspberry Pi 5. Obstacle avoidance performance varied across rows, with the highest accuracy (100%) in the first row, followed by 66.7% in the third row and 50% in the second row. ObstaSense aids visually impaired, elderly, and cognitively impaired users, aligning with Sustainable Development Goals (SDGs) 3 and 10 for inclusive well-being.

Digital twins (DT) technology has shown considerable growth in recent years. Previous studies have examined technologies in a variety of areas, including health care. However, limited studies have attempted to provide a thorough discussion of strategies for the seamless integration of DT into health care, particularly in the context of interoperability of heterogeneous medical data. This review examines the underlying concept of DT and its possible integration in healthcare, particularly in the context of healthcare interoperability. It also analyzes the main problems such as the lack of standardized protocols, the non-homogeneity of data formats and technical complexity. Finally, potential opportunities are highlighted such as standardized protocol, the creation of an open data platform and the empowerment of semantic interoperability. In conclusion, this review has provided valuable insights for many professionals, including researchers and healthcare providers, which will contribute to empowering patient-centered or personalized medicine and to the development of digital health.

Abstract Maritime Search and Rescue (SAR) operations are often challenged by vast search zones, poor visibility, and extreme lighting conditions, especially during nighttime missions. This study investigates the use of computer vision and object detection algorithms to automate life jacket detection and improve SAR effectiveness. To address the absence of domain-specific datasets, a custom image dataset featuring multiple life jacket types was developed. A two-fold methodology was adopted: evaluating the performance of YOLO object detection models (versions 5 through 12) on the dataset, and incorporating advanced image preprocessing techniques to enhance detection under challenging lighting conditions. The results demonstrate that preprocessing significantly improves detection performance in both overexposed and underexposed scenarios. Among all evaluated models, YOLOv10 achieved the strongest combination of precision and real-time inference speed (43.9 FPS on Tesla T4 GPU), making it a promising candidate for time-sensitive rescue applications. While individual cells of Tables 5, 6, 7, 8 and 9 show other detectors achieving higher precision under specific lighting × preprocessing combinations, YOLOv10 offers the best aggregate trade-off across the evaluated criteria. This work contributes a scalable benchmark solution for improving SAR outcomes by enabling faster and more reliable identification of individuals in distress at sea.

Chemical Exchange Saturation Transfer (CEST) MRI enables noninvasive molecular imaging by selectively saturating exchangeable protons. Conventional tissue-specific CEST analysis often relies on projecting segmentations derived from structural MRI onto the CEST plane, introducing dependence on anatomical priors and additional registration steps. This proof-of-concept study investigates whether voxel-wise brain tissue identity can be inferred directly from CEST Z-spectral features, allowing tissue maps to be reconstructed in the native CEST acquisition space without anatomical input during inference. Eight classifiers spanning classical machine-learning methods and deep-learning architectures were trained and evaluated using 45-point voxel-wise spectra derived from B0-corrected, normalized, and spline-interpolated two-dimensional CEST data. Data were collected from six healthy subjects with 72 scans. Five subjects were used for subject-wise leave-one-subject-out cross-validation, while the sixth was reserved for independent held-out testing. Reference tissue labels for white matter, gray matter, and cerebrospinal fluid were generated from T1-weighted segmentations using FSL FAST and SPM12 and registered to the CEST plane. The evaluated classifiers comprised a one-dimensional Transformer and seven comparator models: SVM, Random Forest, FCNN, CNN, TCN-LSTM, ResNet, and Inception. On the held-out subject, the Transformer achieved a mean 88.48

Vitiligo diagnosis in routine practice remains largely subjective, leading to inter-observer variability, while many existing computational approaches are limited by dataset heterogeneity and poor generalization. This study proposes an automatic vitiligo detection framework based on two pre-trained deep convolutional neural networks, Inception V3 and ResNet-50, fine-tuned via transfer learning. A dermoscopic dataset of 500 images, derived from clinically confirmed cases and expanded through augmentation (rotation, flipping, brightness variation, and zooming), was used to improve robustness and reduce overfitting. All images underwent standardized preprocessing including resizing, normalization, RGB-to-HSV conversion, and histogram-based enhancement to emphasize depigmented regions. The models were evaluated using accuracy, sensitivity, specificity, precision, F1-score, and area under the ROC curve (AUC). Inception V3 achieved 92.7% accuracy, 91.5% sensitivity, 93.8% specificity, 90.2% precision, an F1-score of 90.8%, and an AUC of 0.927, consistently outperforming ResNet-50 across all metrics. Confusion-matrix analysis revealed remaining challenges in early-stage and low-contrast lesions and sensitivity to image artifacts. Despite constraints related to dataset size and anatomical diversity, the findings demonstrate that transfer learning with Inception V3 on carefully preprocessed and augmented dermoscopic images enables reliable, objective vitiligo classification and offers a promising tool to support dermatologists in clinical and teledermatology settings. This research establishes that deep learning, particularly through transfer learning on optimized datasets, offers a promising path for augmenting clinical dermatology by enabling more objective, accurate, and accessible vitiligo diagnosis.

This study aimed to assess the diagnostic value of non-enhanced CT radiomics in preoperatively differentiating early-stage (T1-T2) from locally advanced (T3-T4) colon cancer, addressing the limitations of conventional empirical staging. A retrospective analysis was conducted on 170 patients with surgically confirmed primary colon cancer who underwent non-enhanced CT scans within 1 week before surgery. Three-dimensional segmentation of colonic tumors was performed on the non-enhanced images, followed by automated extraction of radiomic features. Feature selection was executed using the minimum redundancy maximum relevance (mRMR) algorithm, and key features associated with cancer stage were identified using the least absolute shrinkage and selection operator logistic regression. The performance of the radiomics model was compared with conventional T-staging by radiologists. The cohort comprised 170 patients with an average age of 61.69 ± 13.22 years, 43.3% of whom were female, and 75 (44.1%) presented with early-stage disease. Eight radiomic features from non-enhanced imaging were ultimately included. The radiomics model achieved an area under the curve (AUC) of 0.85 (95% confidence interval: 0.78-0.92) in the training set and 0.84 (95% confidence interval: 0.74-0.95) in the test set, with corresponding accuracies of 0.70 and 0.78, sensitivities of 0.87 and 0.87, and specificities of 0.69 and 0.71, respectively. Additionally, in the training set, the radiomics model (AUC = 0.85) significantly outperformed empirical T-staging by radiologists (AUC = 0.71, P < .009). A similar trend was observed in the test set, where the radiomics model (AUC = 0.85) surpassed empirical T-staging (AUC = 0.76), although this difference was not statistically significant (P = .27). Non-enhanced CT radiomics demonstrated superior performance over conventional radiologists' T-staging in distinguishing early from advanced colon cancer stages.

Ophthalmic diseases are a leading cause of vision loss, with retinal damage being irreversible. Retinal blood vessels are vital for diagnosing eye conditions, as even subtle changes in their structure can signal underlying issues. Retinal vessel segmentation is key for early detection and treatment of eye diseases. Traditionally, ophthalmologists manually segmented vessels, a time-consuming process based on clinical and geometric features. However, deep learning advancements have led to automated methods with impressive results. This systematic review, following PRISMA guidelines, examines 79 studies on deep learning-based retinal vessel segmentation published between 2020 and 2024 from four databases: Web of Science, Scopus, IEEE Xplore, and PubMed. The review focuses on datasets, segmentation models, evaluation metrics, and emerging trends. U-Net and Transformer architectures have shown success, with U-Net's encoder-decoder structure preserving details and Transformers capturing global context through self-attention mechanisms. Despite their effectiveness, challenges remain, suggesting future research should explore hybrid models combining U-Net, Transformers, and GANs to improve segmentation accuracy. This review offers a comprehensive look at the current landscape and future directions in retinal vessel segmentation.

Addiction is a chronic relapsing brain disease associated with substantial individual and societal burden. It includes complex phyco-physiological conditions and can be recognized by compulsive and harmful behaviors towards a substance or activity, despite negative consequences or “bio-psycho-social-spiritual” disorder. This paper aims to comprehensively review previous research which applied machine learning to electroencephalography (EEG) signals for automatic detection of addiction. A systematic search was done on 200 relevant papers published between 2018 - 2023. The previous studies applied a variety of machine learning algorithms including support vector machines, neural networks, logistic regressions and brain networks to classify addicted individuals against controls based on resting state, cue-reactivity paradigms and neurofeedback tasks. The features utilized spectral power, functional and effective connectivity, graph theoretical measures and event-related potentials. The output accuracy often exceeded 95% across multiple substances including alcohol, nicotine, cannabis and opioids. In this review, the challenges and open questions around data quality, model interpretation and transition to clinical settings were also discussed. In overall, automatic EEG analysis shows significant potential as an objective and accessible tool for addiction diagnosis, treatment monitoring and relapse prevention that projects worthy of continued refinement and validation.

The synthesis of silver nanoparticles (AgNPs) using traditional physical and chemical methods often involves toxic reagents, high energy consumption, and poor biocompatibility, making them unsuitable for many biomedical applications. Moreover, existing green synthesis approaches frequently lack control over nanoparticle size, shape, and stability, limiting their reproducibility and scalability. To address these limitations, this study employed a green synthesis route using Azadirachta indica (Neem) leaf extract as a natural reducing and stabilizing agent. Silver nitrate (AgNO₃) solutions of varying concentrations (1 mM, 5 mM, and 10 mM) were reacted with the Neem extract under ambient conditions. UV–Visible spectroscopy confirmed the formation of AgNPs with a characteristic surface plasmon resonance peak at 402 nm. Scanning Electron Microscopy (SEM) showed that the 5 mM AgNO₃ concentration produced the most desirable morphology, uniformly spherical nanoparticles with an average size of 98 nm. Energy Dispersive X-ray (EDX) analysis further confirmed the presence of pure elemental silver with no silver compounds. Moreover, antibacterial testing, conducted against Total Coliform bacteria and Propionibacterium acnes, revealed that the synthesized AgNPs, particularly in powder form, effectively inhibited bacterial growth over extended incubation periods. In conclusion, this study demonstrates that Neem-mediated synthesis is a viable, sustainable, and efficient approach for producing biologically active silver nanoparticles.

The application of Transcranial Magnetic Stimulation (TMS) with Electroencephalography (EEG) provides a powerful framework for investigating brain function and neural dynamics. While most TMS-EEG studies have focused on TMS-evoked potentials (TEPs) to examine immediate cortical effects, limited research has explored the influence of TMS on the auditory cortex, particularly regarding the persistence of its effects over time. This study investigates the impact of TMS on Auditory Evoked Potentials (AEPs), focusing on responses measured 1- and 2-seconds poststimulation. Single-pulse TMS was administered with a 4 second interstimulus interval (ISI), concurrently with auditory stimuli presented at a 1 -second ISI. AEPs were analyzed using both time-domain averaging and phase synchronization stability methods. Results showed minimal differences in the averaged N100 and P200 components between TMS and sham conditions at 2 seconds post-stimulation (ANOVA, $p=0.08)$. In contrast, phase stability analysis revealed significantly higher synchronization in the N100 component 1 second after TMS $(p =0.01)$, along with increased stability in the P200 and P300 components compared to the sham condition. These findings emphasize the value of advanced analytical techniques in detecting subtle, time-sensitive neural changes induced by TMS. By revealing the temporal profile of TMS-induced modulation in auditory processing, this study contributes to the refinement of non-invasive brain stimulation techniques for diagnostic and therapeutic applications.

OBJECTIVES:Contaminated apparatus and surgical tools pose serious health risks. For such purpose, disinfection chambers are employed. However, these systems rely on mercury-based UV lamps which comes with various drawbacks. These limitations have driven interest in Ultraviolet-C Light Emitting Diode (UV-C LED) technology as a safer and more efficient alternative. However, existing studies have not thoroughly explored the impact of varying intensities of pulse width modulation (PWM) on disinfection efficacy. METHODS:To addess this, the present study designed and tested a LED-based disinfection chamber by employing 4-W 275 nm Surface Mount Device (SMD) LEDs against frequently isolated bacteria. By following prior approach, irradiation time was alternated at 30-s intervals and antibacterial efficacy was assessed through various parameters. Additionally, scanning electron microscopy (SEM) was performed to examine the morphological changes. RESULTS:Results indicated that the reduction was significantly influenced (p<0.05) with varying PWM levels (60-100 %), achieving 2.05-log10 and 1.54-log10 inactivation against Escherichia coli and Staphylococcus aureus, respectively, upon exposure to 51.24 mJ/cm2 under maximum exposure settings. Moreover, complete cellular damage leading to bleb protrusion and cell-leakage confirmed the disruption of bacterial DNA. CONCLUSIONS:In conclusion, UV-LEDs show great potential for disinfection, with efficiency influenced by PWM and dosage.

The integration of Transcranial Magnetic Stimulation (TMS) with Electroencephalography (EEG) offers a robust framework for investigating brain function and underlying neural mechanisms. While most TMS-EEG studies have focused on TMS-evoked potentials (TEPs) to examine the immediate cortical effects of TMS, limited attention has been given to its influence on the auditory cortex, particularly in terms of how long the modulation persists. This study aims to analyze the effects of TMS on Auditory Evoked Potentials (AEPs), with a focus on changes observed 1 and 2 seconds after stimulation. Single-pulse TMS was applied with a 4-second interstimulus interval (ISI) and presented simultaneously with auditory stimuli delivered at a 1-second ISI. The influence of TMS on AEPs was assessed using time-domain averaging and phase synchronization stability analysis. Results showed minimal differences in the averaged N100 and P200 components between TMS and sham conditions at 2 seconds post-stimulation with ANOVA p=0.08. However, phase stability analysis revealed significantly higher synchronization in the N100 wave 1 second after TMS ANOVA p=0.01, followed by increased stability in the P200 and P300 components compared to the sham condition. These findings highlight the importance of advanced analytical techniques in detecting subtle and transient neural changes induced by TMS. By providing insights into the temporal characteristics of TMS-modulated auditory processing, this study contributes to the development of more targeted diagnostic and therapeutic applications involving non-invasive brain stimulation.