Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disorder affecting motor neurons, resulting in neuromuscular weakness and paralysis. Electromyography (EMG) is of vital importance for the detection of ALS. In this paper, a refined mixture of experts is proposed that automatically discriminates ALS patients from non-ALS cases using clinical EMG signals from the N2001 EMGLAB open-access dataset. The architecture consists of a 1D convolutional neural network, a temporal convolutional network and a spectrogram-based CNN to collectively learn localised temporal, long-range temporal and spectral features from EMG activity. A gating mechanism dynamically weights expert contributions and performs significantly better than equal-weight fusion. Training with focal loss and exponential moving average stabilisation addresses class imbalance and improves convergence. The proposed approach reached an AUROC of 0.9992, an F1-score of 0.9903 and a balanced accuracy of 0.9905, demonstrating strong discriminative performance and potential for real-time clinical applications.
Skin cancer is among the most common malignant tumors worldwide, and early detection is essential to improve patient survival and recovery. Conventional AI-based diagnostic approaches rely on centralized data, which raises security and privacy concerns. This paper presents an implementation of federated learning that addresses these challenges by enabling collaborative model training among simulated distributed client nodes while ensuring that raw patient data remains confidential. A convolutional neural network architecture, ResNet50, is employed at client nodes, with preprocessing steps including image augmentation, contrast enhancement, and lesion segmentation to improve feature extraction on the International Skin Imaging Collaboration (ISIC) dataset. Multiple aggregation algorithms are implemented at the central server, along with both vertical and horizontal federated learning settings. Their performance is evaluated using metrics such as precision, accuracy, recall, and F1-score. In the horizontal federated learning setting, the FedNova aggregation approach outperformed other methods, achieving an accuracy of 76.45%, F1-score of 0.73, and a recall of 0.72, demonstrating enhanced overall classification performance among the evaluated aggregation algorithms and stable convergence across clients. These results highlight the impact of federated learning settings and aggregation strategies on overall skin cancer detection performance.
In minimally-invasive surgery, consulting preoperative images such as MRI or CT scans often diverts surgeons' attention from the operative field, leading to interruptions and increasing operative time. To overcome this, we propose InteractOR, a mixed-reality system that overlays preoperative images onto endoscopic video and enables surgeons to interact with these images using their surgical instruments. InteractOR integrates (i) real-time laparoscopic instrument segmentation using TernausNet-16 trained on the CholecInstanceSeg dataset, (ii) a mixed-reality interface that overlays preoperative images onto endoscopic video and (iii) an interaction technique that enables surgeons to control user interfaces using their instruments. An evaluation of the segmentation model is conducted, as well as a usability evaluation of the system. These results demonstrate the feasibility of the approach and its potential for interaction with preoperative images.
Accurate classification of motor imagery (MI)-based electroencephalogram (EEG) signals is often challenged by signal non-stationarity, subject-specific variability, and privacy concerns associated with sharing raw neural data. To address these challenges, this study proposes a hybrid Markov chain-spatial statistical (MCSS) machine learning framework for privacy-preserving MI-EEG classification. Spatially filtered EEG signals were discretised into six symbolic amplitude states representing progressively increasing signal intensities. These symbolic sequences were modelled as stochastic processes, from which transition probability matrices (TPMs) were constructed as the primary feature representation. This TPM-based abstraction provides a non-invertible and privacy-friendly feature space, substantially reducing the ability to reconstruct the original neural waveform. To enhance discriminative performance, the Markov features were combined with seven statistical descriptors. The framework was evaluated using three machine learning algorithms on the brain-computer interface (BCI) Competition III Datasets IVa and IVb. The proposed MCSS + support vector machine framework achieved consistently high classification accuracy (>98%) across all subjects, demonstrating strong robustness and cross-subject stability. Privacy robustness was further validated through empirical threat-model evaluation, where membership inference attacks remained close to chance level, and feature inversion attacks showed low reconstruction similarity. Overall, the framework provides an accurate, computationally efficient, and privacy-preserving solution for scalable BCI development.
This study presents a practical, IMU-based control framework for 2-DoF prosthetic wrists, designed to explore the feasibility of supporting functional wrist control in daily tasks. The system employs movement synergies inherent to the upper limb and offers three controller configurations: pronation-supination (PS), radial-ulnar deviation (RUD) and a combined PS-RUD mode. Ten healthy participants, simulating upper-limb amputees, performed four representative activities of daily living in a virtual environment, during which both kinematic performance metrics and user-perceived effort were recorded. Task-specific controllers led to significant reductions in compensatory limb movements, and overall, the combined PS-RUD controller was preferred by users regardless of task completion time. Participants reported lower physical and cognitive effort, underscoring the intuitiveness and usability of the suggested controller. These preliminary findings suggest the potential of a user-centred control strategy for prosthetic wrists that could reduce training burden and facilitate seamless integration into everyday functional use.
Acute lymphoblastic leukaemia (ALL), a common form of cancer, remains a life-threatening condition that affects individuals worldwide, including both adults and children. The prognosis is significantly worse when the disease is identified at an advanced stage, as the chances of successful treatment diminish. Diagnosis typically requires invasive and costly procedures, which can delay timely intervention. Images from peripheral blood smears (PBS) have been useful for the preliminary screening of ALL in probable cases. However, due to the nonspecific presentation of the disease, interpreting these images poses significant challenges, increasing the risk of misdiagnosis. We have proposed an approach that uses 20,000 PBS images to precisely enable the early diagnosis of ALL and its distinct subtypes (benign, malignant-early Pro-B, malignant-Pro-B, and malignant-Pre-B) in order to overcome these problems. Our proposed system uses colour normalized PBS images using Vahadane method along with the feature extraction capabilities of deep neural networks combined with a stacking ensemble learning approach that integrates various machine learning algorithms for classification. This system is capable of distinguishing ALL cases from haematogones and accurately identifying its subtypes. By streamlining the diagnostic process, the platform aims to reduce the effort and time required by clinicians and patients. Our proposed model highlight the system's effectiveness, with a comparative analysis of established machine learning algorithms showcasing its superior performance. The proposed model achieved exceptional accuracy (99.95%), recall (99.95%), precision (99.95%) and F1-score (99.95%) for the early detection of ALL and its variants, indicating its potential as a proof-of-concept system, while requiring further validation before clinical deployment.
Accurate segmentation of the left atrial appendage (LAA) is essential for device occlusion planning in atrial fibrillation patients who cannot receive anticoagulation. Yet 3D echocardiography suffers from low signal-to-noise ratio, anisotropy and marked morphological variability, increasing overfitting risk and reliance on operator-dependent post-processing. Existing U-Net variants capture local detail but often miss long-range dependencies, while transformers improve context at the cost of boundary precision; semi-automated pipelines still require experts. We present a fully automated, AERO-optimised DAT-DAD U-Net with SE-augmented skip fusion. Deformable attention transformers (DAT) provide content-adaptive global context, while a dual attention with deformable convolution (DAD) block refines rims and addresses shape irregularities; spatial-channel squeeze-and-excitation improves multi-scale fusion. Hyperparameters are selected by AERO, a surrogate- and multi-fidelity-driven optimiser balancing exploration and exploitation under limited data in practice. Validation used a 22-patient 3D echocardiography cohort from King's College Hospital. Volumes were reformatted into axial 2D slices, trained with on-the-fly anatomy-preserving augmentation and evaluated using strict patient-wise splits. The model achieved Dice 0.8925 ± 0.0144, IoU 0.8026 ± 0.0156 and HD95 9.14 ± 1.96 mm. Ablations confirmed additive gains from DAT, DAD and SE, with faster convergence and a lower error floor, supporting operator-light, time-sensitive LAA workflows.
Emerging mobile health and STEM engagement platforms are increasingly recognised as catalysts for improving public understanding of biomedical technologies and supporting digitally enabled models of healthcare delivery. The Bionics Bus (https://bionicsbus.org/) is a novel mobile laboratory designed to democratise access to robotics, artificial intelligence, prosthetics and digital diagnostics by delivering hands-on experiences directly to schools, communities and healthcare settings. This study provides the first transatlantic comparison of public attitudes toward the Bionics Bus within the United Kingdom and the United States of America (USA), two countries with contrasting healthcare systems and digital transformation priorities. Using a mixed-methods cross-sectional design, 78 participants completed a survey assessing technology perceptions, healthcare accessibility, enthusiasm for STEM and views on the usefulness and potential of the Bionics Bus. Quantitative analysis demonstrated high levels of excitement and strong recognition of the relevance of technology in healthcare across both countries, with USA respondents more likely to report geographical barriers to accessing healthcare. Thematic analysis revealed shared optimism for the Bus as a tool for education, accessibility and community-based innovation. These findings highlight the potential of mobile biomedical platforms to support digital inclusivity, strengthen science capital and enhance public engagement with technology-enabled healthcare.
Tinnitus is a common auditory symptom - sound perceived without an external source. Chronic tinnitus affects about 65 million adults in the EU, often impairing quality of life through distress, sleep problems, anxiety, and depression. Yet, management remains inconsistent across Europe due to patient heterogeneity and lack of standardised care. Health literacy is increasingly recognised as key to empowering patients in self-management and reducing burden. The Erasmus+ project TinWise addresses this by creating a gamified health literacy platform for tinnitus patients and educating healthcare professionals on digital self-help strategies. Building on Tin-TRAC, which developed an open-access e-learning platform, TinWise integrates gamification to enhance motivation, engagement, and knowledge retention. The project applies the OPHELIA framework to optimise literacy strategies and follows the ASPIRE learning construct for co-creating content. TinWise will deliver interactive, co-created digital games to promote self-help literacy, an AI-powered chatbot for real-time guidance, and a patient-linking service to enhance community support. It also aims to train healthcare professionals using reusable learning objects, supporting the integration of digital tools into tinnitus care. Through modern technology and inclusive design, TinWise supports European priorities in digital education, patient empowerment, and healthcare innovation, ultimately improving tinnitus care and outcomes across Europe.
To evaluate the impact of digital interventions used to support care transitions on health outcomes and hospital costs compared with usual clinical care for patients with chronic diseases, we conducted a systematic review and meta-analysis. A systematic search of Cochrane, ACM, Embase, MEDLINE, and PsycINFO was performed from database inception until 6 October 2023. Meta-analysis was conducted for hospitalisation outcomes, mortality rates, and cost-effectiveness, while quality of life, mobility and qualitative outcomes were summarised narratively. Following full-text screening, 34 studies were eligible for review. After excluding eight studies due to a high risk of bias, the final analysis included 25 studies comprising a total of 29,724 patients. The meta-analysis revealed a marginal but statistically significant reduction in hospitalisation rates (SMD = -0.07, 95% CI -0.13 to -0.01, p = 0.02) and a statistically significant reduction in mortality (OR = 0.71, 95% CI 0.58 to 0.85, p = 0.0003), favouring digital interventions. Digitally supported transitional care interventions suggest a potential trend toward supporting quality of life, although evidence was not conclusive. Future research should focus on digitalising established care transition models and evaluating technology's added value for effectiveness and user satisfaction.
Electroencephalogram (EEG) signals are critical for diagnosing neurological disorders like depression but require substantial storage, posing challenges for telemedicine. This paper proposes a novel lossy compression method integrating two-dimensional zigzag and spiral rearrangement, wavelet transform (bior4.4), and adaptive filtering to minimise storage while preserving depression recognition accuracy. Using the Hospital University Sains Malaysia (HUSM) dataset (34 depressed, 30 healthy subjects), EEG signals were converted into 2D matrices, compressed with four wavelet encoders (SPIHT, STW, EZW, LVL-MMC), and reconstructed. A feedforward artificial neural network (ANN) with 80 features (relative wavelet energy and entropy) classified depressed and healthy individuals. The combination of zigzag and STW methods achieved a compression ratio (CR) of 89.30, a percentage root-mean-square distortion (PRD) of 0.23, and a peak signal-to-noise ratio (PSNR) of 58.80 dB, with depression recognition accuracy of 90.2%. Classification was performed on both original and reconstructed signals to evaluate the effect of compression on diagnostic performance. Notably, compressed signals occasionally outperformed original signals, likely due to noise reduction. This method offers a robust solution for efficient EEG storage and accurate depression diagnosis in resource-constrained settings.
A substantial amount of literature has been published on epileptic seizures. However, adequate evidence is still lacking to demonstrate that utilising explainable artificial intelligence for epileptic seizures can ensure an individual's safety. Furthermore, there is a need to define the fundamental challenges and opportunities present in the current state-of-the-art solutions and guide efforts towards responsible artificial intelligence. To identify fundamental challenges and opportunities in the existing state-of-the-art solutions available for explainable artificial intelligence-based epileptic seizure onset early warning: towards responsible artificial intelligence. The PRISMA checklist was utilised to develop this report. Papers were extracted from original articles and prior conference studies published in reputable databases such as PubMed, IEEE Xplore, ScienceDirect, Scopus and Google Scholar from January 2019 to 17 November 2024. Rayyan's online platform was used to identify duplicates, inclusions and exclusions of papers. This systematic review protocol was registered with the PROSPERO database. The included papers were assessed based on Microsoft's Responsible Artificial Intelligence template. The Responsible AI Impact Assessment Template, Principle 3 (transparency and explainability), determined a high-risk rating. A total of 26 studies are included based on the established inclusion and exclusion criteria. This study investigated 14.29% of responsible artificial intelligence principles applied in at least one paper with a high-risk rate. The results indicate that to transform researched solutions into practical applications, epileptic monitoring applications should be tested within the eight principles set by Microsoft. The black box explanation lacks insight into the deep internal features and operational methods, suggesting that further investigation is necessary. Systematic Review Registration ID: CRD42024544.
Mobile-based applications are increasingly used to support patient self-management and medication adherence. Despite advantages, there is limited research on their essential functional and non-functional requirements. This study aimed to identify functional and non-functional requirements of mobile-based medication management applications. This scoping review was conducted in 2025 by searching Scopus, PubMed, ProQuest, Web of Science databases, Google Scholar, and Persian-language databases, including Scientific Information Database (SID) and Magiran. Studies published in English and Persian between 2010 and 2024 were screened for eligibility. Relevant studies which met the inclusion criteria were selected, reviewed, and analysed descriptively. A total of 38 articles were included. Functional requirements included medication information, recommendations, alerts, reminders, medication search, data sharing with healthcare providers, drug interaction information, and side effect reporting. Non-functional requirements consisted of ease of use, scalability, synchronisation, accessibility, compatibility, security, and reliability. Defining functional and non-functional requirements is essential for designing effective medication management applications. The findings provided actionable insights for developers and healthcare stakeholders to build user-centred, secure, and reliable mobile-based solutions, ultimately improving patient outcomes in medication adherence and self-management.
This paper aims to demonstrate the validity of the use of a cylindrical approximation for human fingers and a parallelepiped approximation of the palm, from a modelling perspective, for the determination of the mass of the hand. The goal is to provide an intuitive way of determining the length and radius of missing segments from partial hand amputees based on palm dimensions and determine the corresponding mass based on the previous two approximations. In-vivo hand mass measurements were taken from 23 able-bodied participants using Archimedes' water displacement method to verify the geometric approximations used. Furthermore, an anthropometric investigation on how segment length and radius change with respect to palm dimensions was undertaken and the estimates found, which can then be used in supporting the design of prosthetic segments in a personalised context.
This paper contributes to a relatively simple and efficient referencing method for a decision-maker needs to impose or adjust the policies in fighting against a pandemic crisis. A Bayesian network (BN) model is built via referring to a pre-condition evaluated by canonical correlation analysis to diagnose the likelihood of a severe pandemic by exploring the relationship between Hofstede's national culture dimensions and mortality rate (MR) as well as some other policy-related index. Sample data retrieved from 90 countries and areas up to 13, July 2020, were used for this worldwide cross-sectional study during the COVID-19 pandemic. Another four countries and areas were employed to examine the accuracy of our model. Results also suggest that a strict isolation policy (PI) is possibly not an efficient way to contain the COVID-19 pandemic, especially for those countries or areas with loose ties between individuals. The probability of a high MR, derived from three Hofstede cultural dimensions using the BN model, serves as a reliable indicator of policy implementation effectiveness. Additionally, the individualism versus collectivism (IDV) dimension, which reflects societal group integration (with lower IDV values indicating stronger cohesion), constitutes a key metric for policy assessment. The findings are intended to provide a foundation for developing strategies that strengthen preparedness and resilience in addressing comparable public health crises in the future.
Wearable electronics have been on the rise for personal monitoring in healthcare and sports, allowing real-time tracking. However, developing flexible, conductive, biocompatible, and suitable for continuous, long-term use (bio)electrodes remains a challenge. In this sense, carbon materials offer a promising solution due to their excellent electrical conductivity, mechanical strength, and natural biocompatibility. Moreover, they are cost-effective, modifiable, and align well with environmentally friendly practices. This work presents a simple and sustainable fabrication method for custom-formulated carbon black-chitosan (CB-CH) ink, enhanced with multi-walled carbon nanotubes (MWCNTs). The formulation avoids toxic chemicals, high energy input, and lengthy processing, supporting a greener approach. The resulting ink enables the fabrication of free-standing and textile-based electrodes with high conductivity, mechanical durability, and application-dependent biocompatibility, supporting extended use for CB-CH and short- to medium-term wearable applications (≤24 h) when MWCNTs are incorporated. Their performance was validated through real-time monitoring of electrophysiological signals such as electrocardiograms and electromyograms, showing signal quality comparable to conventional silver electrodes while overcoming gel dehydration and skin irritation. Overall, this work offers a scalable, cost-effective, and eco-friendly pathway for producing multifunctional electrodes, paving the way for next-generation wearable sensing platforms in clinical diagnostics, rehabilitation therapies, and athletic performance monitoring.
Background and Aims:Haemophilia is a congenital bleeding disorder requiring extensive self-management. Mobile applications can support these processes. This study aimed to design and evaluate a mobile application to facilitate self-management in persons with haemophilia. Methods:The study was conducted in two phases. First, a literature review was performed in PubMed, Web of Science, and Scopus, and searches were made in Google Play, Apple Store, and the websites of the World Federation of Hemophilia and Hemophilia Federation of America to identify information-educational needs and necessary capabilities for app design. A questionnaire based on these needs was then provided to physicians and patients for approval. In the second phase, the application prototype was designed according to the approved information-educational needs and necessary capabilities and evaluated for usability using the standard Questionnaire for User Interface Satisfaction with 17 patients. Data were analysed using SPSS 20. Results:Sixty-two information-educational needs and capabilities with scores above 50% were included in the design. The designed application can provide the conditions for recording demographic information, medical history, treatment information, bleeding information, and injuries and accidents and provide the education on self-management for patients. Moreover, it has the capabilities to set various reminders (taking medication, rehabilitation-physical activities, and appointments with the therapist), set the treatment plan, send reports to the therapist, communicate with other persons with haemophilia and therapists, and record notes. Patients evaluated the application's usability at a "good" level. Conclusion:The haemophilia self-management app can enhance patients' quality of life by addressing their educational needs and supporting self-management.
Objective:Ensuring critical data integrity practices are often missing from the confidence of decision-making in the recorded clinical data. Thus, we developed an application to report clinical data errors to provide effective healthcare IT services and maintain clinical data integrity. Materials and Methods:The RadOnc E-Variance web-based application was designed and developed to report data inconsistencies and variances noted during clinical operations. The data inconsistency issues can be reported in 3 broad categories: Patient-related, Non-Patient related, and other suggestions/open-ended, allowing uploading of images or document files. The contents presented on the web are dynamically created in the application and are stored in a database. In the back-end, data analytics and detailed assessments are available on the dashboard. Users reporting variances remain anonymous to ensure reporting freely and frequently. Results:Once implemented clinically, we have found significant improvements with fewer errors, ensured patient treatment safety, and improved workflows in our state-wide clinics. Conclusion:The RadOnc E-Variance application effectively increases patient safety by ensuring data integrity and minimises risks in the EMR applications at multifaceted enterprise levels of hospital environments by reducing the inconsistencies in data recording and errors recorded in electronic health records.
Employment is a crucial part of recovery for individuals with severe mental illness. Individual placement and support (IPS) is the gold standard for vocational rehabilitation, yet IPS reaches only a fraction of who could benefit. Large language models (LLMs) have been proposed as potential tools for vocational guidance, but their utility for vulnerable populations is unknown. We conducted an analysis of LLM-generated job recommendations for individuals with schizophrenia spectrum disorders, and for a matched control cohort without psychiatric diagnoses. We used discharge summaries from 450 patients with a primary diagnosis of schizophrenia spectrum disorder and 50 control cases in the MIMIC-IV database, fitting three independent job recommendations per case with Gemini 2.0 Flash and Claude Sonnet 4. Recommendations were summarised as a frequency and LLM-automated content analysis was used to analyse reasoning patterns, workplace accommodations, and alignment with supported employment principles. Both, Gemini and Claude, showed little diversity and strong bias toward entry-level roles. In the schizophrenia cohort, Gemini mostly recommended data entry and other clerical jobs while Claude produced a similarly narrow pattern with the majority suggesting library-related. The controls revealed comparable clustering, with Gemini defaulting to clerical work and medical secretary roles, and Claude to customer service. There was limited diversity in the role settings, which almost uniformly suggested flexible schedules and minimal social interaction. Nor was there diversity in how roles were tailored to patient strengths, qualifications, or prior experience; instead, demographic stereotypes such as age-based framing, gendered role allocation, and assumptions about language skills often shaped the recommendations. Based on our data and procedures, preliminary evidence does not support immediate deployment of LLMs for job recommendations for the tested population; further evaluation is needed after integrating human oversight and bias-mitigation steps.
Diabetes has become a critical global health concern, particularly in regions where access to diagnostic facilities is limited. In this work, we propose a hybrid framework that combines extreme gradient boosting (XGBoost) and deep neural networks (DNNs) for early-stage diabetes detection, using soft voting to generate the final ensemble predictions. The proposed framework was evaluated on two datasets: the widely used Diabetes UCI dataset and a newly collected dataset from Nepal. The ensemble method achieved 99% accuracy (ACC) with an area under the curve (AUC) of 1.00 on the Diabetes UCI dataset, and 91% ACC with a 0.96 AUC on the Nepal diabetes dataset, demonstrating its strong generalisability across distinct populations. Compared to individual models, the hybrid approach offered increased stability and a lower rate of false negatives, which is particularly important in clinical contexts. These findings highlight the potential of hybrid machine learning-deep learning models as cost-effective, scalable and generalisable decision-support tools for diabetes risk assessment.
