Publications

Early diagnosis of rare lysosomal storage disorders, such as Gaucher disease (GD), acid sphingomyelinase deficiency (ASMD), and Saposin C deficiency, is critical but often delayed. Electronic health records (EHRs) provide opportunities for screening, but class imbalance challenges computational methods. The real-world performance of one-class classification (OCC) machine learning (ML) models versus clinician-derived expert rules remains poorly understood. We compare a clinician-curated expert rule system, based on phenotypes like organomegaly and cytopenias, against eight OCC models on 572 patients (13 GD, 8 ASMD, 3 Saposin C, 548 controls) from a United Arab Emirates health system. The evaluation used imbalance-aware metrics including recall, precision, F1- and F2-scores, balanced accuracy, the area under the precision–recall curve, and the number needed to screen (NNS). The expert rule system demonstrated high, stable recall (76.9%–100.0%) with a clinically feasible workload (NNS: 3.0–7.7). In a matched comparison, rules consistently outperformed the best OCC models (F1-score: GD 0.749 vs 0.626; ASMD 0.751 vs 0.580; Saposin 0.631 vs 0.360). ML models exhibited severe instability as rarity increased, with catastrophic failures (0% precision) on the ultra-rare Saposin cohort. Across all three disorders in this EHR cohort, expert rules delivered more stable recall and a more clinically feasible screening workload (NNS) than one-class ML as disease rarity increased. This stability proved critical, as the primary clinical objective in rare disease screening is to ensure no cases are missed (i.e., to maximize recall). The expert rules provided this dependable, high-recall screen, whereas selected OCC models that raised precision did so at an unacceptable cost of reduced sensitivity and stability. Given the ultra-rare Saposin cohort, OCC methods were unreliable in our dataset, whereas expert rules preserved sensitivity. Overall, our results motivate a hybrid strategy in which expert rules set a reliable recall floor and ML is used only as a secondary tool to re-rank or triage within the flagged subset.

Understanding learners' cognitive and affective states underpins adaptive educational systems and effective teaching. Although research links nonverbal cues to internal states, no framework calibrates them to evidence. We present the Nonverbal Syntax Framework, drawn from a systematic review of 908 studies and 17,043 cue-state mappings (Turaev et al., 2026). The framework addresses three challenges: terminological fragmentation (behaviors described inconsistently), evidence heterogeneity (single observations to replicated findings), and state ambiguity (similar patterns indicating multiple states). Normalization consolidated 5,537 state labels into 2,010 canonical states (63.7%) and 11,521 cues into 6,434 normalized cues (44.2%) across nine behavioral channels. Dual-evidence assessment separately evaluates Component Evidence (coverage of cues and states) and Relationship Evidence (independent studies per cue-state link). 52% of "Very High" relationships rest on one paper, so separation enables calibrated rather than overconfident inference from preliminary findings. The framework's four levels comprise a Cue Vocabulary of 6,434 indicators classified as observable/instrumental; State Clusters linking 2,010 states to indicative cues; State Profiles with multimodal behavioral signatures and actionable specifications; and Discriminative Analysis distinguishing 1,215 confusable state pairs. We identify 480 actionable R1-R4 relationships (three or more independent papers), the replicated core of six decades of research, covering 35.5% of mappings across 47 key learning states and 111 distinct indicators. The remaining 91.5% (9,653 single-paper findings) form exploratory hypotheses for replication. The framework gives researchers an empirical foundation for identifying gaps, practitioners evidence-based tools for state inference, and technologists validated features for multimodal detection.

Modeling microscopic cerebrovascular networks is essential for understanding cerebral blood flow and oxygen transport. High-resolution imaging modalities, such as Optical Coherence Tomography (OCT) and Two-Photon Microscopy (TPM), are widely used to capture microvascular structure and topology. Although TPM angiography generally provides better localization and image quality than OCT, its use is impractical in studies involving fluorescent dye leakage. Here, we exploit generative adversarial learning to produce high-quality TPM angiographies from OCT vascular stacks. We investigate the use of 2D and 3D cycle generative adversarial networks (CycleGANs) trained on unpaired image samples. We evaluate the generated TPM vascular structures based on image similarity and signal-to-noise ratio. Additionally, we evaluated the generated vascular structures after applying vessel segmentation and extracting their 3D topological models. Our results demonstrate that the 2D adversarial learning model outperforms the 3D model in terms of image quality. However, our statistical comparisons of vascular network features show the 3D model's consistent superiority in generating vascular structures. Our work provides a complementary approach to enhance vascular analysis when only OCT imaging is available.

The impact of Artificial Intelligence (AI) in healthcare is undeniable, aiding physicians in diagnosing diseases. This study aims to synthesize the literature to examine the progress of AI in diagnosing Tuberculosis (TB), one of the deadliest diseases in the world. The review also provides a taxonomy for AI-based studies for TB diagnosis. A scoping review approach was adopted using the PRISMA-Scoping guidelines, focusing on types of studies, focus area, algorithms/models, and key results. Relevant articles published from 2013 to 2022 were sought using PubMed, Web of Science, and Scopus, resulting in 199 articles included in the review. The use of AI, especially deep learning, has increased since 2016, particularly for diagnosing TB using chest X-Rays (CXR). Most studies focused on diagnosing TB using CXR, Computed Tomography, biomarkers, sputum smear, and drug resistance and recovery. Convolutional Neural Networks (CNNs) and their variants were commonly used in deep learning, while Support Vector Machine (SVM), Decision Trees, and ensemble algorithms performed well in machine learning. CNN outperformed other variants for CXRs (accuracy from 80 to 100%) probably due to their ability to handle high-dimensional data and extract features. In contrast, simpler algorithms like Naive Bayes underperformed compared to other algorithms (accuracy range: 79–89%), showing their limitations in dealing with complex TB diagnostic data. The use of AI approaches, namely machine and deep learning are expected to increase in the future, hence promising a rapid and cost-effective (and potentially sustainable) alternative solution for efficient TB diagnosis.

Background: Neuroimaging segmentation is increasingly important for diagnosing and planning treatments for neurological diseases. Manual segmentation is time-consuming, apart from being prone to human error and variability. Transformers are a promising deep learning approach for automated medical image segmentation. Objective: This scoping review will synthesize current literature and assess the use of various transformer models for neuroimaging segmentation. Methods: A systematic search in major databases, including Scopus, IEEE Xplore, PubMed, and ACM Digital Library, was carried out for studies applying transformers to neuroimaging segmentation problems from 2019 through 2023. The inclusion criteria allow only for peer-reviewed journal papers and conference papers focused on transformer-based segmentation of human brain imaging data. Excluded are the studies dealing with nonneuroimaging data or raw brain signals and electroencephalogram data. Data extraction was performed to identify key study details, including image modalities, datasets, neurological conditions, transformer models, and evaluation metrics. Results were synthesized using a narrative approach. Results: Of the 1246 publications identified, 67 (5.38%) met the inclusion criteria. Half of all included studies were published in 2022, and more than two-thirds used transformers for segmenting brain tumors. The most common imaging modality was magnetic resonance imaging (n=59, 88.06%), while the most frequently used dataset was brain tumor segmentation dataset (n=39, 58.21%). 3D transformer models (n=42, 62.69%) were more prevalent than their 2D counterparts. The most developed were those of hybrid convolutional neural network-transformer architectures (n=57, 85.07%), where the vision transformer is the most frequently used type of transformer (n=37, 55.22%). The most frequent evaluation metric was the Dice score (n=63, 94.03%). Studies generally reported increased segmentation accuracy and the ability to model both local and global features in brain images. Conclusions: This review represents the recent increase in the adoption of transformers for neuroimaging segmentation, particularly for brain tumor detection. Currently, hybrid convolutional neural network-transformer architectures achieve state-of-the-art performances on benchmark datasets over standalone models. Nevertheless, their applicability remains highly limited by high computational costs and potential overfitting on small datasets. The heavy reliance of the field on the brain tumor segmentation dataset hints at the use of a more diverse set of datasets to validate the performances of models on a variety of neurological diseases. Further research is needed to define the optimal transformer architectures and training methods for clinical applications. Continuing development may make transformers the state-of-the-art for fast, accurate, and reliable brain magnetic resonance imaging segmentation, which could lead to improved clinical tools for diagnosing and evaluating neurological disorders.

The rapid adoption of online learning in higher education has resulted in significant cybersecurity challenges. As educational institutions increasingly rely on digital platforms, they are facing cyber threats that can compromise sensitive data and disrupt operations. This systematic literature review explores the integration of artificial intelligence (AI) into traditional methods to address cybersecurity risks in online higher education. The review integrates a qualitative synthesis of relevant literature and a quantitative meta-analysis using PRISMA guidelines, ensuring comprehensive insights into the integration process. The most prevalent cybersecurity threats are examined, and the effectiveness of AI-based and conventional approaches in mitigating these challenges is compared. Additionally, the most effective AI techniques in cybersecurity solutions are analyzed, and their performance is compared across different contexts. Furthermore, the review considers the key ethical and technical considerations associated with integrating AI into traditional cybersecurity methods. The findings reveal that while AI-based techniques offer promising solutions for threat detection, authentication, and privacy preservation, their successful implementation requires careful consideration of data privacy, fairness, transparency, and robustness. The importance of interdisciplinary collaboration, continuous monitoring of AI models—by automated systems and humans—and the need for comprehensive guidelines to ensure responsible and ethical use of AI in cybersecurity are highlighted. The findings of this review provide actionable insights for educational institutions, educators, and students, helping to facilitate the development of secure and resilient online learning environments. The identified ethical and technical considerations can serve as a foundation for the responsible integration of AI into cybersecurity within the online higher-education sector.

Outlier detection plays a vital role in data analysis by identifying data points that differ from the expected patterns. In healthcare, this is especially important, as spotting these anomalies can help detect unusual patient conditions, errors in data, or even early signs of fraud. However, even though this is crucial, little research has compared how well different outlier detection methods work in healthcare. This study aims to fill that gap by testing 37 different outlier detection algorithms on 28 healthcare datasets, covering everything from common conditions like diabetes to rare diseases like thyroid disorders. We measured their performance using metrics like ROC-AUC, Average Precision, and F1 Score, along with practical considerations like training time, prediction speed, and memory usage. Our results show that no single model works best across all scenarios. However, methods like HBOS, COPOD, and IForest achieved average ROC-AUC scores of 0.85, 0.83, and 0.82, respectively, performing well in various situations. More complex models like AnoGAN and DeepSVDD achieved ROC-AUC scores up to 0.90 on specific datasets, particularly excelling in datasets with higher dimensionality. This variety highlights the need to choose outlier detection methods based on the specific healthcare context, balancing accuracy and efficiency to meet different clinical and practical needs.This paper provides a solid foundation for future research and development of outlier detection systems in healthcare.

Pompe disease (OMIM #232300), a rare genetic disorder, leads to glycogen buildup in the body due to an enzyme deficiency, particularly harming the heart and muscles. Infantile-onset Pompe disease (IOPD) requires urgent treatment to prevent mortality, but the unavailability of these methods often delays diagnosis. Our study aims to streamline IOPD diagnosis in the UAE using electronic health records (EHRs) for faster, more accurate detection and timely treatment initiation. This study utilized electronic health records from the Abu Dhabi Healthcare Company (SEHA) healthcare network in the UAE to develop an expert rule-based screening approach operationalized through a dashboard. The study encompassed six diagnosed IOPD patients and screened 93,365 subjects. Expert rules were formulated to identify potential high-risk IOPD patients based on their age, particular symptoms, and creatine kinase levels. The proposed approach was evaluated using accuracy, sensitivity, and specificity. The proposed approach accurately identified five true positives, one false negative, and four false positive IOPD cases. The false negative case involved a patient with both Pompe disease and congenital heart disease. The focus on CHD led to the overlooking of Pompe disease, exacerbated by no measurement of creatine kinase. The false positive cases were diagnosed with Mitochondrial DNA depletion syndrome 12-A (SLC25A4 gene), Immunodeficiency-71 (ARPC1B mutation), Niemann–Pick disease type C (NPC1 gene mutation leading to frameshift), and Group B Streptococcus meningitis. The proposed approach of integrating expert rules with a dashboard facilitated efficient data visualization and automated patient screening, which aids in the early detection of Pompe disease. Future studies are encouraged to investigate the application of machine learning methodologies to enhance further the precision and efficiency of identifying patients with IOPD.

In recent years, online reviews have become increasingly important in promoting various products and services. Unfortunately, writing deceptive reviews has also become a common practice to promote one's own business or tarnish the reputation of competitors. As a result, identifying fake reviews has become an intense and ongoing area of research. This paper proposes a node embedding approach to detect online fake reviews. The approach involves extracting features from the input data to create a distance matrix, which is then used to construct a Graph. From the graph, we extract graph nodes and use the Node2Vec biased random walk algorithm to create a model. We retrieve node embeddings from the Node2Vec model using Word2Vec and use different classifiers to classify the nodes. We trained and evaluated the machine learning models on the Deceptive Opinion Spam Corpus and YelpChi datasets and achieved superior results compared to prior work for detecting fake reviews, with SVM using the Hamming distance achieving 98.44% accuracy, 98.44% precision, 98.44% recall, and 98.44% F1-score. Furthermore, we explored different methods for explaining our proposed methods using explainable AI, demonstrating the interpretability of our approach. Our proposed node embedding approach shows promising results for detecting fake reviews and offers a transparent and interpretable solution for the problem.

This study proposes an innovative approach to the early identification of at-risk students in higher education settings by augmenting traditional machine learning classifiers with topological features and node embeddings derived from graph-based representations of student data from programming courses at the United Arab Emirates University. Utilizing a dataset comprising student demographic characteristics and course performance results, we construct an adjacency matrix using cosine and Canberra distance metrics, which is then thresholded to form a binary, unweighted graph. Topological features and node embeddings are extracted from this graph, representing the relationships and interactions between students. The extracted features are combined with the original dataset to train various machine learning classifiers, aiming to enhance their predictive accuracy. Experimental results demonstrate a significant improvement in prediction performance, with an increase of almost 10% in accuracy when node embeddings are incorporated. The most notable improvement was observed when using a Multi-Layer Perceptron classifier with the original dataset supplemented with both topological features and node embeddings, achieving 93.5% accuracy. Our findings highlight the potential of graph-based methods to enrich the feature set used by machine learning models, thereby enhancing their capacity to identify at-risk students early. Future work will focus on refining the feature extraction process, exploring other graph methods, and incorporating additional types of data. This study lays the foundation for more comprehensive and effective early-warning systems in higher education, aiming to enhance student support services and improve overall educational outcomes.

Body language refers to the unspoken communication conveyed through human body actions like body movements and postures, limb gestures, and facial and other bodily expressions. It acts as a transparent medium, exposing an individual’s emotions, attitudes, true thoughts, intentions, and physical and mental health states. A person may express pain using hand movements or other bodily cues, their facial expressions potentially offering insights into the intensity of the pain. Additionally, various diseases and pains can induce abnormalities in body movements, postures, and expressions, signaling distress or discomfort. Therefore, investigating the cause-effect relationships between diseases/pains and patients’ abnormal body language holds significant relevance, promising to enhance our understanding and management of these conditions. This importance has been reflected in numerous healthcare and artificial intelligence (AI) research articles. AI studies investigate this and related topics by detecting, recognizing, and analyzing patients’ abnormal activities and body actions using machine-learning techniques. However, most AI studies do not consider comprehensive domain knowledge that describes a complete and accurate list of patients’ abnormal actions caused by a disease or pain. Though these results appear consistent and stable from an AI outlook, they fall short when viewed through the prism of healthcare, primarily because the limited domain knowledge incorporated in the AI studies makes the findings partially incomplete. To overcome these drawbacks, this paper comprehensively reviews healthcare and medical studies centered on patients’ body language from an AI outlook. It presents a thorough descriptive and exploratory analysis of the findings, yielding a more accurate and comprehensive understanding of the causational connections between diseases and abnormal body actions and the strength of the evidence supporting these connections. The analysis enables us to define “disease-to-abnormality” and “abnormality-to-disease” mappings that result in building exhaustive and accurate lists of abnormal body actions induced by diseases and pains as well as lists of diseases and pains causing particular abnormal body actions. The generation of these lists is assisted by the concepts of “correlation strength index” and “strongly correlated selection” defined in this paper. The paper’s results have significant implications for developing machine learning systems that can more accurately analyze patients’ physical and mental health states, correctly identify external signs and symptoms of diseases, and effectively monitor health conditions.

Computational quantification of magnetic resonance imaging (MRI) response from neurovascular structures is used to investigate potential biomarkers for different types of cerebrovascular deteriorations at the microscopic scale. Simulation-based MRI requires fully resolved microvascular structures, with geometric and physiological parameters, from tissue volumes captured using microscopic imaging modalities, e.g., optical coherence tomography (OCT). The preparation of such input models hinders large cohort studies and requires extensive manual effort. Here, we propose using 3D neural networks as an alternative learning-based solution over MRI simulation schemes. We trained state-of-the-art 3D neural networks to predict the spin echo (SE) MRI response from OCT microvascular volumes. By validating against simulated signals, our result demonstrates that the 3D ResNet-based regression network achieves a high accuracy to predict MRI signals with an average mean square error (MSE) <1%, R2 of 82.8% and explained variance score of 82.9%.

Cardiovascular diseases (CVDs) are prevalent disorders affecting the heart or blood arteries. Early disease detection significantly enhances survival prospects, thus emphasizing the necessity for accurate prediction methods. Emerging technologies, such as machine learning (ML), present promising avenues for more precise prediction of CVDs. However, a critical challenge lies in developing models that not only ensure optimal predictive performance but also conform to well-established domain knowledge, thereby enhancing their credibility. Single classifiers often fall short due to issues like overfitting and bias. In response, this study proposes a domain knowledge-based feature selection integrated with a stacking ensemble classifier. The Framingham Heart Study, UCI Heart Disease and UAE retrospective cohort study datasets were utilized for training and evaluation of the ML algorithms. The results indicate that the proposed domain knowledge-based feature selection performs on par with frequently adopted feature selection techniques. Moreover, the proposed stacked ensemble, in conjunction with domain knowledge-based feature selection, achieved the highest metrics with 89.66% accuracy, and 89.16% F1-score on the Framingham dataset. Similarly, the proposed method achieved an F1-score of 85.26% and 96.23% on the UCI Heart Disease and UAE datasets. Furthermore, this study employs explainable AI techniques to illuminate the decision-making process of the predictive models. Thus, the study establishes that domain knowledge-based feature selection promotes the credibility of ML models without compromising predictive performance.

Body language is a nonverbal communication process consisting of movements, postures, gestures, and expressions of the body or body parts. Body language expresses human feelings, thoughts, and intentions. It also reveals physical and psychological health conditions: abnormal activities inform peoples' health conditions, facial expressions indicate their emotional states and abnormal body actions convey specific diseases' external signs and symptoms. We can observe the importance of studying the body language of people with health conditions through many reports in literature written by healthcare (medical) and artificial intelligence researchers. This paper comprehensively reviews artificial intelligence-based articles that have studied patients' body language. We also conduct different descriptive and exploratory examinations of the findings using data analysis techniques, which provide more authentic domain knowledge of abnormal activities, abnormal body actions, and more precise analysis of methodologies used in machine learning tasks for studying these abnormalities. The paper's results are essential for developing intelligent automated systems that accurately evaluate patients' physical and psychological conditions, precisely identify external signs and symptoms of diseases, and adequately monitor patients' health conditions.

Green space is any green infrastructure consisting of vegetation. Green space is linked with improving mental and physical health, providing opportunities for social interactions and physical activities, and aiding the environment. The quality of green space refers to the condition of the green space. Past machine learning-based studies have emphasized that littering, lack of maintenance, and dirtiness negatively impact the perceived quality of green space. These methods assess green spaces and their qualities without considering the human perception of green spaces. Domain-based methods, on the other hand, are labour-intensive, time-consuming, and challenging to apply to large-scale areas. This research proposes to build, evaluate, and deploy a machine learning methodology for assessing the quality of green space at a human-perception level using transfer learning on pre-trained models. The results indicated that the developed models achieved high scores across six performance metrics: accuracy, precision, recall, F1-score, Cohen’s Kappa, and Average ROC-AUC. Moreover, the models were evaluated for their file size and inference time to ensure practical implementation and usage. The research also implemented Grad-CAM as means of evaluating the learning performance of the models using heat maps. The best-performing model, ResNet50, achieved 98.98% accuracy, 98.98% precision, 98.98% recall, 99.00% F1-score, a Cohen’s Kappa score of 0.98, and an Average ROC-AUC of 1.00. The ResNet50 model has a relatively moderate file size and was the second quickest to predict. Grad-CAM visualizations show that ResNet50 can precisely identify areas most important for its learning. Finally, the ResNet50 model was deployed on the Streamlit cloud-based platform as an interactive web application.

Cardiovascular diseases (CVDs) are the leading cause of death globally, with millions of lives lost yearly. CVDs are a group of disorders of the heart and blood vessels. Although there are no exact causes of CVDs, there are risk factors associated that increase the likelihood of getting CVDs. Clustering and association rule mining are among the methods used for pattern discovery. However, not much research has been proposed to compare clustering and association algorithms regarding risk factors of CVDs. Hence, this study presents a comparative analysis of clustering and association on the risk factors of CVDs to assess which factors are significant. The Framingham Heart Study dataset was used for clustering and association rule mining. The clustering results using three clusters show that older age, high BMI, and high systolic blood pressure are the significant risk factors. Smoking and hypertension are among the risk factors contributing to angina and heart attack based on association analysis with minimum support, minimum confidence, and maximum items of 25%, 60% and 4, respectively. This study successfully adopted clustering and association for pattern discovery to assess the most critical risk factors of CVDs.