Mesopotamian Journal of Artificial Intelligence in Healthcare

The COVID-19 outbreak has shown how urgently good models are needed to understand and project the dissemination of infectious diseases. Graph theory\'s excellent basis for presenting and analyzing complex networks helps one to grasp COVID-19 transmission dynamics. This work investigates the simulation and evaluation of the COVID-19 spread using multiple graph theory approaches. We utilize algorithms, including centrality assessments, community detection, and epidemic spreading models, to identify significant transmission channels and viable intervention sites; we also study the usage of network-building strategies to show relationships between individuals and communities. By incorporating real-world data with graph-based models, we demonstrate how these approaches could increase the accuracy of infection estimates and direct public health strategies. Moreover, the paper discusses the advantages and limitations of graph theory approaches within the framework of pandemic modeling, therefore leading the next direction of investigation to improve disease outbreak reactions. This work highlights the important contribution of graph theory to increase our understanding of COVID-19 dissemination and offers a scientific foundation for handling associated problems in upcoming public health-catastrophes

Lung cancer is a leading cause of cancer-related mortality, often diagnosed at advanced stages. This systematic review explores AI applications in lung cancer diagnosis, focusing on medical imaging, pathology, and genetic analysis. A systematic literature review methodology was employed, analyzing studies from databases such as PubMed, IEEE Xplore, and Scopus (2018–2024). Findings indicate that AI-powered diagnostic models, particularly deep learning techniques, outperform conventional methods in accuracy, sensitivity, and early detection capabilities. However, integration into clinical practice presents challenges, including data privacy concerns, model biases, and regulatory limitations. This review highlights the potential of AI in lung cancer screening andprovides insights into future research directions.

Parkinson's disease (PD) is a movement disorder characterized by motor dysfunction commonly bradyphemia, tremor, rigidity, akinesia, or slowness of movement. Noting that motor states can fluctuate in PD the primary aim of this current paper was to differentiate multiple states using wearable sensors in the patients and detection of this PD based on deep learning (CNN). Methodology: In this paper, the researchers recorded the signals of the accelerometer and gyroscope fixed on the wrist of PD in their regular daily functioning after using this dataset collection. The deep learning architecture developed was to optimize a CNN for analyzing the sensor data for motor status like rest, tremor, and dyskinesia. Results and Conclusion: The availability of high accuracy in defining different degrees of motor state was achieved through the deep learning algorithm hence providing a high option in monitoring pd symptoms. These outcomes also show and suggest the possibility and practicality of the wearable sensor for daily activity monitoring for the identification of PD. It can also provide appropriate information to patients other than doctors regarding their sickness which gives them a platform to effectively handle Parkinson’s disease patients. The error has been determined to be within the vicinity of 98%.

In the present day, humans are confronted with a variety of diseases as a result of their lifestyle and the current environmental conditions. Therefore, it is crucial to identify and predict these diseases in their early phases in order to prevent their severe manifestations. Manually identifying maladies is a challenging task for physicians on a regular basis. Predicting chronic illnesses is the aim of this article. This goal is applicable through a state-of-the-art approach to classification correctly identifies people with chronic illnesses. Predicting maladies is also a difficult endeavor. Therefore, disease prediction is significantly influenced by data mining. To get data, a collection of disease symptoms, the individual's lifestyle, and information regarding medical consultations are considered in this general disease prediction. In conclusion, this paper analyzes that model with a variety of algorithms, including (Naiva Bayes) and (RF-Random Forest).

This study presents an AI-driven multi-objective optimization approach using the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to enhance hospital operational efficiency in Syria. Using real-world data from the Tishreen University Hospital over a 60-day period, the research addresses three conflicting objectives: minimizing average patient waiting time, reducing daily operational costs, and maximizing the number of patients treated. Six key operational variables were selected to build the optimization model, including bed availability, physician count, and daily admissions. The NSGA-II algorithm successfully generated a set of Pareto-optimal solutions, each reflecting different trade-offs among the objectives. Statistical analysis and visualizations confirmed the complexity and nonlinearity of hospital operations, showing that increases in resources or costs do not always lead to improved outcomes. The results offer decision-makers a range of efficient operational configurations tailored to various institutional priorities. This model provides a valuable decision-support tool, especially in resource-constrained healthcare environments like Syria. Future research will focus on integrating real-time data, expanding operational variables, and validating the model across different institutions to support broader policy implementation and operational standardization.

Palliative care seeks to improve the quality of life of patients with life-threatening illnesses by addressing their physical, emotional, and psychological needs. However, global challenges such as workforce shortages, limited access to specialized care, and inconsistent care quality demand innovative solutions. Advances in artificial intelligence (AI)-powered robotics offer transformative potential to overcome these barriers and strengthen palliative care delivery. This study explores how AI-driven robotic technologies support palliative care through applications in symptom monitoring, clinical decision-making, emotional companionship, and personalized care planning. It reviews cutting-edge robotic systems, including assistive, companion, diagnostic, nursing, procedural, service, and rehabilitation robots. Enabled by machine learning, deep learning, natural language processing, and computer vision breakthroughs, these systems help monitor vital signs, manage symptoms, plan end-of-life care, deliver medication, alleviate pain, and support mobility through robotic exoskeletons. They also assist patients with daily activities and offer respite to caregivers. Despite their promise, AI-powered robotics face significant challenges, including ethical concerns, algorithmic bias, data privacy risks, cultural resistance, and resource limitations. When integrated ethically and thoughtfully, AI-powered robotics can extend the reach of palliative services, support human caregivers, and enhance outcomes for patients and families. Collaboration among healthcare professionals, AI researchers, engineers, and policymakers is crucial to ensure that robotic technologies remain patient-centered, safe, and accessible. By merging technological innovation with compassionate care, AI and robotics can redefine the future of palliative care globally.

This study developed an XGBoost-based risk assessment model to enhance early tumor detection among Egyptian patients, addressing the challenges of late diagnosis and limited healthcare resources. Utilizing a retrospective dataset of 178 patients, the model incorporated demographic, clinical, and biochemical variables, including AFP levels, viral hepatitis status (HBV/HCV), and liver function markers. The model demonstrated strong predictive performance, achieving an accuracy of 0.833, precision of 0.846, and an AUC of 0.86, though recall remained moderate (0.688), indicating room for improvement in identifying high-risk cases. Feature importance analysis highlighted AFP levels and hepatitis status as the most influential predictors, aligning with Egypt’s high prevalence of liver cancer. The findings underscore the potential of AI-driven tools for early cancer screening in resource-limited settings, offering a scalable and cost-effective solution. However, future work should focus on expanding datasets, optimizing recall, and validating the model across diverse populations to ensure clinical applicability. This research contributes to the growing integration of AI in oncology, providing a framework for tailored risk stratification in high-burden regions

This paper explores the potential of Artificial Intelligence (AI), including natural language processing (NLP), computer vision, and robotics, to transform nursing workflow in endoscopy units. AI technologies offer numerous advantages, including predictive analytics for efficient scheduling and resource allocation, real-time decision support during procedures, and enhanced diagnostic accuracy through image analysis. Additionally, AI simplifies administrative tasks such as patient scheduling and medical documentation, improving operational efficiency and patient safety. Despite these benefits, challenges related to data privacy, algorithm accuracy, and ethical considerations must be addressed. Effective AI integration requires interdisciplinary collaboration, continuous education for nurses, and robust governance frameworks to ensure responsible and ethical use. Addressing these challenges will enable AI to profoundly impact nursing workflows, significantly enhance patient care, and reshape the operational future of endoscopy units.

Cardiovascular disease (CVD) continues to be the predominant cause of mortality globally, underscoring the critical necessity for prompt and precise diagnostic techniques. This paper introduces an innovative machine learning framework for categorizing heartbeat sounds into four classifications—normal, murmur, additional heart sound, and artifact—utilizing audio recordings from the PhysioNet/CinC Challenge 2016 dataset. The methodology employs Mel-Spectrograms and Mel-Frequency Cepstral Coefficients (MFCCs) for feature extraction, converting raw heart sound data into comprehensive time-frequency representations. A Convolutional Neural Network (CNN) is utilized for classification, with its hyperparameters refined by the recently developed Frilled Lizard Optimization (FLO) method, a bio-inspired metaheuristic that emulates the hunting and climbing behaviors of the frilled lizard. The suggested method illustrates the capability of integrating deep learning with sophisticated optimization techniques to improve the diagnostic precision of cardiac auscultation, particularly in non-clinical and remote environments. This can facilitate the development of intelligent, scalable, and proactive intervention methods in preventative cardiovascular healthcare.

Type 2 diabetes mellitus (T2DM) is a growing global health concern requiring early detection strategies. This study applies a Random Forest machine learning model to predict diabetes progression using a structured clinical dataset of 100 patients. The dataset includes demographic, physiological, and biochemical variables such as age, BMI, blood pressure, glucose levels, and lipid profiles. After preprocessing and training, the model achieved strong performance metrics: accuracy of 0.80, precision of 0.84, recall of 0.94, and an AUC of 0.88. Feature importance analysis revealed that systolic blood pressure, fasting glucose, and BMI are the most critical predictors. These findings are consistent with previous literature and demonstrate the model’s clinical relevance in identifying high-risk patients. The model’s ability to generalize across cross-validation folds highlights its robustness and potential integration into decision-support systems for proactive diabetes management. The study concludes that machine learning models, when properly validated, can offer significant improvements in chronic disease prediction and personalized care delivery.

This study presents of Decision Tree classifiers for predictive modeling in medicine, focusing on model depth optimization, pruning techniques, and performance evaluation. On the basis of a synthetic healthcare dataset containing over 55,000 records, each with features such as age, gender, blood type, bill amount, and medical condition, we investigate the impact of varying tree depth from 1 to 5 on predictive accuracy, interpretability, and generalizability. Shallow models have strong transparency but poor classification strength, and deep models obtain stronger interactions but suffer from overfitting. With pruning, we find a balance between model simplicity and precision and yield strong classifiers for practice. With comparative examination through confusion matrices, feature importance graphs, and accuracy measures, the research presents thorough details about how Decision Tree configurations affect healthcare prediction tasks. The findings emphasize the need for interpretable artificial intelligence (XAI) methods in medical machine learning, stressing the attractiveness of models that find a balance between interpretability and sound performance. The research contributes a pragmatic approach to model selection and optimization in medical analytics that supports the creation of decision-support systems that are clinically valid, ethical, and regulatory compliant.

Healthcare and robotics dynamic task domains are also known to be both complex and time-varying systems where reasoning about causal links are important for making reliable decisions. But causal structures in these domains are difficult to ascertain with limited data, distribution shifts, and changing environments. Meta-learning (or learning to learn) is a promising approach to combating these two obstacles by learning from experience across tasks to quickly adapt to new task environments. In this paper we provide a new framework that combines ideas from meta-learning and causal discovery, whose objective is to discover the underlying cause effect relationships in sequential, real-world domains. Our method learns a transferable causal discovery model based on simulated tasks from health monitoring and robotic control that can then be used to swiftly estimate causal graphs for new table top environments unseen during training. We test the framework empirically on simulated experiments, involving patient health monitoring and robotic manipulation, where we show that the meta-learned model can recover the causal structure with orders of magnitude fewer samples than standard approaches. We find that we obtain better structural accuracy (in terms of AUC and Structural Hamming Distance) and systemic adaptation in comparison with non-meta-learning baselines even in an environment where the system dynamics change over time. This work contributes to the novelty of interleaving meta-learning with causal discovery in dynamical domains: the knowledge to infer causality across tasks enables the model to generalize to new contexts -- a characteristic common to both healthcare and robotics. We also present comparisons of our method with prior work on synthetic benchmarks and interpret the learned representations. We conclude with discussions about future directions, such as the use of active interventions and real-world testing data, to advance the meta-learning-based Causal Discovery for complex dynamic systems.

This paper introduces BioGPT, a generative transformer-based framework designed to advance personalized genomic medicine and rare disease diagnosis. Unlike conventional models that process either genomic sequences or clinical narratives in isolation, BioGPT employs a cross-modal architecture that effectively fuses both data streams, enabling precise classification and interpretable natural language generation. The model is pre-trained on large-scale genomic and electronic health record datasets and fine-tuned for rare disease tasks. Comprehensive experiments demonstrate BioGPT’s superiority over state-of-the-art biomedical models, including RarePT, BioBERT, and DNABERT, with improvements of up to 10% in F1-score and over 20 BLEU points in justification fluency. Ablation studies highlight the essential contribution of cross-attention mechanisms in enhancing multi-modal synergy. Furthermore, attention-based interpretability techniques show strong alignment with expert clinical markers, ensuring trust and transparency in diagnostic outputs. With sub-second inference times and compatibility with edge deployment strategies, BioGPT proves both effective and deployable in real-world clinical settings. This work establishes BioGPT as a robust, scalable, and explainable AI solution, setting a new benchmark for intelligent diagnostic systems in precision medicine.

This study investigates the application of a machine learning model—specifically the Light Gradient Boosting Machine (LightGBM)—to predict 30-day hospital readmissions using structured electronic health record (EHR) data. Hospital readmissions remain a critical challenge in healthcare systems, often indicating gaps in continuity of care and contributing to higher costs. By leveraging demographic, clinical, and diagnostic variables from 350 anonymized patient records, the model aimed to accurately identify individuals at high risk of readmission. Key features included age, number of previous admissions, length of stay, number of medications, chronic disease status, and gender. Data preprocessing, model training, and evaluation were conducted using Python-based libraries, ensuring both reproducibility and scalability. The model achieved a ROC AUC of 0.89, precision of 0.78, recall of 0.65, and F1 score of 0.71, indicating strong performance and balance between sensitivity and specificity. A confusion matrix analysis confirmed high accuracy in both positive and negative predictions. SHapley Additive exPlanations (SHAP) values were used to enhance interpretability by quantifying the contribution of each variable to the model’s output. Feature importance analysis revealed that age and previous hospitalizations had the greatest impact on prediction, which aligns with prior clinical evidence. The study's results are consistent with existing research and highlight the potential of explainable AI models in healthcare risk prediction. These findings support the integration of machine learning into hospital decision-support systems to enable early intervention strategies, reduce preventable readmissions, and improve overall patient outcomes. Recommendations include adopting predictive models for discharge planning, prioritizing high-risk patient groups, and further validation across broader datasets. The study emphasizes transparency, clinical relevance, and operational feasibility, demonstrating how data-driven tools can be effectively deployed in real-world clinical environments

Bone sarcomas represent aggressive malignancies with complex clinical presentations where traditional prognostic methods often lack precision in predicting patient outcomes. This study developed and validated machine learning models for predicting bone sarcoma patient outcomes using a comprehensive dataset from Memorial Sloan Kettering Cancer Center spanning 2010-2020, comparing multiple algorithms including AdaBoost, Logistic Regression, Ridge Classifier, Quadratic Discriminant Analysis, and Linear Discriminant Analysis. AdaBoost demonstrated superior performance with 0.84 accuracy, 0.875 sensitivity, 0.7955 specificity, 0.8448 precision, 0.8333 negative predictive value, and 0.8596 F1 score, outperforming other algorithms which achieved 0.83 accuracy and 0.8496 F1 score. Statistical analysis confirmed significant differences between classifiers with F-statistic of 21.9130 and p-value less than 0.0001. The study concludes that AdaBoost-based classification provides a reliable framework for bone sarcoma outcome prediction with superior performance, demonstrating potential for clinical integration to support treatment planning and establishing a foundation for precision medicine applications in orthopedic oncology.

Human-Centric Artificial Intelligence (HCAI) is rapidly emerging as a transformative paradigm, shifting the focus of AI development from mere algorithmic optimization to ethical alignment, user trust, and societal integration most notably within the critical domain of healthcare. This review offers a comprehensive examination of the principles, architecture, challenges, and future directions underpinning HCAI, with an emphasis on its applications in health-related contexts. We begin by exploring the conceptual foundations of human-centricity, including core values such as transparency, fairness, autonomy, and privacy, all of which are essential in sensitive environments like clinical decision-making and patient data management. The paper then surveys key enabling technologies such as Explainable AI (XAI), Human-in-the-Loop learning, affective computing, and multi-agent collaboration demonstrating how these approaches operationalize human alignment in real-world systems, especially in personalized healthcare delivery and diagnostic support. Societal implications are critically evaluated, encompassing trust, data sovereignty, algorithmic bias, regulatory compliance, and cross-cultural adaptability, which are particularly pronounced in global health systems. We highlight the limitations of existing benchmarks and propose a multi-metric, user-centered evaluation framework capable of assessing both technical robustness and alignment with human values in healthcare and beyond. Finally, we identify open research challenges and outline a strategic agenda that integrates cognitive science, ethical theory, and participatory design. This review aims to serve as both a foundational reference and a forward-looking roadmap for researchers, developers, and policymakers striving to build AI systems that are not only intelligent, but also responsible, inclusive, and aligned with human dignity, particularly in life-critical domains like healthcare

This study illustrates a data‐driven approach to the segmentation of health insurance policyholders based on K-Means clustering of an open insurance dataset. Key demographic and financial features like age, body mass index (BMI), dependents, annual medical spending, and premium payment were normalized first to ensure comparability. The optimal number of clusters (k = 3) was determined using silhouette analysis, and three clusters were formed: (1) young, low‐cost individuals, (2) middle‐aged medium‐cost individuals, and (3) old, high‐cost individuals. Cluster centroids provide actionable profiles that can be utilized by insurers for target marketing, risk profiling, and development of customized plans. A set of visualizations scatter plots, boxplots, histograms, and bar charts illustrate the separation and within‐distribution nature of these segments. The preprocessing workflow (missing value treatment, encoding of categorical features, and feature scaling) was encoded in a flowchart for reproducibility. Results demonstrate that straightforward-to-implement unsupervised learning techniques can yield interpretable customer segmentations, offering a foundation for more advanced predictive modeling and individualized insurance policies.

Early and adequate identification of patients at risk of cardiac arrest remains a fundamental challenge in medical environments. Within this piece of work, we design and evaluate a hybrid machine-learning pipeline integrating Random Forest (RF) and Support Vector Machine (SVM) classifiers by soft voting to detect cardiac arrest occurrences from a publicly available dataset of 303 patient records with 13 clinical features. After median imputation and data cleaning, continuous features were normalized and the cohort was divided into 70 % training and 30 % test sets. RF scored 97 % with an AUC of 1.00 on its own, and SVM scored at 95 % with an AUC of 0.97. The hybrid ensemble matched RF's perfect AUC but slightly improved recall on positive instances with enhanced sensitivity while maintaining specificity. Permutation-based and Gini-based feature-importance analysis identified major-vessel count, chest-pain type, and peak heart rate as the most important predictors. Distribution plots and heatmaps of correlations further elucidated associations between clinical measures and event status. The proposed hybrid approach is an effective, interpretable way to stratify risk early in high-risk patient populations.

This paper reports a comparative analysis of four supervised machine learning algorithms: RF, SVM (using radial and linear kernels), Logistic Regression, and Multi-Layer Perceptron, for breast cancer recurrence prediction on a carefully curated clinical dataset. The data set, first collected by Royston and Altman and subsequently released on Kaggle, has patient age, menopausal status, tumor size, histological grade, lymph node status, estrogen and progesterone receptor levels, hormone therapy for treatment, recurrence-free survival time, and a binary recurrence outcome. The data set was then divided after the elimination of identifiers and z-score normalization in an 80:20 ratio using stratified sampling. Models were compared based on accuracy, precision, recall, F1-score, and area under the ROC curve, with RF and Logistic Regression having the highest test-set accuracy of 0.703. Feature significance analysis Gini impurity in R F, linear model absolute coefficients, and permutation importance in neural networks all showed lymph node count, survival time, and hormone receptor levels to be significant predictors. Visualized confusion matrices, ROC curves, and correlation heatmaps enhanced interpretability. The results illustrate the potential of explainable machine learning to enhance individualized surveillance and treatment planning in breast cancer care.

Public Health Inspectors (PHIs) serve a critical role in enforcing health and safety regulations, particularly under the growing pressures of climate change. With rising exposure to occupational hazards such as heat waves, air pollution, and vector-borne diseases, PHIs now face escalating stress and burnout. Geographical variability, limited resources, and institutional gaps in training and support further shape their complex risk profile. Despite growing concern, systematic, confidence-based evaluations of these occupational risks—especially tailored to PHIs—remain rare. This study addresses that gap using a Confidence in Network Meta-Analysis (CINeMA)-enhanced framework to assess domain-specific occupational risk profiles of PHIs working in climate-stressed environments. Drawing on an empirically collected dataset of emotional pressure, cognitive fatigue, organizational support, and environmental exposure, we conducted regression analyses and domain-level CINeMA confidence ratings. These included dimensions such as indirectness, imprecision, bias, and heterogeneity. Findings revealed moderate imprecision and within-study bias, with weak model fit suggesting latent variables beyond traditional exposure metrics. Semi-urban PHIs reported the highest climate-related impact scores (CCF mean= 2.91), while PhD-level PHIs showed lower susceptibility. We also propose a novel integration of AI-based topic modeling with CINeMA bias classification to support a future-ready hybrid risk evaluation system. Additionally, we introduce the Novel AI-based modeling Network Meta-Analysis, Adamopoulos–Valamontes Classification and Assessment Model (AV-CA Model)—a structured framework for classifying and assessing environmental, psychosocial, and organizational risks linked to climate impacts in PHI settings. These results support evidence-based OSH policy and training reforms that enhance the resilience of frontline public health systems amid escalating climate challenges.