We analyze the global innovation landscape of point-of-care (POC) blood-testing devices and evaluate their technological principles, diagnostic coverage, usability, regulatory pathways, deployment feasibility, and data/AI integration, in the context of precision healthcare and predictive medicine. We identified and reviewed 86 commercially available POC blood-testing devices worldwide. Devices were compared across core technology platforms, diagnostic panel breadth, usability attributes, regulatory classifications, approval timelines, and a custom deployment-capability index incorporating training burden, maintenance requirements, and integration readiness. Data/AI features and geographic innovation patterns were also analyzed. Electrochemical biosensors were the most prevalent technology (29.1%) and showed strong enrichment for glucose testing. Spectroscopic and microfluidic systems remained inaccessible or under-represented due to higher cost and workflow complexity. Regulatory approvals concentrated in moderate-risk classes, while approval times increased sharply with risk category. The review highlights (1) strong associations between sensing modalities and clinical applications driven by underlying physicochemical constraints, with downstream implications for healthcare disparities, accessibility barriers, and care gaps, and (2) the emerging integration of AI-enabled analytics with POC systems to enhance diagnostic precision, scalability, and real-time clinical decision-making. Understanding these patterns enables more data-driven, clinical translation of POC diagnostics and prognostics, particularly in the context of noisy and limited patient data. These quantitative findings help guide next-generation diagnostic design, scalable detection, and equitable deployment of AI-enabled POC systems in digital, precision, and predictive medicine. Our framework supports wide-scale, routine access to advanced detection systems in underserved and low-resource healthcare settings, helping bridge critical gaps in patient-centered care.
Optical Coherence Tomography (OCT) is a critical imaging modality for the early detection of retinal diseases; however, existing automated analysis approaches remain constrained by speckle noise, heterogeneous lesion morphologies, and subtle inter-class textural variations. This study introduces DenseWave-OCT, a hybrid classification framework that integrates deep semantic representations extracted from a pretrained DenseNet-121 backbone with multi-scale wavelet scattering coefficients designed to capture fine-grained structural and textural patterns. The two complementary feature sets are fused at the feature level into a unified descriptor, which is subsequently passed to a fully connected classifier. The proposed framework was evaluated on two publicly available benchmark datasets OCT2017 and Retinal OCT-C8 and benchmarked against established convolutional neural network baselines. DenseWave-OCT achieved classification accuracies of 99.48% and 98.29% on the two datasets, respectively, outperforming all evaluated baselines. These results demonstrate that combining deep convolutional features with wavelet-based textural representations yields a more discriminative and robust descriptor for OCT-based retinal disease classification, offering a promising direction for clinically reliable automated diagnostic tools.
Gliomas, a common type of brain tumor, present a complex diagnostic challenge owing to their heterogeneity and the need for accurate grading to guide treatment decisions. This research explores the application of ensemble Machine Learning (ML) techniques to enhance the accuracy of glioma grading using the “Glioma Grading Clinical and Mutation Features” dataset. We introduce a rigorous methodology that includes the derivation of a novel feature, “Number_of_mutations”, to better capture the aggregate impact of genetic alterations. Our proposed framework involves comprehensive data preprocessing to prevent data leakage, extensive base model training with 5-fold cross-validation, and the implementation of stacking and voting ensemble architectures. The results indicate that a stacking model utilizing Gradient Boosting (GB) and Logistic Regression (LR) as base learners, with a meta-learner aggregation, achieves a high testing accuracy of 94.05%, precision of 88.46%, recall of 97.18%, F1-score of 93.24%, Matthews Correlation Coefficient (MCC) of 0.88, and an Area Under the Curve (AUC) score of 0.96. The model also demonstrates excellent calibration with a Mean Square Error (MSE) of 0.06 and Root Mean Square Error (RMSE) of 0.24, outperforming prior studies on the same dataset. Feature contributions are further illuminated through Explainable AI (XAI) techniques, separating global feature importance via SHAP (SHapley Additive exPlanations) and local interpretability via LIME (Local Interpretable Model-agnostic Explanations). These highlight IDH1 and IDH2 as significant positive indicators and TP53 and PTEN as negative indicators. This study provides a robust, reproducible, and interpretable framework for glioma grading, specifically targeting the improvement of diagnostic accuracy and supporting informed clinical decision-making.
Detecting early-stage sarcoma can be challenging, given that it rarely shows clear symptoms. Further, for pathologists, accurately diagnosing and classifying sarcoma from medical images can also be difficult. To address these issues, researchers have used many AI models to predict and classify diseases from imaging data, including machine learning, deep learning, Transformers, the GPT family, large language models (LLMs), and one of the latest advances, multimodal LLMs. To contextualise this research, this study provides a comprehensive scoping review of recent studies on sarcoma diagnosis and classification. We located relevant articles in notable journals, books, and conference proceedings and summarised each study’s objectives, datasets, methods used, results, and limitations. Our objective was to identify recent AI-based techniques in sarcoma diagnosis that rely on different imaging modalities, including MRIs, CT scans, X-rays, and pathology slides. We have found that CNNs, SVMs, k-NN, decision trees, and their variants have been widely used for image classification. Also, Transformer and hybrid models are increasingly being used to improve performance, with high levels of accuracy reported in several studies. However, generalisability remains a key challenge given the limited number of studies that externally validate their model’s performance or use multi-institution data. The limitations of these studies are also discussed along with potentially useful future directions of research to assist experts. A dedicated section summarises early applications of GPT/LLMs to diagnosing sarcoma, highlighting their preliminary nature and the need for human oversight, reliability assessment, and validation. The article concludes by outlining the challenges of applying AI to solve problems in the current climate, noting issues such as data privacy, regulatory uncertainty, dataset transparency, and the need for external validation. In terms of diagnosing sarcoma from radiology and pathology imaging, there is also a need for calibrated reporting.
Hospital discharge planning plays a central role in patient flow, quality of care, and hospital efficiency. However, discharge assessment is often fragmented across teams and performed under tight time constraints. Existing machine-learning models trained on electronic health record (EHR) data have been applied to forecast near-term discharge events, but they primarily estimate timing or probability and provide limited structured clinical interpretability. To address these limitations, we propose an LLM-based multi-agent clinical decision support system for discharge readiness assessment. The architecture decomposes discharge reasoning into five clinically distinct specialist agents, each producing structured, evidence-linked claims grounded in specific EHR fields. A deterministic, schema-based claim verification layer cross-checks the asserted field values against the patient record prior to arbitration, and a reliability-aware arbiter synthesizes verified assessments into a final readiness determination. This design replaces a single scalar score with multidimensional, auditable reasoning while explicitly separating generative reasoning from value-level data checking. Evaluated on a retrospective MIMIC-IV cohort across four LLM backbones, binary prediction of proxy safe early discharge under the SAFE_D24 definition ranged from 0.490 to 0.620 F1, with PPV from 0.521 to 0.628. GPT-4.1 achieved the strongest performance (0.620 F1, 75.8% specificity, 61.3% sensitivity). Ablation analysis demonstrates measurable performance gains from deterministic verification. Beyond discrimination metrics, the system produces traceable barrier identification, claim-level verification statistics, and configurable human-review triggers, supporting traceable and auditable discharge assessment with explicit safety and operational trade-offs.
