Artificial intelligence (AI) is increasingly being integrated into healthcare systems and has the potential to improve health outcomes. In Sub-Saharan Africa (SSA), however, concerns remain that AI may either reduce or exacerbate existing health inequities depending on how it is developed, governed, and implemented. This scoping review aimed to map and synthesise the existing evidence on the implications of AI for health equity among marginalised populations in Sub-Saharan Africa. PubMed, Web of Science, Scopus, and selected grey literature sources were searched between February and March 2026. Peer-reviewed and grey literature examining AI applications, governance, or implementation in healthcare involving marginalised populations or health systems within SSA were eligible for inclusion. The review followed the Arksey and O'Malley methodological framework and the PRISMA-ScR reporting guideline. Two reviewers independently screened sources of evidence and extracted data using a standardised charting form, and findings were synthesised thematically. Twenty-three sources of evidence met the inclusion criteria. Two dominant narratives emerged. AI may reinforce existing inequities through digital infrastructure gaps, algorithmic bias, under-representation of African datasets, weak governance, and data colonialism. Conversely, AI has the potential to improve health equity by expanding healthcare access, strengthening disease surveillance, supporting health system planning, and improving access to specialised services. Across the literature, AI's impact consistently depended on equitable infrastructure, inclusive governance, and context-specific implementation. AI has considerable potential to advance health equity in SSA. However, achieving equitable benefits requires investment in digital infrastructure, representative data systems, ethical governance, and inclusive policies. This paper situates AI within the context of health equity in Sub-Saharan Africa.Main findings: Artificial intelligence in African health systems presents both opportunities to expand healthcare access and risks of reinforcing existing inequalities driven by infrastructure gaps, data bias, and governance challenges.Added knowledge: This review provides a focused synthesis of how artificial intelligence specifically affects marginalised populations in Sub-Saharan Africa, highlighting the structural conditions that shape equitable or inequitable outcomes.Global health impact for policy and action: The findings emphasise the need for deliberate policy action on inclusive data systems, digital infrastructure investment, and contextually grounded governance to ensure artificial intelligence advances health equity rather than deepening disparities.
In the field of 3D printing technology in the medical application is becoming more and more widely, especially in orthopaedic surgery, its importance is increasingly prominent. This technology through precise biological tissue engineering, and can be produced with the patient's own tissue matching the artificial bone implants, thus in the bone graft surgery and cosmetic surgery play a key role. With the deepening of the research found that different material and printing method can significantly affect the artificial bone implant biological specificity and clinical effect. In addition, the personalization of 3D printing implants can better adapt to the patient's anatomical structure, improve the success rate of surgery and patient satisfaction. In recent years, 3D printing technology in expanding the application of orthopedic surgery, its potential value also gradually by mining. Therefore the author through the reviews in recent years, 3D biological technology and 3D printing to print and the research status of the artificial bone implants, as well as their application in orthopedic surgery and potential value, in order to provide new ideas for the research of the field.
Recent advances in protein structure determination and prediction, large-scale structural databases, and artificial intelligence have reshaped structure-based drug discovery. Structure-aware artificial intelligence models integrate molecular representation learning with three-dimensional protein information to model interactions, predict complex structures and binding poses, and generate novel molecules. In this review, we follow this paradigm along a continuum from protein-ligand modeling to the de novo design of biomolecular binders. We first outline the molecular and protein representations that render structures computable, together with the growing collection of structural data resources. We then survey state-of-the-art methods across drug-target interaction prediction, protein-ligand complex modeling and docking, de novo molecular generation, and biomolecule design, examining the convergence of docking, structure prediction, and molecular generation within co-folding and diffusion-based frameworks. Despite these advances, prospective experimental validation remains scarce, and persistent limitations such as biased structural coverage, limited and ambiguous negative supervision, fragmented benchmarking, and insufficient mechanistic interpretability continue to constrain real-world utility. Progress in data quality and supervision design, evaluation rigor, and design-relevant interpretability will be essential to translate methodological innovation into practical impact.
Given the pivotal role of imaging in diagnosing urological cancers, artificial intelligence (AI) has emerged as a promising tool to improve diagnostic accuracy and reliability. This study systematically evaluates the diagnostic performance of AI models in radiologic imaging of urological cancers. A systematic search was conducted in four electronic databases up to June 2026 to identify studies that applied AI algorithms for the diagnosis of urological cancers using CT, MRI, or ultrasound. Eligible studies reported diagnostic accuracy metrics for AI models, with clinician comparator data extracted when available. A bivariate random-effects model was used to pool sensitivity, specificity, and AUC values. Subgroup analyses were conducted to examine diagnostic performance across different cancer types and imaging modalities, and to explore potential sources of heterogeneity. Study quality was assessed using the QUADAS-2 tool. A total of 110 studies were included in the meta-analysis. AI models achieved pooled sensitivity and specificity of 0.85 (95% CI: 0.83-0.87) and 0.83 (95% CI: 0.80-0.86), with an AUC of 0.91 (95% CI: 0.88-0.93). Clinicians demonstrated a pooled sensitivity of 0.82 (95% CI: 0.79-0.85) and specificity of 0.68 (95% CI: 0.62-0.73), with an AUC of 0.83 (95% CI: 0.80-0.86). Subgroup analyses indicated that AI models showed overall diagnostic advantages across cancer types and imaging modalities, particularly in specificity, AUC, and diagnostic odds ratios, although clinicians demonstrated higher sensitivity in the prostate cancer and MRI subgroups. AI models demonstrate strong diagnostic performance across various urological cancers and imaging modalities, showing potential as supportive tools in radiological workflows. Further prospective, standardized, and multi-center evaluations are warranted to confirm AI's clinical utility across diverse diagnostic tasks in urological oncology.
Artificial intelligence (AI) has achieved remarkable success in the diagnosis of Alzheimer's disease (AD) in the literature, where many of the models use multi-modal methods including neuroimaging, cerebrospinal fluid, genetics, and cognitive assessment. But clinical adoption of these systems is still limited since most systems are developed in an idealized setting, as cost-effective and specialized diagnostic studies are not universally accessible. We discuss the translation of benchmark performance of AI to real-world dementia care pathways. A practical framework that would be useful for scalable, equitable, and clinically deployable AI-assisted dementia care. In fact, recent advancements in blood-based biomarkers such as plasma phosphorylated tau, glial fibrillary acidic protein, and neurofilament light chain are providing new opportunities for a flexible and minimally invasive diagnosis method. Based on these advances, we propose a clinically grounded AI-assisted cascading model which mirrors real-world workflows via progressive screening, biomarker-guided assessment, selective imaging escalation, and longitudinal prognostic monitoring. We further discuss enabling methods such as sequential decision-making, reinforcement learning, cost-sensitive learning, missing-modality robustness, and explainable AI. Finally, we outline the challenges for data design, for future validation and integration into healthcare systems, and ethical use.
There is a significant intra-provider variability in colonoscopy performance but this is difficult to measure in routine practice. We describe an artificial intelligence tool (AI-CQ) that measures colonoscopy quality via analysis of recorded colonoscopy procedures and compare AI-CQ assessment of quality metrics with manual measurement in a large cohort of colonoscopists. Colonoscopy procedures were performed at one of two endoscopy locations at a single academic medical center. Select analyses were restricted to higher volume screening colonoscopists performing ≥100 screening or surveillance colonoscopies over the 11-month study period. Colonoscopy quality metrics were calculated from recorded colonoscopy videos using the AI-CQ tool and compared to manually calculated metrics (via nurse documentation) including adenoma detection rate (ADR) and withdrawal time (WT). A total of 18,597 colonoscopy procedures performed by 55 unique attendings were recorded with 31 higher volume screening colonoscopists performing 12,456 screening or surveillance colonoscopies (median colonoscopist ADR 43.2%). AI-Insertion time (AI-IT) and AI-WT strongly correlated with manually calculated IT (r=0.60) and WT (r=0.91). AI-polyps per colonoscopy (AI-PPC) was 1.47 (SD ±0.54) and strongly correlated with ADR (0.54) and serrated detection rate (0.67). The AI-CQ accurately measured performance of any polypectomy and cold snare polypectomy with a mean cold snare polypectomy rate of 84.0% (range 63.0-95.3%). The AI-CQ can accurately measure commonly utilized quality metrics using recorded colonoscopy videos. Use of this AI tool provides a novel feasible approach to reliably measuring colonoscopy quality.
暂无摘要（点击查看详情）
This Viewpoint explores how artificial intelligence may mitigate or exacerbate inequities in correctional health systems and outlines policy and implementation considerations necessary to ensure health equity for incarcerated populations.
暂无摘要（点击查看详情）
暂无摘要（点击查看详情）
Previous research has demonstrated changes in neural oscillations associated with varying levels of roughness during active-touch exploration of surfaces. In the present study, we aimed to investigate changes in neural oscillatory activity and softness perception during touch exploration of skin. Two experiments were conducted. Study 1 evaluated active touch of artificial skin samples mounted to a purpose-built touch sensor, whereas Study 2 investigated active stroking of one's own forearm. In both experiments, the substrates were treated with commercially available bar soaps to deliver either a soft or draggy skin feel. Oscillatory brain activity was measured using a 129-channel electroencephalography system. In 31 participants, changes in oscillatory band power were evaluated in relevant frequency bands during touch exploration periods. For the artificial skin study, the soft condition led to lower alpha-band power over bilateral somatosensory cortices, which has previously been proposed as a marker of reduced roughness, compared to the draggy condition. Similar results were obtained during the self-touch paradigm, which additionally led to reduced theta-band changes over the frontal and central-parietal electrodes indicating modulation of activity involved in affective (pleasant) touch. Using a novel and highly controlled experimental approach using a novel artificial skin paradigm and active exploration during self-touch of participants own skin, we were able to demonstrate for the first time the neural correlates associated with softness perception of skin and their impact on brain activity associated with affective touch, thereby advancing our understanding of brain oscillatory activity during active-touch exploration of skin. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
BackgroundWith population aging and an increase in sports injuries, articular cartilage wear has become increasingly severe, significantly impairing patients' quality of life. Cross-shear motion is a common loading pattern in daily joint activities, yet its effects on cartilage and counterpart materials remain insufficiently investigated.ObjectiveTo investigate the wear behavior and underlying mechanisms of articular cartilage under cross-shear motion, and to compare the wear resistance of different artificial joint materials, so as to propose material optimization strategies suitable for this motion pattern.MethodsFresh bovine knee femoral cartilage was used to simulate the human cartilage environment under various cross-shear conditions and loading regimes. Stepwise loading, microhardness testing, cartilage compression deformation measurements, and surface wettability tests were performed to systematically analyze the tribological characteristics and wear mechanisms of each material. The evaluated materials included CoCrMo, ultra-high molecular weight polyethylene, and polyether-ether-ketone (PEEK).ResultsCross-shear motion significantly increased wear in all materials. CoCrMo exhibited the largest increase (ΔK = 230.51 ± 25.67%), while PEEK showed the smallest increase in wear rate (138.37%). No significant linear correlation was found between material hardness and cross-shear wear rate (r = -0.32, p > 0.05). In terms of dynamic wettability, PEEK performed best, with a hysteresis angle 38.46% lower than that of natural cartilage, and its energy dissipation parameters were close to those of natural cartilage, indicating the best cartilage compatibility.ConclusionUnder cross-shear conditions, PEEK demonstrates the best overall wear resistance and biomechanical compatibility, suggesting its potential as a promising material for joint repair. The wear mechanisms revealed in this study provide experimental evidence and reference for performance optimization and clinical selection of artificial joint materials.
There is an emerging understanding of neurodegenerative diseases as complex diseases with a combination of multiple interrelated signaling pathways as opposed to one causative factor. This review examines the idea that neurons do not die in a single event, but through convergence of signals, which outlines the different pathological events such as oxidative stress, mitochondrial dysfunction, excitotoxicity, calcium imbalance, impaired proteostasis and neuroinflammation that interact to determine the fate of neurons. The processes are closely connected by molecular nodes like ROS, NF-κB, and MAPK signaling pathways, Nrf2/Keap1 antioxidant axis, and dysregulated autophagy and endoplasmic reticulum stress responses. The review also discusses the contribution of neuron glia interactions and the impact of microglial activation, astrocyte malfunction and cytokine networks in enhancing neuronal damage in a feedback mechanism. Mechanisms that have been mentioned as the oxidative stress inflammation cycle, mitochondrial damage, ROS feedback, and protein aggregation cellular stress loop are cited to be the major contributors to disease progression. Also, the review mentions new biomarkers, multi-omics methods, and sophisticated research instruments, such as artificial intelligence and organoid models, which can contribute to our knowledge of disease pathogenesis and help diagnose it earlier. On the whole, this review presents a complete paradigm on how to perceive neurodegeneration as a systems-level phenomenon. It combines molecular, cellular, and clinical perspectives and offers the rationale for the need to consider the therapeutic approach to neurodegenerative diseases in a holistic and multi-dimensional manner.
Telemedicine is conventionally modeled as a dyadic clinician-patient encounter, yet a third party-caregiver, community health worker, nurse, or increasingly an artificial intelligence (AI) conversational agent-frequently participates. No principled basis exists for determining when such a third party is a genuine facilitator versus an instrument of one party, rendering cross-study comparison incommensurable and deployment decisions poorly grounded. We conducted a narrative synthesis of literature spanning triadic clinical communication, shared decision-making, and AI-mediated interaction to derive a technology-neutral conceptual framework and classification model. We propose conversational capacity, operationalized through four functions (interpret, translate, advocate, adapt), as the minimum criterion for triadic facilitation. The framework defines a structural boundary separating genuine triadic architectures (Modes A and B) from augmented dyadic models (Mode C); a five-level Conversational Capacity Spectrum classifying human and AI facilitators; and a Clinical Situation Matrix mapping architecture to morbidity complexity, patient vulnerability, and decision complexity. Advocacy emerges as the discriminating function AI is least able to perform, making it the current limiting dimension of AI facilitation. The framework generates three testable hypotheses and identifies an equity paradox whereby high-vulnerability populations most in need of human facilitation are those most likely to be assigned AI on resource grounds. Triadic telemedicine should be defined by conversational capacity rather than mere third-party presence. Replacing the binary triadic-versus-dyadic distinction with two gradable, measurable constructs is a prerequisite for cumulative research and responsible AI deployment in clinical consultations.
Ergonomics is essential for ensuring safe, high-quality surgical care. However, existing studies vary widely in their clinical settings, terminology, and assessment methods. This scoping review aimed to map the terminology and methods used to evaluate surgeons' ergonomics in real-world surgical environments. We systematically searched MEDLINE, Embase, CENTRAL, IEEE Xplore, the WHO International Clinical Trials Registry Platform, and ClinicalTrials.gov from inception to July 1, 2025 for studies that assessed physical, cognitive, or organizational ergonomics. Two reviewers independently screened titles and abstracts, assessed full texts, and extracted study characteristics and outcomes. Assessment tools were categorized as self-reported, observational, direct or instrumental, or computer based. The review protocol was registered in PROSPERO (CRD420251088370). Ninety-nine studies met the inclusion criteria. Minimally invasive surgery predominated: 60 studies involved laparoscopic surgery (including thoracoscopic and other endoscopic minimally invasive approaches), and 36 evaluated robotic surgery. Comparative designs most frequently compared laparoscopic versus robotic surgery (n = 14). Physical, cognitive, and organizational ergonomics were assessed in 64, 49, and 19 studies, respectively, with 31 addressing multiple domains. Considerable variation was observed in terminology and assessment methods. Physical ergonomics was most commonly evaluated using observational or instrumental approaches, whereas cognitive ergonomics primarily relied on self-reported measures. Computer-based assessments, including artificial intelligence-assisted tools, have recently emerged. This review provides a structured framework for assessing ergonomics in surgery and highlights substantial heterogeneity in terminology and assessment methods. Standardized, multimodal, and objective approaches are needed to enhance comparability across studies.
Precise regulation of Cas12a activity is crucial for expanding its application in molecular diagnostics. However, existing split crRNA systems exhibit hardly any activation efficiency at low-abundance target and lack a well-defined regulated mechanism, representing a persistent bottleneck for practical application. This work proposes a DNA-guided spatially ordered assembly of split crRNA for activating CRISPR/Cas12a (DIRECTOR) strategy. This work combines artificial intelligence-driven AlphaFold3 structure prediction, computer-powered molecular dynamics simulations with fluorescence analysis to demonstrate that the 3' terminal extension of activator acts as a spatial director, utilizing DNA-guided spatially ordered assembly of split crRNA and stabilizing key Cas12a domains, thereby activating Cas12a. Conversely, the 5' terminal extension serves as a spatial misdirector, inhibiting Cas12a activation by destabilizing the protein structure and introducing the steric hindrance to shield the catalytic center. Furthermore, the structural and energy thresholds required for effective Cas12a activation were identified. Finally, utilizing the spatial director as an energy amplification element, DIRECTOR achieves a limit of detection as low as 42.1 fM for single-target miR-155 and dual-response detection of wide-scope nucleic acids. Owing to its direct activation strategy, DIRECTOR provides mechanistic insights for affordable and programmable CRISPR molecular diagnostics.
Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental condition characterized by diverse behavioral, cognitive, sensory, and communication profiles, making early diagnosis and personalized intervention challenging. Recent advances in machine learning (ML) and deep learning (DL) have enabled the development of computational tools for ASD screening, classification, severity assessment, and intervention monitoring. This review synthesizes findings from 50 recent studies that applied ML and DL techniques to ASD-related datasets, including electroencephalography (EEG), eye-tracking, behavioral video, microbiome, voice acoustic, demographic, and multimodal data. The review addresses three key questions: (i) which data modalities and computational approaches are most frequently used, (ii) how diagnostic performance is evaluated across different study designs, and (iii) what methodological challenges limit clinical translation. The literature is organized according to data modality, algorithmic approach, and clinical readiness. Approaches examined include conventional ML methods, convolutional neural networks, graph neural networks, hybrid deep learning architectures, federated learning, explainable artificial intelligence, topological data analysis, and multimodal fusion. The findings suggest that multimodal and graph-based approaches provide a more comprehensive representation of ASD phenotypes than single-modality methods. Explainability and privacy-preserving learning have also emerged as important considerations for clinical deployment. However, many reported high-performance models are based on small sample sizes, repeated use of the ABIDE dataset, class imbalance, single-site validation, or limited external testing, raising concerns regarding generalizability. Beyond diagnostic accuracy, this review evaluates model interpretability, calibration, scalability, validation rigor, and clinical applicability. Overall, the analysis highlights the need for standardized benchmarks, externally validated multimodal datasets, clinically relevant evaluation metrics, and decision-support systems that complement rather than replace expert clinical assessment in ASD diagnosis and management.
This article helps neurologists understand modern approaches to diagnosing obstructive sleep apnea, including the clinical role and limitations of home sleep apnea testing, and learn how they can integrate wearable and noncontact technologies into patient care to improve diagnostic efficiency, monitor treatment, and reduce health disparities. Advances in home sleep apnea testing have expanded beyond traditional type III monitors to include wearable devices such as wrist sensors and smart rings and noncontact "nearable" systems that use radar or acoustic signals. Since 2019, the US Food and Drug Administration (FDA) has cleared numerous software-as-a-medical-device platforms that leverage artificial intelligence and multisignal integration to estimate sleep parameters. These tools improve accessibility, particularly for patients who are unable or unwilling to undergo in-laboratory polysomnography. However, awareness of racial bias in pulse oximetry, regulatory gaps, variable accuracy, and privacy concerns highlight the need for further validation to ensure equitable and reliable clinical use. Obstructive sleep apnea is common and underdiagnosed, particularly in patients with neurologic conditions. In-laboratory polysomnography remains the gold standard for the diagnosis of obstructive sleep apnea but is limited by cost and access, whereas home sleep apnea testing offers a validated alternative for appropriate patients. Wearable and nearable technologies expand diagnostic options, improving usability and scalability, although their accuracy and validation vary. Neurologists play a key role in identifying patients at high risk, selecting the appropriate test, and interpreting results with awareness of limitations such as pulse oximetry bias. Consumer devices can raise awareness but should not replace clinical evaluation, and modern continuous positive airway pressure (CPAP) platforms enhance remote monitoring and management.
Artificial intelligence generated advertising is widely adopted in retailing and consumer services, yet its effects on trust, engagement, and purchase intention remain theoretically inconsistent. This study addresses this anomaly by reconceptualising advertising effectiveness under algorithmic authorship as a process of signal resolution rather than additive persuasion. Drawing on signalling theory and advertising value theory, the study specifies advertising value as a formative signal system composed of informativeness, entertainment, and executional credibility, which simultaneously activates competing inferential pathways of perceived credibility and perceived eeriness. Using a theory-driven PLS-SEM model estimated on a quota-based U.S. consumer sample (N = 412), the results provide associative evidence of systematic asymmetry and suppression effects: identical executional cues strengthen credibility while concurrently amplifying eeriness, with trust patterns consistent with the relative dominance of these opposing inferences rather than from overall message quality. Trust functions as a conditional transmission mechanism to engagement and purchase intention, and AI disclosure is associated with shifts in signal weighting by attenuating credibility-based pathways and amplifying eeriness-based suppression. By identifying signal competition as a structural feature associated with AI-generated advertising, the study extends current theoretical understanding of algorithmic persuasion by introducing signal competition as a structural feature associated with AI-generated advertising, departing from human-centric models and clarifying why creative AI execution often fails to yield behavioural conversion.