Artificial Intelligence Research Articles

Leveraging artificial intelligence to promote COVID-19 appropriate behaviour in a healthcare institution from north India: A feasibility study

Background & Objectives Non-pharmacological interventions (NPI) were crucial in curbing the initial COVID-19 pandemic waves, but compliance was difficult. The primary aim of this study was to assess the changes in compliance with NPIs in healthcare settings using Artificial intelligence (AI) and examine the barriers and facilitators of using AI systems in healthcare. Methods A pre-post-intervention study was conducted in a north-Indian hospital between April and July 2022. YOLO-V5 and 3D Cartesian distance algorithm-based AI modules were used to ascertain compliance through several parameters like confidence threshold, intersection-over-union threshold, image size, distance threshold (6 feet), and 3D Euclidean Distance estimation. Validation was done by evaluating model performance on a labelled test dataset, and accuracy was 91.3 per cent. Interventions included daily sensitization and health education for the hospital staff and visitors, display of information, education and communication (IEC) materials, and administrative surveillance. In-depth interviews were conducted with the stakeholders to assess the feasibility issues. Flagged events during the three phases were compared using One-way ANOVA tests in SPSS. Results Higher social distancing (SD) compliance events were flagged by the module in the intervention phase compared to the pre-intervention and post-intervention phases (P<0.05). Mask non-compliance was significantly lower (P <0.05) in the pre-intervention phase and highest in the post-intervention phase, with varied differences between different intervention phases in the registration hall and medicine out-patient department (OPD). The modules’ data safety, transfer, and cost were the most common concerns. Interpretation & conclusions AI can supplement our efforts against the pandemic and offer indispensable help with minimal feasibility issues that can be resolved through adequate sensitization and training.

Developing Trustworthy Artificial Intelligence Models to Predict Vascular Disease Progression: the VASCUL-AID-RETRO Study Protocol.

Abdominal aortic aneurysms (AAAs) and peripheral artery disease (PAD) are two vascular diseases with a significant risk of major adverse cardiovascular events and mortality. A challenge in current disease management is the unpredictable disease progression in individual patients. The VASCUL-AID-RETRO study aims to develop trustworthy multimodal predictive artificial intelligence (AI) models for multiple tasks including risk stratification of disease progression and cardiovascular events in patients with AAA and PAD. The VASCUL-AID-RETRO study will collect data from 5000 AAA and 6000 PAD patients across multiple European centers of the VASCUL-AID consortium using electronic health records from 2015 to 2024. This retrospectively-collected data will be enriched with additional data from existing biobanks and registries. Multimodal data, including clinical records, radiological imaging, proteomics, and genomics, will be collected to develop AI models predicting disease progression and cardiovascular risks. This will be done while integrating the international ethics guidelines and legal standards for trustworthy AI, to ensure a socially-responsible data integration and analysis. A consensus-based variable list of clinical parameters and core outcome set for both diseases will be developed through meetings with key opinion leaders. Blood, plasma, and tissue samples from existing biobanks will be analyzed for proteomic and genomic variations. AI models will be trained on segmented AAA and PAD artery geometries for estimation of hemodynamic parameters to quantify disease progression. Initially, risk prediction models will be developed for each modality separately, and subsequently, all data will be combined to be used as input to multimodal prediction models. During all processes, data security, data quality, and ethical guidelines and legal standards will be carefully considered. As a next step, the developed models will be further adjusted with prospective data and internally validated in a prospective cohort (VASCUL-AID-PRO study). The VASCUL-AID-RETRO study will utilize advanced AI techniques and integrate clinical, imaging, and multi-omics data to predict AAA and PAD progression and cardiovascular events. The VASCUL-AID-RETRO study is registered at www.clinicaltrials.gov under the identification number NCT06206369. The VASCUL-AID-RETRO study aims to improve clinical practice of vascular surgery by developing artificial intelligence-driven multimodal predictive models for patients with abdominal aortic aneurysms or peripheral artery disease, enhancing personalized medicine. By integrating comprehensive data sets including clinical, imaging, and multi-omics data, these models have the potential to provide accurate risk stratification for disease progression and cardiovascular events. An innovation lies in the extensive European data set in combination with multimodal analyses approaches, which enables the development of advanced models to facilitate better understanding of disease mechanisms and progression. For clinicians, this means that more precise, individualized treatment plans can be established, ultimately aiming to improve patient outcomes.

Use of Big Data, Artificial Intelligence and Other Emerging Technologies in Public Health Communication Campaigns: A Systematic Review

Introduction: Public health campaigns have begun to employ various emerging technologies such as artificial intelligence (AI) and Big Data for their design, implementation, and evaluation. These technologies offer new opportunities to enhance the effectiveness of health communication strategies by enabling more precise audience segmentation, personalized messaging, and real-time impact assessment. However, despite their potential, their application in public health campaigns remains a relatively new field of study. This systematic review aims to identify and analyze the scientific literature that examines the role of these tools in optimizing communication strategies aimed at promoting healthy behaviors and attitudes among the population. Methods: Following the PRISMA methodology, 129 potentially relevant articles were initially identified from the search in the WOS, Scopus and PubMed databases. However, only 18 met the established inclusion criteria. Results: We identified a scarcity of scientific articles focused on presenting findings on the application of emerging technologies in public health campaigns. Most of the research focused on the use of these tools for audience segmentation and impact assessment, while their use in campaign design was a minority. Conclusions: This study supports the usefulness of the application of these technologies at any stage of the campaign implementation process, and presents some limitations that can help optimize this approach and improve the reach and effectiveness of public health communication strategies.

Atopic dermatitis: a comprehensive updated review of this intriguing disease with futuristic insights.

Atopic dermatitis (AD) is a paradigmatic prevalent, long-lasting, and inflammatory skin condition with a diverse range of clinical manifestations. The etiology and clinical symptoms of AD are influenced by complex pathophysiological processes, which involve a strong genetic component, epidermal dysfunction, and immunological dysregulation, and a strong influence of other physiological and environmental factors. The FDA has approved targeted and well-tolerated immunomodulators including biologics like dupilumab and crisaborole, and small molecules such as baricitinib, as novel therapies for AD. They effectively treat AD but are too expensive for most patients. The review provides an update on the state of knowledge of AD pathogenesis, discusses the available diagnostic and scoring indices, and provides a scientific foundation for treatment methods for AD. This review also presents data on clinical efficacy of innovative treatments' considering recent guidelines, emphasizing the newest medications and ongoing trials. Finally, the new implication of artificial intelligence (AI) in AD management is explored, where AI can speed up diagnosis and therapy. The PubMed, Google Scholar, and ScienceDirect databases were used for this review.

The implementation of knowledge-based planning with partial OAR contours for prostate radiotherapy.

Intra- and inter-observer contour uncertainty is a continuous challenge in treatment planning for radiotherapy. Our proposed solution to address this challenge is the use of partial contours for treatment planning, focusing on uninvolved or non-overlapping portions of the organs-at-risk (OARs) with the planning target volume (PTV). The partial contours systematically eliminate overlapping regions. The partial contours were evaluated against fully contoured OARs. We incorporated advanced tools like knowledge-based planning (KBP) to create treatment plans and artificial intelligence (AI) to create auto-segmented contours. We developed two models, Rapid Plan (RP) and Rapid Plan partial uninvolved (RP_Part_Un), using 70 previous clinically approved volumetric arc therapy (VMAT) plans each prescribed with 70 Gy/28 fractions. From these models, we created three plans, RP, RP_Part_Un, and MIM AI_Part_Un. In this retrospective study, 60 prostate patients were analyzed using the three plans. For determining OAR sparing, Dmax and Dmean along the percent volume receiving a dose over a range (V10 Gy V70 Gy) between each plan were compared. Geometric evaluations, dice similarity coefficient (DSC), and overlay index (OI) between the OAR contours from partial-contoured manual structure sets and partial-contoured AI structure sets were analyzed. When comparing the DSC and OI for full contours to the partial contours, in both groups, all comparisons were significantly increased for both organs. This indicated the partial contours had a higher degree of concordance. In patients with SpaceOAR, RP_Part_Un plans exhibited significantly reduced bladder Dmax and Dmean compared to RP plans, while rectum Dmax and Dmean showed no significant differences. For patients without SpaceOAR, RP_Part_Un significantly lowered rectum Dmean. MIM AI_Part_Un plans demonstrated lower rectum Dmax in both patient groups. Partial contours, defined at a specified distance from the PTV, yielded dosimetry comparable to fully contoured plans, highlighting their potential efficacy in treatment planning.

Thessaly Graft Index: An Artificial Intelligence-Based Index for the Assessment of Graft Integrity in ACL-Reconstructed Knees.

Magnetic resonance imaging (MRI) has proven to be a valuable noninvasive tool to evaluate graft integrity after anterior cruciate ligament (ACL) reconstruction. However, MRI protocols and interpretation methodologies are quite diverse, preventing comparisons of signal intensity across subsequent scans and independent investigations. The purpose of this study was to create an artificial intelligence (AI)-based index (Thessaly Graft Index [TGI]) for the evaluation of graft integrity following ACL reconstruction. The cohort study included 24 patients with an isolated ACL injury that had been treated with a hamstring tendon autograft and followed for 1 year. MRI was performed preoperatively and 1 year postoperatively. The clinical and functional evaluations were performed with use of the KT-1000 and with the following patient-reported outcome measures (PROMs): the Knee Injury and Osteoarthritis Outcome Score (KOOS), the International Knee Documentation Committee Subjective Knee Function form (IKDC), the Lysholm score, and the Tegner Activity Scale (TAS). An AI model, based on the YOLOv5 Nano version, was designed to compute the probability of accurately detecting, in the sagittal plane, a healthy ACL (on a percentage scale) and was trained on healthy and injured knees from the KneeMRI dataset. The model was used to assess the integrity of ACL grafts, with a maximum score of 100. The results were compared with the MRI assessment from an independent radiologist and were correlated with PROMs and KT-1000 laxity. The mean preoperative and postoperative TGI scores were 64.21 ± 8.96 and 82.37 ± 3.53, respectively. A mean increase of 15% in the TGI scores was observed between preoperative and postoperative images. The minimum threshold for TGI to categorize a graft as healthy on the postoperative MRI was 79.21%. Twenty-two grafts were characterized as intact and 2 as reruptured, with postoperative TGI scores of 71% and 42%. The radiologist's assessment was in total agreement with the TGI scores. The correlation of the TGI ranged from moderate to good with the TAS (0.668), IKDC (0.516), Lysholm (0.521), KOOS total (0.594), and KT-1000 (0.561). The TGI is an AI tool that is able to accurately recognize an ACL graft rupture. Moreover, the TGI correlated with the KT-1000 postoperative values and PROM scores. Diagnostic Level IV. See Instructions for Authors for a complete description of levels of evidence.

Artificial intelligence (AI) for reinforced autoclaved aerated concrete (RAAC) crack defect identification

PurposeReinforced autoclaved aerated concrete (RAAC) panels have been extensively used in the UK since the 1960s as structural roofs, floors and walls. The lack of a longitudinal, objective, consistent defect data capture process has led to inaccurate, invalid and incomplete RAAC data, which limits the ability to survey RAAC within buildings and monitor performance. Therefore, an accurate, complete and valid digital data capture process is needed to facilitate better RAAC performance and defect monitoring. This paper presents the development of an artificial intelligence (AI)-driven RAAC crack defect capture tool for improving the quality of RAAC survey data.Design/methodology/approachRAAC crack defect image data were collected, curated and trained. A deep learning approach was employed to train RAAC surveyed defects (cracks) images from two hospitals. This approach mitigated unavoidable occlusions/obstructions and unintended “foreign” objects and textures.FindingsAn automatic RAAC crack identification tool has been developed to be integrated into RAAC survey processes via an executable code. The executable code categorises RAAC survey images into “crack” or “non-crack” and can provide longitudinal graphical evidence of changes in the RAAC over time.Originality/valueThis paper identifies the role of AI in addressing the intrinsic defects data capture issues for RAAC and extends current debates on data-driven solutions for defect capture and monitoring.

Influence of scan extension on the accuracy of maximum intercuspal position recorded by using intraoral scanners or an artificial intelligence-based program.

Intraoral scanners (IOSs) and artificial intelligence (AI) based programs can be used to locate the maximum intercuspal position (MIP). However, the influence of scan extension on the accuracy of the MIP located by using these technologies is uncertain. The purpose of this in vitro study was to analyze the effect of scan extension on the accuracy of the MIP located by using 3 IOSs and an AI-based program. Stone casts mounted in an articulator in MIP were digitized (T710). Two groups were created: complete- (CA group) and half arch (HA group) scan. In the CA-group, complete arch scans of the reference casts were captured with each IOS tested. The nonarticulated scans were duplicated 20 times. In the HA-groups, the right half arch scans of the reference casts were captured with each IOS tested. Six subgroups were generated: 3 IOS (Primescan-IOS, i700-IOS, and Aoralscan3-IOS) and 3 AI (Primescan-AI, i700-AI, and Aoralscan3-AI) subgroups. In the CA-Primescan-IOS subgroup, 10 duplicated scans were articulated in MIP by recording a bilateral occlusal record. In the CA-Primescan-AI subgroup, 10 duplicated scans were articulated in MIP by using an AI-based program (Bitefinder). In the CA-i700-IOS, CA-Aoralscan3-IOS, CA-i700-AI, and CA-Aoralscan3-AI subgroups, the same procedures as in the CA-Primescan-IOS and CA-Primescan-AI subgroups were completed, respectively. In the HA-Primescan-IOS subgroup, 10 duplicated scans were articulated in MIP by capturing a right occlusal record. In the HA-Primescan-AI subgroup, 10 duplicated scans were articulated in MIP by using the AI-based program. In the HA-i700-IOS, HA-Aoralscan3-IOS, HA-i700-AI, and HA-Aoralscan3-AI subgroups, the same procedures as in the HA-Primescan-IOS subgroups were completed, respectively. A program (Geomagic) was used to calculate 36 interlandmark measurements on the virtual articulated casts (control) and each specimen. Three-way ANOVA and Tukey tests were used to analyze trueness (α=.05). The Levene and pairwise multiple comparison tests were used to analyze precision (α=.05). MIP trueness discrepancies were found between the IOS (P<.001), groups (P<.001), and subgroups (P<.001), with a significant interaction IOS×subgroup (P<.001), group×subgroup (P<.001), and IOS×group×subgroup (P<.001). The Primescan and i700 (P=.014) and the Primescan and Aoralscan3 (P<.001) were different from each other. The CA and HA groups (P<.001) were different from each other. The IOS and AI subgroups (P<.001) were different from each other. The Levene test showed significant precision discrepancies between the groups (P<.001) and subgroups (P<.001). The HA scans demonstrated significantly worse precision than the CA scans (P<.001). Additionally, the AI-based program obtained significantly worse precision than the IOS programs tested (P<.001). Scan extension and program impacted the trueness and precision of the MIP. CA groups demonstrated better MIP trueness and worse precision than the HA groups. Primescan obtained better MIP trueness than the i700 and Aoralscan3 systems. The IOSs revealed better MIP trueness and precision than the AI-based program tested.

Artificial intelligence with ChatGPT 4: a large language model in support of ocular oncology cases.

To evaluate ChatGPT's ability to analyze comprehensive case descriptions of patients with uveal melanoma and provide recommendations for the most appropriate management. Retrospective analysis of ocular oncology patients' medical records. Forty patients treated for uveal melanoma between May 2019 and October 2023. We uploaded each case description into the ChatGPT interface (version 4.0) and asked the model to provide realistic treatment options by asking the question, "What type of treatment do you recommend?" The accuracy of decisions produced by ChatGPT was compared to those recorded in patients' files and the treatment recommendations provided by three ocular oncologists, each with more than 10years of experience. The primary objective of this research was to assess the accuracy of ChatGPT replies in ocular oncology cases, analyzing its competence in both straightforward and intricate situations. Our secondary purpose was to assess the concordance between the responses of ChatGPT and those of ocular oncology specialists when faced with analogous clinical scenarios. ChatGPT's surgical choices matched those in patients' files in 55% of cases (22 out of 40). ChatGPT options were agreed upon by 50%, 55%, and 57% of the three ocular oncologyspecialists. The investigation revealed significant differences between ChatGPT's responses and those of the three cancer specialists when compared to patients' files (p = 0.003, p = 0.001, and p = 0.001). ChatGPT's surgical responsesmatched with patient data in 18 out of 24 cases (75%), excluding enucleation cases. The decisions matched with the three ocular oncology specialists in 17/24, 18/24, and 18/24 cases, reflecting agreements of 70%, 75%, and 75%, respectively. The decisions made by ChatGPT were not significantly different from those of the three professionals in this cohort (p = 0.50, p = 0.36, and p = 0.36 for ChatGPT compared to specialists 1, 2, and 3). ChatGPT exhibited a level of proficiency that was comparable to that of trained ocular oncology specialists. However, it exhibited its limitations when evaluating more complex scenarios, such as extrascleral extension or infiltration of the optic nerve, when a comprehensive evaluation of the patient is therefore necessary.

Modelling and optimization of well hole cleaning using artificial intelligence techniques

Ineffective hole cleaning in deviated and horizontal well drilling can lead to issues like stuck pipes, reduced rate of penetration (ROP), and drill bit damage, resulting in increased non-productive time (NPT) and operational costs. Traditional methods for assessing hole cleaning rely on experimental and empirical models that often fail to account for all influencing factors and lack real-time applicability. This study aims to improve the accuracy and practicality of hole cleaning assessment by applying Artificial Intelligence (AI) techniques, specifically Artificial Neural Networks (ANN) and Genetic Algorithms (GA), to predict downhole parameters and optimize drilling processes. These AI methods analyze the impact of key drilling parameters—such as weight on bit (WOB), ROP, rock geomechanics, drilling fluid characteristics, and rig hydraulics—on hole cleaning. Results demonstrated that AI-driven models provide high-precision predictions and enable real-time optimization, significantly reducing NPT and enhancing drilling efficiency and safety. In conclusion, AI techniques like ANN and GA offer a robust solution to improve hole cleaning, overcoming limitations of traditional methods and contributing to safer, more cost-effective drilling operations.

Establishing methodological standards for the development of artificial intelligence-based Clinical Decision Support in emergency medicine.

Artificial intelligence (AI) offers opportunities for managing the complexities of clinical care in the emergency department (ED), and Clinical Decision Support has been identified as a priority application. However, there is a lack of published guidance on how to rigorously develop and evaluate these tools. We sought to answer the question, "What methodological standards should be applied to the development of AI-based Clinical Decision Support tools in the ED?". We conducted an iterative consensus-establishing activity involving a subcommittee with AI expertise followed by surveys and a live facilitated discussion with participants of the 2024 Canadian Association of Emergency Physicians Research Symposium in Saskatoon. We augmented analysis of participant feedback with large language models. We established 11 recommendations AI-based Clinical Decision Support development including the selection of a relevant problem and team of experts, standards of data quality and quantity, novel AI-specific reporting guidelines, and adherence to principles of ethics and privacy. We removed the recommendation regarding model interpretability from the final list due to a lack of consensus. These 11 recommendations provide guiding principles and methodological standards for emergency medicine researchers to rigorously develop AI-based Clinical Decision Support tools and for clinicians to gain knowledge and trust in using them.

Speech Acts and Hidden Meaning: A Journey into Pragmatics

This article examines the core principles of pragmatics, focusing on how speech acts and hidden meanings guide communication across diverse contexts. It explores John Austin and John Searle’s foundational work on speech acts—locutionary, illocutionary, and perlocutionary—demonstrating how utterances perform actions beyond mere information exchange. The discussion then turns to implicature, introduced by H. P. Grice, highlighting the ways in which inferred meanings depend on shared knowledge and context. By integrating cultural considerations, the article underscores how pragmatic competence shapes both cross-cultural understanding and effective discourse. Practical applications include challenges in translation, advances in artificial intelligence, and strategies in media and political communication. Ultimately, the study affirms that pragmatics offers crucial insights into the power of language to form relationships, shape societies, and influence global interaction.

Multiple objective programming and goal programming: making better decisions with artificial intelligence and business analytics

Multiple objective programming and goal programming: making better decisions with artificial intelligence and business analytics

Accurately Models the Relationship Between Physical Response and Structure Using Kolmogorov-Arnold Network.

Artificial intelligence (AI) in science is a key area of modern research. However, many current machine learning methods lack interpretability, making it difficult to grasp the physical mechanisms behind various phenomena, which hampers progress in related fields. This study focuses on the Poisson's ratio of a hexagonal lattice elastic network as it varies with structural deformation. By employing the Kolmogorov-Arnold Network (KAN), the transition of the network's Poisson's ratio from positive to negative as the hexagonal structural element shifts from a convex polygon to a concave polygon wasaccurately predicted. The KAN provides a clear mathematical framework that describes this transition, revealing the connection between the Poisson's ratio and the geometric properties of the hexagonal element, and accurately identifying the geometric parameters at which the Poisson's ratio equals zero. This work demonstrates the significant potential of the KAN network to clarify the mathematical relationships that underpin physical responses and structuralbehaviors.

Enhancing feature selection for multi-pose facial expression recognition using a hybrid of quantum inspired firefly algorithm and artificial bee colony algorithm

Facial expression recognition (FER) has advanced applications in various disciplines, including computer vision, Internet of Things, and artificial intelligence, supporting diverse domains such as medical escort services, learning analysis, fatigue detection, and human-computer interaction. The accuracy of these systems is of utmost concern and depends on effective feature selection, which directly impacts their ability to accurately detect facial expressions across various poses. This research proposes a new hybrid approach called QIFABC (Hybrid Quantum-Inspired Firefly and Artificial Bee Colony Algorithm), which combines the Quantum-Inspired Firefly Algorithm (QIFA) with the Artificial Bee Colony (ABC) method to enhance feature selection for a multi-pose facial expression recognition system. The proposed algorithm uses the attributes of both the QIFA and ABC algorithms to enhance search space exploration, thereby improving the robustness of features in FER. The firefly agents initially move toward the brightest firefly until identified, then search transition to the ABC algorithm, targeting positions with the highest nectar quality. In order to evaluate the efficacy of the proposed QIFABC algorithm, feature selection is also conducted using QIFA, FA, and ABC algorithms. The evaluated features are utilized for classifying face expressions by utilizing the deep neural network model, ResNet-50. The presented FER system has been tested using multi-pose facial expression benchmark datasets, including RaF (Radboud Faces) and KDEF (Karolinska Directed Emotional Faces). Experimental results show that the proposed QIFABC with ResNet50 method achieves an accuracy of 98.93%, 94.11%, and 91.79% for front, diagonal, and profile poses on the RaF dataset, respectively, and 98.47%, 93.88%, and 91.58% on the KDEF dataset.

The Evolution of Modern Trends in Education Quality Management

Education quality management has undergone significant transformations in recent decades, influenced by global demands, technological advancements, and changing educational paradigms. This research paper explores the evolution of modern trends in education quality management (EQM), focusing on the key drivers that have shaped its development. Initially, education systems were designed with a traditional, teacher-centered approach, but the need for more dynamic, student-centered learning environments led to the introduction of quality assurance (QA) mechanisms such as accreditation, performance-based assessments, and standards-setting initiatives. The paper examines how the rise of digital tools, data analytics, and artificial intelligence (AI) has enabled more personalized learning experiences and real-time tracking of educational outcomes. Furthermore, the shift towards inclusivity and the recognition of diverse learner needs have prompted educational institutions to adopt more flexible and adaptive systems for quality management. In addition, globalization has encouraged cross-border collaborations, resulting in the development of international frameworks and best practices for maintaining educational standards. By reviewing contemporary models, including Total Quality Management (TQM) and Six Sigma, the paper assesses how EQM has become increasingly focused on continuous improvement and stakeholder engagement. The research also discusses the challenges and opportunities posed by current trends, including the integration of sustainability goals and the importance of preparing students for a rapidly evolving workforce. Ultimately, this paper highlights the importance of adapting education quality management to meet the evolving demands of society and the global economy, ensuring that education remains accessible, equitable, and of the highest possible standard.

Review of selected methods for estimating risk from occupational exposure to carcinogenic substances

Assessing the expected level of cancer risk associated with occupational exposure to chemical substances is extremely important, as it forms the basis for the proper and safe organization of work in a company. It allows the translation of scientific research findings into legal regulations, enabling the implementation of appropriate preventive measures and optimizing the protection of workers' health. This article discusses various methods for estimating the risk associated with the occurrence of cancer due to exposure to carcinogenic substances in workplaces. An analysis, summary, and synthesis of the literature were conducted, based on which 7 methods for estimating cancer risk were selected. These methods utilize descriptors (features or variables used to describe data), numerical modeling techniques measured during observations, and approaches based on artificial intelligence. The decision to select these specific methods was driven by their representativeness of different approaches to risk assessment, allowing for a comprehensive analysis and comparison of their strengths and weaknesses. The analysis revealed that there is no universal best method for risk estimation, and the choice largely depends on the data available from experimental animal studies. Key factors include the species and number of animals used in the experiment giving statistical power, the route of administration, the duration of the experiment, the incidence of spontaneous tumors, and the appropriate selection of tumors that might occur in humans. There is a need for continuous improvement and harmonization of risk assessment methods and the incorporation of the latest scientific research to effectively manage health risks associated with exposure to carcinogenic substances in the workplace. Med Pr Work Health Saf. 2025;76(1).

Widening the divide, reloading dependency: the political economy of data and AI capability building in Latin America

ABSTRACT The availability of data and AI-based applications has increased in recent years, but the people with the skills to use these technologies remain limited. To fill this gap, several Latin American universities and training centres have expanded their academic offerings on data and AI in recent years. Based on a bibliometric analysis, a hundred in-depth interviews with experts from Ecuador, Chile, and Mexico, and an extensive documentary analysis of academic programmes and public policy initiatives in these countries, this paper aims to identify and characterize the capability building processes in Latin America. Our results show the growing asymmetries in the production and distribution of knowledge and know-how on big data and artificial intelligence in the world and in the region. These emerging asymmetries are leading to a steady transfer of resources and professionals trained by Latin American public institutions to national and transnational private companies, updating old structures of dependency.

Predictive career guidance and entrepreneurial development for university students using artificial intelligence and machine learning

Predictive career guidance and entrepreneurial development for university students using artificial intelligence and machine learning

Unveiling GPT-4V's hidden challenges behind high accuracy on USMLE questions: Observational Study.

Recent advancements in artificial intelligence, such as GPT-3.5 Turbo (OpenAI) and GPT-4, have demonstrated significant potential by achieving good scores on text-only United States Medical Licensing Examination (USMLE) exams and effectively answering questions from physicians. However, the ability of these models to interpret medical images remains underexplored. This study aimed to comprehensively evaluate the performance, interpretability, and limitations of GPT-3.5 Turbo, GPT-4, and its successor, GPT-4 Vision (GPT-4V), specifically focusing on GPT-4V's newly introduced image-understanding feature. By assessing the models on medical licensing examination questions that require image interpretation, we sought to highlight the strengths and weaknesses of GPT-4V in handling complex multimodal clinical information, thereby exposing hidden flaws and providing insights into its readiness for integration into clinical settings. This cross-sectional study tested GPT-4V, GPT-4, and ChatGPT-3.5 Turbo on a total of 227 multiple-choice questions with images from USMLE Step 1 (n=19), Step 2 clinical knowledge (n=14), Step 3 (n=18), the Diagnostic Radiology Qualifying Core Exam (DRQCE) (n=26), and AMBOSS question banks (n=150). AMBOSS provided expert-written hints and question difficulty levels. GPT-4V's accuracy was compared with 2 state-of-the-art large language models, GPT-3.5 Turbo and GPT-4. The quality of the explanations was evaluated by choosing human preference between an explanation by GPT-4V (without hint), an explanation by an expert, or a tie, using 3 qualitative metrics: comprehensive explanation, question information, and image interpretation. To better understand GPT-4V's explanation ability, we modified a patient case report to resemble a typical "curbside consultation" between physicians. For questions with images, GPT-4V achieved an accuracy of 84.2%, 85.7%, 88.9%, and 73.1% in Step 1, Step 2 clinical knowledge, Step 3 of USMLE, and DRQCE, respectively. It outperformed GPT-3.5 Turbo (42.1%, 50%, 50%, 19.2%) and GPT-4 (63.2%, 64.3%, 66.7%, 26.9%). When GPT-4V answered correctly, its explanations were nearly as good as those provided by domain experts from AMBOSS. However, incorrect answers often had poor explanation quality: 18.2% (10/55) contained inaccurate text, 45.5% (25/55) had inference errors, and 76.3% (42/55) demonstrated image misunderstandings. With human expert assistance, GPT-4V reduced errors by an average of 40% (22/55). GPT-4V accuracy improved with hints, maintaining stable performance across difficulty levels, while medical student performance declined as difficulty increased. In a simulated curbside consultation scenario, GPT-4V required multiple specific prompts to interpret complex case data accurately. GPT-4V achieved high accuracy on multiple-choice questions with images, highlighting its potential in medical assessments. However, significant shortcomings were observed in the quality of explanations when questions were answered incorrectly, particularly in the interpretation of images, which could not be efficiently resolved through expert interaction. These findings reveal hidden flaws in the image interpretation capabilities of GPT-4V, underscoring the need for more comprehensive evaluations beyond multiple-choice questions before integrating GPT-4V into clinical settings.