FDA-cleared artificial intelligence and machine learning-based medical devices and their 510(k) predicate networks

Artificial intelligence and machine learning (AI/ML) medical devices are able to optimise their performance by learning from past experience. In healthcare, such devices are already applied within controlled settings in image analysis systems to detect conditions like diabetic retinopathy, for instance. In examining the regulatory governance of AI/ML medical devices in the United States, it is argued that the development and application of these devices as a technical and social concern whether in research or in clinical care must proceed in tandem with their identities in regulation. In the light of emerging regulatory principles and approaches put forward by the International Medical Device Regulators Forum, and endorsed by the US Food and Drug Administration, conventional thinking about clinical research and clinical practice as distinct and separate domains needs to be reconsidered. The high connectivity of AI/ML medical devices that are capable of adapting to their digital environment in order to optimise performance suggest that the research agenda persist beyond what may be currently limited to the pilot or feasibility stages of medical device trials. If continuous risk-monitoring is required to support the use of software as medical devices in a learning healthcare system, more robust and responsive regulatory mechanisms are needed, not less.

Purpose Law enforcement agencies in the United States are relying on state fusion centers for intelligence to develop actionable, data-driven reports that increase efficiency and improve investigations in crime prevention and homeland security. This study assesses the extent to which artificial intelligence and machine learning (AI/ML) are increasingly shaping intelligence operations in law enforcement and the functions of state fusion centers in supporting intelligence-led policing (ILP). The reliance and integration of AI/ML is improving analytic accuracy, situational awareness and information and data sharing and collaboration among law enforcement and homeland security agencies. This study examines the state of contemporary academic literature, assesses AI/ML applications used in law enforcement and builds a conceptual and theoretical framework centered on ILP policing. It also relies on empirical data, case study applications, and DHS assessments to explore the degree to which AI-driven processes and analytics enhance criminal intelligence, investigative efficiencies, situational awareness and predictive policing. The analysis, while focusing on the opportunities and challenges of using AI/ML tools in law enforcement, also highlights the need for ethical governance, transparency and accountability when relying on advanced technologies for crime prevention and policing. Design/methodology/approach This study utilizes qualitative methods, including a thematic content analysis of government and think tank/practitioner reports as well as academic literature on the benefits, costs and ethical factors regarding variations in the implementation of AI/ML tools for law enforcement intelligence products and resource allocation. For cross-validation of operational outcomes, it examines publicly available information in the DHS Fusion Center Annual Assessment, Bureau of Justice Statistics, LEMAS and RAND Corporation assessments of intelligence-led policing. Findings Qualitative Results Federal and state sources report fusion centers and law enforcement agencies integrating advanced analytic and ML-enabled tools into each step in the criminal intelligence lifecycle process. However, ethical and structural challenges limit and constrain technology-driven narratives in fusion centers. Given these challenges, there is a consistent qualitative and thematic pattern: state fusion centers now function as criminal intelligence analytic hubs or resources that leverage the most contemporary analytic and data-driven tools for criminal intelligence and law enforcement investigations. Interrelated themes describe AI/ML technologies in terms of shaping, constraining and complicating the intelligence lifecycle in fusion centers and law enforcement operations. Seven specific themes emerged from latent coding are illustrated in the chart. Research limitations/implications There are limitations on the collection of quantitative data since DHS, leading think tanks and NGOs do not disclose specific figures on the proportion of AI/ML tools. The DHS Fusion Center Annual Assessment process monitors technology adoption and the growth of analytic capabilities throughout the national network of fusion centers; however, the specific quantitative statistics are not disclosed in public summaries (DHS, 2024). Second, publicly available data and information constitute the bulk of empirical sources. Consequently, this study relies primarily on qualitative narrative reporting, not quantitative performance metrics. Third, publication bias is likely present in industry and government sources as these reports may provide overly optimistic observations and conclusions while overlooking ethical dilemmas, failures and challenges. Moreover, qualitative thematic analysis could reflect broader structural narratives as opposed to empirical outcomes. Finally, since AI/ML adoption varies across fusion centers and according to technology levels, qualitative themes identified in this study must be read as representative patterns and not as universally generalizable. Originality/value Fusion center utilization of AI/ML technologies is as much an operational tool as it is a policy, governance and ethical challenge. Successful and professional use in support of law enforcement is about placing technological innovations firmly within institutional accountability and constitutional guardrails. On the one hand, AI/ML tools are enhancing analytical intelligence production by accelerating analytic workflows, predictive modeling and expanding data/information integration capabilities. AI/ML are extending ILP concepts by offering improvements in situational awareness and threat identification and operational efficiencies. On the other, substantial constraints hinder responsible use of these technologies. In the absence of standardized oversight frameworks, data-quality issues, algorithmic bias and the lack of professional development, workforce capacity and critical skills on the part of fusion center analysts will cancel the benefits of these tools.

American Indian and Alaska Native (AI/AN) communities are at a critical juncture in health research, where combining participatory methods with advancements in artificial intelligence and machine learning (AI/ML) can promote equity. Community-based participatory research methods which emerged to help Alaska Native communities navigate the complicated legacy of historical research abuses provide a framework to allow emerging AI/ML technologies to align with their unique world views, community strengths, and healthcare goals. A consortium of researchers (including Alaska Native Tribal Health Consortium, the Center for Alaska Native Health Research at University of Alaska, Fairbanks, Stanford University, Southcentral Foundation, and Maniilaq Association) is using community-engaged AI/ML methods to address air medical ambulance (medevac) utilization in rural communities within the Alaska Tribal Health System (ATHS). This mixed-methods convergent triangulation study uses qualitative and quantitative analyses to develop AI/ML models tailored to community needs, provider concerns, and cultural contexts. Early successes have led to a second funded project to expand community perspectives, pilot models, and address issues of governance and ethics. Using the Ethical, Legal, and Social Implications of Research framework to address implementation of AI/ML in AI/AN communities, this second grant expands community engagement, technical capacity, and creates a body within the ATHS able to provide recommendations about AI/ML security, privacy, governance and policy. These two projects have the potential to provide equitable AI/ML implementation in Alaska Native healthcare and provide a roadmap for researchers and policy makers looking to effect similar change in other AI/AN and marginalized communities.

ABSTRACTIntroductionEvidence syntheses are crucial in healthcare and elsewhere but are resource‐intensive, often taking years to produce. Artificial intelligence and machine learning (AI/ML) tools may improve production efficiency in certain review phases, but little is known about their impact on entire reviews.MethodsWe performed prespecified analyses of a convenience sample of eligible healthcare‐ or welfare‐related reviews commissioned at the Norwegian Institute of Public Health between August 1 2020 (first commission to use AI/ML) and January 31 2023 (administrative cut‐off). The main exposures were AI/ML use following an internal support team's recommendation versus no use. Ranking (e.g., priority screening), classification (e.g., study design), clustering (e.g., documents), and bibliometric analysis (e.g., OpenAlex) tools were included, but we did not include or exclude specific tools. Generative AI tools were not widely available during the study period. The outcomes were resources (person‐hours) and time from commission to completion (approval for delivery, including peer review; weeks). Analyses accounted for nonrandomized assignment and censored outcomes (reviews ongoing at cut‐off). Researchers classifying exposures were blinded to outcomes. The statistician was blinded to exposure.ResultsAmong 39 reviews, 7 (18%) were health technology assessments versus systematic reviews, 19 (49%) focused on healthcare versus welfare, 18 (46%) planned meta‐analysis, and 3 (8%) were ongoing at cut‐off. AI/ML tools were used in 27 (69%) reviews. Reviews that used AI/ML as recommended used more resources (mean 667 vs. 291 person‐hours) but were completed slightly faster (27.6 vs. 28.2 weeks). These differences were not statistically significant (relative resource use 3.71; 95% CI: 0.36–37.95; p = 0.269; relative time‐to‐completion: 0.92; 95% CI: 0.53–1.58; p = 0.753).ConclusionsAssociations between AI/ML use and the outcomes remains uncertain. Multicenter studies or meta‐analyses may be needed to determine if these tools meaningfully reduce resource use and time to produce evidence syntheses.

<div class="section abstract"><div class="htmlview paragraph">Currently, there are no safe and suitable fuel sources with comparable power density to traditional combustible fuels capable of replacing Internal Combustion Engines (ICEs). For the foreseeable future, civilian and military systems are likely to be reliant on traditional combustible fuels. Hybridization of the vehicle powertrains is the most likely avenue which can reduce emissions, minimize system inefficiencies, and build more sustainable vehicle systems that support the United States Army modernization priorities. Vehicle systems may further be improved by the creation and implementation of artificial intelligence and machine learning (AI/ML) in the form of advanced predictive capabilities and more robust control policies. AI/ML requires numerous characterized and complete datasets, given the sensitive nature of military systems, such data is unlikely to be known or accessible limiting the reach to develop and deploy AI/ML to military systems. With the absence of data, AI/ML may still be developed and deployed to military systems if supported by near-real-time or real-time computationally efficient and effective hardware and software or cloud-based computing. In this research, an OPAL real-time (OPAL-RT) simulator was used to emulate a compression ignition (CI) engine simulation architecture capable of developing and deploying advanced AI/ML predictive algorithms. The simulation architecture could be used for developing online predictive capabilities required to maximize the effectiveness or efficiency of a vehicle. The architecture includes a real-time simulator (RTS), a host PC, and a secondary PC. The RTS simulates a crank angle resolved engine model which utilized pseudo engine dynamometer data in the form of multi-dimensional matrices to emulate quasi-steady state conditions of the engine. The host PC was used to monitor and control the engine while the secondary PC was used to train the AI/ML to predict the per-cylinder generated torque from the crank shaft torque, which was then used to predict the in-cylinder temperature and pressure. The results indicate that using minimal sensor data and pretrained predictive algorithms, in-cylinder characterizations for unobserved engine variables may be achievable, providing an approximate characterization of quasi-steady state in-cylinder conditions.</div></div>

Driven by the emerging trend for transparent, open and programmable communications, Open Radio Access Network (O-RAN) constitutes the dominant architectural approach for deploying the future wireless networks. Towards standardizing and specifying the building blocks and principles of O-RAN, a coordinated global effort has been observed, mainly comprised of the O-RAN Alliance, the operators and several research activities. This paper presents the architectural aspects and the current status of O-RAN deployments, integrating both existing and ongoing activities from the O-RAN enablers. Furthermore, since the Artificial Intelligence and Machine Learning (AI/ML) act as key pillars for realizing O-RANs, a comprehensive view on the AI/ML functionality is provided as well. Additionally, a Network Telemetry (NT) architecture is also proposed to ensure end-to-end data collection and real-time analytics. To concretely illustrate the O-RAN supporting mechanisms for hosting AI/ML, we implemented two realistic ML algorithms: (i) a Supervised Learning (SL) based algorithm for cell traffic prediction using the training data of an open dataset and (ii) a Deep Reinforcement Learning (DRL) based algorithm for energy-efficiency maximization using a 5G-compliant simulator to obtain RAN measurements. We schematically demonstrate the AI/ML workflow for both ML-assisted algorithms through the usage of xApps running on the Radio Intelligent Controller (RIC), as well as we outline the role of the O-RAN components involved in the AI/ML loop. Combining the high-level architectural descriptions with a detailed presentation of ML-empowered resource allocation schemes, the paper discusses and summarizes the O-RAN disaggregation principles and the role of AI/ML embedded in future O-RAN deployments.

Skin cancers occur commonly worldwide. The prognosis and disease burden are highly dependent on the cancer type and disease stage at diagnosis. We systematically reviewed studies on artificial intelligence and machine learning (AI/ML) algorithms that aim to facilitate the early diagnosis of skin cancers, focusing on their application in primary and community care settings. We searched MEDLINE, Embase, Scopus, and Web of Science (from Jan 1, 2000, to Aug 9, 2021) for all studies providing evidence on applying AI/ML algorithms to the early diagnosis of skin cancer, including all study designs and languages. The primary outcome was diagnostic accuracy of the algorithms for skin cancers. The secondary outcomes included an overview of AI/ML methods, evaluation approaches, cost-effectiveness, and acceptability to patients and clinicians. We identified 14 224 studies. Only two studies used data from clinical settings with a low prevalence of skin cancers. We reported data from all 272 studies that could be relevant in primary care. The primary outcomes showed reasonable mean diagnostic accuracy for melanoma (89·5% [range 59·7-100%]), squamous cell carcinoma (85·3% [71·0-97·8%]), and basal cell carcinoma (87·6% [70·0-99·7%]). The secondary outcomes showed a heterogeneity of AI/ML methods and study designs, with high amounts of incomplete reporting (eg, patient demographics and methods of data collection). Few studies used data on populations with a low prevalence of skin cancers to train and test their algorithms; therefore, the widespread adoption into community and primary care practice cannot currently be recommended until efficacy in these populations is shown. We did not identify any health economic, patient, or clinician acceptability data for any of the included studies. We propose a methodological checklist for use in the development of new AI/ML algorithms to detect skin cancer, to facilitate their design, evaluation, and implementation.

IntroductionThe International Federation of Medical and Biological Engineering created a multidisciplinary working group to discuss assessments of artificial intelligence and machine learning (AI/ML) applications in health care. Engineers, clinicians, and economists identified evidence generation as a critical topic. Heart failure (HF) was selected to investigate the available evidence on the clinical effectiveness and safety of AI/ML applications. Attention was paid to transparency of AI/ML methods and their data sources.MethodsA scoping review was conducted on AI/ML algorithms developed for the management of HF. A search for systematic reviews, scoping reviews, and meta-analyses published from 1976 to October 2022 was conducted in Embase, MEDLINE, and Scopus.ResultsOf 456 relevant publications, 21 papers were included in the final analysis. Most papers (10 systematic reviews, five meta-analyses, and six non-systematic or scoping reviews) included studies conducted in North America. No study was conducted in Africa. The healthcare setting was not clearly stated in approximately half of the studies. A lack of agreement was noticed regarding the quality assessment tools used among the reviews. The most common data source for AI/ML algorithms was electronic health records, but in some cases data sources were not reported. While deep learning emerged as the most common adopted methodology, covariates were not always included in the algorithm development. The review demonstrated that comparative assessment of algorithms requires further investigation, given the high variability in the comparator used (e.g., clinical gold standard, other AI/ML algorithms, or other statistical methods). The main investigated endpoints were the incidence of HF and the number of hospital admissions.ConclusionsWhen assessing innovative health technologies such as AI/ML applications in health care, evidence is among the main challenges. Our scoping review, focusing on algorithms developed to manage HF, showed that the biggest challenges relate to the quality of the studies, the adoption of a comparative approach, and transparency of methods.

Artificial intelligence and machine learning (AI/ML) have grown in popularity in recent decades as a result of advances in high-performance computing (HPC) and open-source software. Earth science research has benefited from these advancements and continues to do so. However, entry-level AI/ML projects frequently have a significant level of complexity that prevents them from being realized. Furthermore, HPC workflows oftentimes require specialized knowledge for the installation and management of software dependencies. Our group has been working on closing this gap, particularly focused in remotely sensed applications, by leveraging high performance computing containers to make code portable across environments. In this paper we present the design and implementation of a set of containers to both speed up and optimize the development of remotely sensed data applications. These containers have been made publicly available for the use of the science community and include common AI/ML frameworks and hardware acceleration libraries. Here we present several use cases and experiments of how containers can be leveraged to speed up the development and deployment of AI/ML software and some examples of containerized applications performance.

Immuno-oncology involves the study of approaches which harness the patient’s immune system to fight malignancies. Immuno-oncology, as with every other biomedical and clinical research field as well as clinical operations,...

Clinical medicine requires the integration of various forms of patient data including demographics, symptom characteristics, electrocardiogram findings, laboratory values, biomarker levels, and imaging studies. Decision-making on the optimal management should be based on a high probability that the envisaged treatment is appropriate, provides benefit, and bears no or little potential harm. To that end, personalized risk-benefit considerations should guide the management of individual patients to achieve optimal results. These basic clinical tasks have become more and more challenging with the massively growing data now available; artificial intelligence and machine learning (AI/ML) can provide assistance for clinicians by obtaining and comprehensively preparing the history of patients, analysing face and voice and other clinical features, by integrating laboratory results, biomarkers, and imaging. Furthermore, AI/ML can provide a comprehensive risk assessment as a basis of optimal acute and chronic care. The clinical usefulness of AI/ML algorithms should be carefully assessed, validated with confirmation datasets before clinical use, and repeatedly re-evaluated as patient phenotypes change. This review provides an overview of the current data revolution that has changed and will continue to change the face of clinical medicine radically, if properly used, to the benefit of physicians and patients alike.

e14028 Background: For 70 years, the best indicator of glioblastoma (GBM) survival has remained age at diagnosis. Factors across the entire genome affect every aspect of the disease. But typical artificial intelligence and machine learning (AI/ML) would require 3B-patient training sets to generate predictive models from the whole 3B-nucleotide genome. As a result, all other attempts to associate a tumor’s DNA copy-number alterations (CNAs) with the patient’s outcome failed. Methods: A genome-wide pattern of DNA CNAs in primary GBM tumors was recently validated in a retrospective clinical trial as the most accurate and precise predictor of survival and response to treatment [doi: 10.1063/1.5142559 ]. Applicable to the general population, this biomarker, the first to encompass the whole genome, and biomarkers in lung, nerve, ovarian, and uterine cancers, were repeatedly identified in open-source datasets from as few as 50–100 patients by using our data-agnostic unsupervised AI/ML, which extends the mathematics of quantum mechanics to overcome the limitations of typical AI/ML [doi: 10.1063/1.5099268 , 10.1073/pnas.0530258100 ]. Results: At 75–95% concordance, our biomarker is more accurate than and independent of age and all other indicators, including the one-gene tests for MGMT, IDH1, and TERT. Platform- and reference genome-agnostic, the biomarker’s >99% precision is greater than the community consensus of <70% reproducibility. It describes disease mechanisms and identifies drug targets and combinations of targets to sensitize tumors to treatment. Now, in follow-up results from the trial we, first, show correct prospective prediction of the outcome of the five of the 79 patients who were alive four years earlier, at the time of first results (log-rank P-value=3.9×10–2). Two patients, who were predicted to have shorter survival, lived less than five years from diagnosis, whereas of the three patients predicted to have longer survival, one lived more than five, and the remaining two are alive >11.5, years from diagnosis. Second, we demonstrate 100%-precise clinical prediction for the 59/79 patients with remaining tumor DNA, by using whole-genome sequencing in a Clinical Laboratory Improvement Amendments (CLIA)/College of American Pathologists (CAP) laboratory. Third, we establish that the risk that a tumor’s whole genome confers upon outcome, as is reflected by the biomarker’s univariate Cox hazard ratio of 4.2 and Kaplan-Meier median survival difference of 2.25 years (log-rank P-value=6.0×10–4), is greater than that conferred by the patient’s Karnofsky performance score and access to chemotherapy, and the tumor’s percent resection, and is surpassed only by the patient’s access to radiotherapy. Conclusions: This is a proof of principle that our AI/ML is uniquely suited for personalized medicine.

Background and rationale There is widespread enthusiasm to improve health through the application of artificial intelligence and machine learning (AI/ML) methods to large population-level health datasets. Achieving this may require successful collaboration between institutions as well as between computer scientists (CS), machine learning researchers (MLR) and health service researchers (HSR).
 Main Aim Describe lessons learned in creating the Health Artificial Data and Analysis Platform (HAIDAP) in Ontario, Canada.
 Methods/Approach A partnership between a HSR institute (ICES), an AI/ML institute (Vector) and a high-performance computing center (HPC4H) was initiated in 2017 to enable the application of AI/ML methods to population-level health data for the province of Ontario (population 14M). We describe lessons learned (and being learned) following the HAIDAP’s launch.
 Results The HAIDAP was launched in 2019. Major learnings include: 1) importance of institutional partnerships and alignment with institutional strategies; 2) potential of joint institutional risk-sharing models; 3) need for scientific collaborations bridging disciplines around joint research projects; 4) sensitivity to different scientific cultures (e.g., academic prestige of conference proceedings for MLR vs journal publications for HSR; traditional statistical vs. ML model assumptions); 5) differences in research timeline expectations; 6) different experience with and expectations for access to de-identified routinely collected data (e.g., need for research ethics committee project approvals and privacy impact assessments); 7) developing data access models that enable greater flexibility (e.g., importing code or using open source tools); 8) broadening data access models to allow modern high-dimensional exploratory data analysis; 9) obtaining support of information/privacy regulator; 10) the hardware is the (relatively) easy part compared to other success factors.
 Conclusion The HAIDAP has enabled multi-disciplinary collaborations and novel AI/ML research of Ontario’s population-level health data. Collectively we have learned that additional effort is required to develop systems and processes enabling more efficient access to data and analytic tools for the analysis of administrative health data.

<div class="section abstract"><div class="htmlview paragraph">This paper introduces an innovative digital solution for the categorization and analysis of fractures in Auto components, leveraging Artificial Intelligence and Machine Learning (AI/ML) technologies. The proposed system automates the fracture analysis process, enhancing speed, reliability, and accessibility for users with varying levels of expertise.</div><div class="htmlview paragraph">The platform enables users to upload images of fractured parts, which are then processed by an AI/ML engine. The engine employs an image classification model to identify the type of fracture and a segmentation model to detect and analyze the direction of the fracture. The segmentation model accurately predicts cracks in the images, providing detailed insights into the direction and progression of the fractures.</div><div class="htmlview paragraph">Additionally, the solution offers an intuitive interface for stakeholders to review past analyses and upload new images for examination. The AI/ML engine further examines the origin of the fracture, its progression pattern, and the structures formed during the fracture’s development. The results are stored as attributes within the system, facilitating easy review and future analysis.</div><div class="htmlview paragraph">A notable feature of the system is its possibility to learn from user feedback. If the model provides an incorrect failure recommendation, users can correct it. These corrections can be used to retrain the model, improving its accuracy over time. This feedback loop will ensure that the system continually evolves and adapts to new data, enhancing its reliability and performance.</div><div class="htmlview paragraph">This innovative approach not only streamlines the fracture analysis process but also ensures that the insights gained are comprehensive and easily accessible. By integrating advanced AI/ML techniques, the solution represents a significant advancement in the field of fracture analysis, offering a robust tool for both novice and experienced users. The system’s adaptability and user-friendly interface make it an asset for improving the accuracy and efficiency of fracture analysis subsequently deskilling the judgement process for failures in Auto Components.</div></div>

Artificial intelligence and machine learning (AI/ML) can be used to automatically analyze large image datasets. One valuable application of this approach is estimation of plant trait data contained within images. Here we review 39 papers that describe the development and/or application of such models for estimation of stomatal traits from epidermal micrographs. In doing so, we hope to provide plant biologists with a foundational understanding of AI/ML and summarize the current capabilities and limitations of published tools. While most models show human-level performance for stomatal density (SD) quantification at superhuman speed, they are often likely to be limited in how broadly they can be applied across phenotypic diversity associated with genetic, environmental, or developmental variation. Other models can make predictions across greater phenotypic diversity and/or additional stomatal/epidermal traits, but require significantly greater time investment to generate ground-truth data. We discuss the challenges and opportunities presented by AI/ML-enabled computer vision analysis, and make recommendations for future work to advance accelerated stomatal phenotyping.