Mapping the Bibliometrics Landscape of AI in Medicine: Methodological Study (Preprint)

Abstract

BACKGROUND Artificial intelligence (AI) was coined in the 1950s. Since then, AI has made its way into many industries in several waves and has become more mainstream in parallel with the increase in hardware’s computing power. However, AI in medicine lags behind other industries. In recent years AI in medicine has gained massive attention among researchers and practitioners, and thus, the studies regarding medical AI have shown expositional growth. OBJECTIVE Given that there was no current framework as a blueprint guideline, this research aimed to map the current research status and future trends in medical AI by screening all the AI-related studies within PubMed in the past two decades. This article also provided possible approaches to acquiring data and analyzing it via Python in similar studies in the future. METHODS Our approach was twofold: (1) we obtained the metadata of the publications on AI from PubMed between 2000 and 2021 via Python, including titles, abstracts, author names, journal names, and publishing year, and then conducted the keyword occurrences counting; and (2) we classified the relevant topics by unsupervised machine learning approach – latent Dirichlet allocation (LDA) and identified the research scope on AI in medicine. As there is no universal medical AI taxonomy, we generated an AI dictionary based on the European Commission Joint Research Centre (JRC) AI Watch report. RESULTS A total of 249,243 eligible publications were included in this research. The result showed exponential growth of medical AI research on PubMed. Articles in the (machine) learning domain constantly account for half of all AI research. The number of research in communication, integration and interaction, and service has also increased over the past years, but the overall shares remain similar during the past decades. In the reasoning domain, research articles increased every year. However, the share dropped significantly. AI ethics appeared to be the new research area based on increased published articles. CONCLUSIONS Along with the increasing number of studies on AI each year, the research topics have shown different trends. Machine learning was the focus of medical AI research and would still be the center of future research. Deep learning will continue to play a crucial role in AI in medicine. With the help of predictive algorithms, pattern recognition power, and imaging analysis capabilities, research regarding a medical diagnosis, robotic intervention, and disease management would likely be the future focuses.

Evaluating global data policies around non-personal data on social and public good

“…There are several other use cases in different sectors such as agriculture, manufacturing, telecom services as well as energy and process industries where NPD application has already added immense public and economic value and has the potential to add even more. However, studies have pointed out that the standards to make this assessment have to be improved and based on principles of democratic governance, deliberation, fairness and public justification among others (Züger and Asghari, 2023; Umbrello and Van de Poel, 2021; Floridi et al , 2020; Shi et al , 2020; Leslie, 2019; Taddeo and Floridi, 2018; Hager et al , 2017). For instance, in the context of AI, Züger and Asghari (2023) argue based on the theory of public interest that for any technology to be considered serving the public good, it must have a democratic justification agreed to upon by the public, must ensure equality and safeguard human rights and follow a process of deliberation accommodating different interests.…”

Beyond ‘AI for Social Good’ (AI4SG): social transformations—not tech-fixes—for health equity

“…Such technical processes may be the basis for both AI4SG projects as well as AI and big data operations in society more generally. While a fuller discussion of AI techniques and their nuances is beyond the scope of this paper, there is evidence that machine learning capabilities are increasingly prominent in AI4SG efforts across domains, especially in healthcare (Shi, Wang, and Fang 2020, 7).…”

“…A 2018 McKinsey AI4SG white paper also found the greatest number of actual or potential use cases in its ‘Health and hunger’ domain, comprising 28 of 160 use cases (Chui et al 2018). Meanwhile the AI4SG literature survey mentioned above found more papers in the ‘Healthcare’ sector (comprising both clinical care and public health) than in any of the eight application domains it identified (Shi, Wang, and Fang 2020, 5–6), with this sector accounting for 32% of the 2019 AI4SG literature surveyed, as well as having the highest rate of growth in recent years, such that ‘the difference between it and other domains appears to be widening’ (Shi, Wang, and Fang 2020, 5). Articulating their enthusiasm for leveraging big data for public health, Muin Khoury and John Ioannidis write, ‘Big Data stands to improve health by providing insights into the causes and outcomes of disease, better drug targets for precision medicine, and enhanced disease prediction and prevention’ (Khoury and Ioannidis 2014, 1054).…”

“…The medical model, conceptually and practically, emphasizes downstream, individualized interventions in health care settings, increasingly wedded to emerging precision medicine treatments, for which big data and AI are enabling technologies (Ho et al 2020; Hager et al 2017; Schork 2019; Sun et al 2018). As noted in the AI4SG literature survey, ‘The majority of research efforts in AI for healthcare are focused on clinical care’, citing the availability of EHRs and medical images as having ‘directly facilitated the AI research in disease diagnosis, clinical treatment, and clinical prediction’ (Shi, Wang, and Fang 2020, 20). Among these applications, ‘Disease diagnosis is the topic in healthcare that has seen the most applications of AI’ (again, with this survey including both clinical care and public health in its ‘healthcare’ category; Shi, Wang, and Fang 2020, 20).…”

“…While AI4SG projects are often just another way to frame the AI activities of conventional profit-oriented business entities, the formal field of AI4SG is gathering momentum, basking in the glow of an apparent ‘AI summer’. A recent survey of the field found over 1000 published papers on AI4SG topics (Shi, Wang, and Fang 2020, 1), growing from 18 papers in 2008 to 246 papers in 2019 (Shi, Wang, and Fang 2020, 5). AI4SG projects might identify endangered species in digital image streams to aid conservation efforts; they might analyze satellite data to monitor manifestations of climate change, such as sea-level rise or desertification; or they might support health diagnostics, for instance, by detecting skin cancers based on mobile phone photos (Chui et al 2018).…”