Study on horizon-scanning with a focus on the development of AI-based medical products: citation network analysis

Abstract

ObjectivesHorizon-scanning for innovative technologies that might be applied to medical products and require new assessment approaches/regulations will help to prepare regulators, allowing earlier access to the product for patients and an improved benefit/risk ratio. In this study, we focused on the field of AI-based medical image analysis as a retrospective example of medical devices, where many products have recently been developed and applied. We proposed and validated horizon-scanning using citation network analysis and text mining for bibliographic information analysis.Methods and analysisResearch papers for citation network analysis which contain “convolutional*” OR “machine-learning” OR “deep-learning” were obtained from Science Citation Index Expanded (SCI-expanded) in the Web of Science (WoS). The citation network among those papers was converted into an unweighted network with papers as nodes and citation relationships as links. The network was then divided into clusters using the topological clustering method and the characteristics of each cluster were confirmed by extracting a summary of frequently cited academic papers, and the characteristic keywords, in the cluster.ResultsWe classified 119,553 publications obtained from SCI and grouped them into 36 clusters. Hence, it was possible to understand the academic landscape of AI applications. The key articles on AI-based medical image analysis were included in one or two clusters, suggesting that clusters specific to the technology were appropriately formed. Based on the average publication year of the constituent papers of each cluster, we tracked recent research trends. It was also suggested that significant research progress would be detected as a quick increase in constituent papers and the number of citations of hub papers in the cluster.ConclusionWe validated that citation network analysis applies to the horizon-scanning of innovative medical devices and demonstrated that AI-based electrocardiograms and electroencephalograms can lead to the development of innovative products.Article SummaryStrengths and limitations of this studyCitation network analysis can provide an academic landscape in the investigated research field, based on the citation relationship of research papers and objective information, such as characteristic keywords and publication year.It might be possible to detect possible significant research progress and the emergence of new research areas through analysis every several months.It is important to confirm the opinions of experts in this area when evaluating the results of the analysis.Information on patents and clinical trials for this analysis is currently unavailable.

Study on Horizon Scanning by Citation Network Analysis and Text Mining: A Focus on Drug Development Related to T Cell Immune Response

“…Thus, this study shows that citation network analysis and text mining using the methods of Sakata et al [16] can be used to understand new research trends not only in fundamental technologies applicable to various fields such as nanocarbon and AI [18,47], but also in biological fields elucidated over a long period, such as immunology.…”

“…This study is the first to show that this method can be used appropriately as a tool for horizon scanning in the medical field. Because it is difficult to extract a limited number of novel topics that may affect pharmaceutical regulations from a vast amount of information on a human basis, it is reasonable and appropriate to use a computer-based method such as the method used in this study as a primary screening [47].…”

“…We extracted scientific papers for citation network analysis from PubMed, which consists of more than 30 million citations to biomedical papers from MEDLINE, life science journals, and online books [48].We also tried to analyze papers extracted from WoS, which consists of 161 million citations across 254 disciplines beyond science, but the obtained papers contained only one of the six key articles, suggesting that the papers that represent the research trends in this field were not extracted as described in METHODS. When we applied this analysis method to the artificial intelligence (AI) field to investigate the R&D of AI-equipped medical devices, we obtained more useful information from WoS than from PubMed [47]. In the study of three-dimensional cell layering using this method, WoS provided information more information on the base material for adhesive culture and bioprinting equipment, while PubMed provided more information on cell functions (data not shown).…”

“…However, they aim to obtain a bird's eye view of the entire scientific field by the top 1% of papers in terms of citations using "co-citation". Sakata's method seems to be more suitable for our purpose since the method applies 'direct citation', which has been reported to be the most appropriate for obtaining leadingedge information on trends, to all papers extracted from the database [47].…”

Identification of Novel Modalities Through Bibliometric Analysis for Timely Development of Regulatory Guidance: A Case Study of T Cell Immunity

“…As shown in Supplementary Figure 1, after forming a citation network from PubMed publications, it was converted into an unweighted network with publications as nodes and citation relationships as links. The network was then divided into several clusters using the topological clustering method with modularity maximization (Louvain method) (21)(22)(23). Subsequently we computed the term frequency-inverse cluster frequency (TFICF) to extract the characteristic keywords of each cluster.…”