Available online 6 December 2022

Author links open overlay panelAbstract

The emergence of massively parallel yet affordable computing devices has been a game changer for research in the field of artificial intelligence (AI). In addition, dramatic investment from the web giants has fostered the development of a high-quality software stack. Going forward, the combination of faster computers with dedicated software libraries and the widespread availability of data has opened the door to more flexibility in the design of AI models. Radiomics is a process used to discover new imaging biomarkers that has multiple applications in radiology and can be used in conjunction with AI. AI can be used throughout the various processes of diagnostic imaging, including data acquisition, reconstruction, analysis and reporting. Today, the concept of “AI-augmented” radiologists is preferred to the theory of the replacement of radiologists by AI in many indications. Current evidence bolsters the assumption that AI-assisted radiologists work better and faster. Interventional radiology becomes a data-rich specialty where the entire procedure is fully recorded in a standardized DICOM format and accessible via standard picture archiving and communication systems. No other interventional specialty can bolster such readiness. In this setting, interventional radiology could lead the development of AI-powered applications in the broader interventional community. This article provides an update on the current status of radiomics and AI research, analyzes upcoming challenges and also discusses the main applications in AI in interventional radiology to help radiologists better understand and criticize articles reporting AI in medical imaging.

Introduction

During the last two decades, the rapid growth in the number of medical imaging examinations and the improvement of calculation capabilities led to a dramatic development of artificial intelligence (AI) in medical imaging [1]. AI is now implicated in all aspects of medical imaging whatever the modality or the organs concerned. Even if most published AI studies refer to diagnostic imaging, there is no doubt about the fact that AI has also a promising future in the field of interventional radiology (IR) [2]. Whether or not AI will overachieve or deceive is still unknown [3]. Bluemke et al. published a brief guide with nine essential questions to address when assessing an AI model in radiology [4] and similar recommendations were also published by Gong et al. [5]. These considerations pave the way for a more standardized and formalized way of publishing results. Essentially, this might alleviate concerns non-specialists feel about the use of specific AI-based solutions.

The purpose of this article was to provide an updated status of radiomics and AI research and upcoming challenges and also discusses the main applications in AI in IR to help radiologists understand and criticize articles reporting AI in medical imaging.

Section snippetsUp-to-date status of AI research and upcoming challenges

The emergence of AI models as useful tools in radiology results from a decade-long industrial effort. From the development of new computing chips to the construction of curated image repositories, this technological leap has been made possible by a conjunction of progresses at multiple levels.

First, the emergence of massively parallel yet affordable computing devices has been a game changer for research in the field. Whereas central processing units (CPU) provide a few dozens of compute units

Radiomics

Radiomics is a high-throughput data mining process used to discover new imaging biomarkers. It is a data-driven, hypothesis-free research field that consists of the extraction of large sets of quantitative imaging descriptors that can feed machine learning algorithms to find correlations with diagnostic, prognostic or predictive targets [11]. The exponential growth of radiomics research has been built on the premises that medical images contain biological information that cannot be analyzed by

Main applications of AI in diagnostic imaging

AI can be used throughout the various processes of diagnostic imaging acquisition, reconstruction, analysis and reporting. AI has a potential to impact all the various steps of the daily radiological workflow, helping radiologists dealing with a constantly increase in workload [23]. Studies report that, usually, an average radiologist must interpret one image every 3–4 seconds in an 8-hour workday to meet workload demands [24]. Therefore, errors are inevitable, especially under such constrained

Main applications in interventional radiology

Most deep learning models are based on large datasets, which are commonly available in diagnostic imaging. Unfortunately, IR does not generate ready-to-use multicentric labelled images, and this may contribute to the idea that IR might be less suited for AI applications. This is a question of perspective, and IR should not be opposed to diagnostic imaging. It should rather be compared to other interventional specialties such as surgery or endoscopy. With this viewpoint, IR becomes a data-rich

Conclusion

AI may enhance the future of radiology throughout every aspect of our daily patient care [1,3,15]. The recent structuration of research in AI for imaging enables a more rigorous development and evaluation of AI-powered solutions, a mandatory turning point in this new field.

CRediT authorship contribution statement

Tom Boeken: Conceptualization, Methodology, Investigation, Writing – original draft. Jean Feydy: Conceptualization, Methodology, Investigation, Writing – review & editing. Augustin Lecler: Conceptualization, Methodology, Investigation, Writing – review & editing. Philippe Soyer: Conceptualization, Methodology, Investigation, Writing – review & editing. Antoine Feydy: Conceptualization, Methodology, Investigation, Writing – review & editing. Maxime Barat: Conceptualization, Methodology,

Declaration of Competing Interest

The authors declare that they have no known competing financial or personal relationships that could be viewed as influencing the work reported in this paper.

Human rights

The authors declare that the work described has been performed in accordance with the Declaration of Helsinki of the World Medical Association revised in 2013 for experiments involving humans.

Informed consent and patient details

The authors declare that this article does not contain any personal information that could lead to the identification of the patients.

Funding

This work did not receive any funding

Contribution of authors

All authors attest that they meet the current International Committee of Medical Journal Editors (ICMJE) criteria for Authorship.

View full text

© 2022 Société française de radiologie. Published by Elsevier Masson SAS. All rights reserved.