ReviewThe future of radiology augmented with Artificial Intelligence: A strategy for success
Section snippets
Background
Radiology is in need of a strategy to future-proof the profession. A diagnostic radiologist is a postgraduate subspecialty-trained medical doctor who is skilled in interpreting medical images such as Digital radiographs, CT scans, Ultrasounds, Nuclear Medicine studies and MRIs and using them to guide management of disease in patients. But recently, experts in Artificial Intelligence (AI) have warned that radiologists may soon be out of a job, one being none other than the grand master of deep
The current state of radiology and the need for a strategy
Radiologists are not unfamiliar with Artificial Intelligence, pioneering work in medical imaging perception in the 1980s [10]. We are domain experts in medical imaging, medical physics and radiation safety. But in the past 5–10 years, there have been substantial new innovations in imaging from deep learning methods of image classification. Current artificial neural networks have accuracy rates which surpass those of human radiologists in narrow-based tasks such as nodule detection [11,12].
The
General use cases, potential impact and implementation strategy
Broadly, several use cases should be targeted for implementation within the scope of radiology. They can be divided into task-based categories:
Impact upon cost leadership, differentiation and focus
One of the most obvious strategies to drive radiology forward is cost leadership. The integration of machine learning in imaging diagnosis has the potential to cut costs for patients and insurance companies by half [27]. It may cost as little as $1000 USD to install machine learning enabled chips capable of processing 260 million images per day [28]. Put into perspective, that is more than the sum of all MRI and CT scans performed in the USA daily. A thousand dollars is the current cost to
Defining roles, technical considerations and requirements for implementation
The individuals involved in implementing these initiatives include the Chief Information Officer (CIO) of each hospital, radiology leadership in committees and academic bodies such as professional colleges and societies, as well as individual radiologists.
The CIO’s role is to ensure that these initiatives can be implemented safely and effectively so that patient safety and privacy is not compromised, integration into existing electronic health data systems and alignment with the rest of the
Organizational aspects of implementation
The main people in charge of implementing these initiatives would be the hospital CIO and/or chief data officer, as well as the department chief at the line-managerial level. The hospital CIO’s duties would include ensuring that the systems are able to integrate into existing IT infrastructure, and purchasing these systems and updates.
The chief of radiology’s duty would be to ensure radiology staff are trained adequately to use these systems and that this new software would not pose a risk to
Roadmap for the implementation of AI in radiology
A few key areas can be automated with AI in the near future with machine learning technologies which already exist:
- 1.
Automated image segmentation, lesion detection, measurement, labelling and comparison with historical images. This technology has already been debuted on the commercial stage at the recent Radiological Society of North America (RSNA) annual meeting 2017 in Chicago.
- 2.
Generating radiology reports: most radiology reports are written in prose rather than in lists, necessitating long
Special considerations, job displacement and risk mitigation
Healthcare is one of the more lucrative business opportunities within most economies worldwide, perhaps even more so in developing nations with growing middle classes who can afford self-funded healthcare. Not surprisingly, medical imaging computing is the most published subject in the scientific literature amongst uses of deep learning in healthcare [34]. What this means is that it is likely, although not yet confirmed, that radiology will be the first medical specialty to be disrupted in the
Safety, privacy, moral and ethical concerns
These remain a large shadow looming over the implementation of AI in healthcare. Notwithstanding these concerns, recent signals from the US FDA portend that governments are keen to support AI technology adoption in the healthcare domain [36]. Borrowing from the Asilomar AI principles [37], the key ethical concerns for imaging AI are:
- 1.
Safety: this is a key imperative for medical AI systems which would be inextricably involved in safeguarding the health of sick individuals at their most vulnerable
Global radiology impact and global RADIOLOGISTS’ response
The impact of AI is beginning to send ripples throughout the international radiology community, dominating industry and academic headlines, as well as becoming sellout ‘standing-room only’ sessions at international radiology meetings. The impact in reading rooms has been more muted, with relatively few departments in academic centers and research institutes being involved in AI research and user acceptance testing. Notably, many efforts have focused on industry and regulation but more is
Conclusion
According to Porter's Generic Strategies model, Cost Leadership, Differentiation and Focus can be used to create a competitive advantage. The roadmap for the future of AI augmented Radiology is guided by the direction provided by these strategies: reduction of overall cost of imaging to the patient/payer by increasing the productivity of radiologists through the automation of time-consuming and low cognitive value tasks and by differentiating Augmented Radiology as the cornerstone of precision
Declarations of interest
None.
Acknowledgements
The author would like to thank the anonymous referees for their helpful comments. The author is also grateful to Dr Angeline Poh and Dr Andrew Tan from the Department of Diagnostic Radiology, Changi General Hospital, Singapore for their insights.
References (37)
A reflection on the 150 anniversary of the birth of Marie Curie
Appl. Radiat. Isot.
(2017)- et al.
Augmenting diagnostic vision with AI
Lancet
(2017) The future of image perception in radiology: synergy between humans and computers
Acad. Radiol.
(2003)- et al.
A hybrid CNN feature model for pulmonary nodule malignancy risk differentiation
J. X-Ray. Sci. Technol.
(2017) - et al.
The effects of changes in utilization and technological advancements of cross-sectional imaging on radiologist workload
Acad. Radiol.
(2015) Imaging 3.0: what is it?
J. Am. Coll. Radiol.
(2013)Machine learning and the market for intelligence
- E. Crooks, P. McGee, GE and Siemens: power pioneers flying too far from the sun, (n.d.)....
The O-ring theory of economic development
Q. J. Econ.
(1993)Why are there still so many jobs? The history and future of workplace automation
J. Econ. Perspect.
(2015)
How competitive forces shape strategy
Strateg. Plan. Read.
A new initiative on precision medicine
N. Engl. J. Med.
Radiomics images are more than pictures, they are data
Radiology
Deep learning in mammography
Invest. Radiol.
Adapting to artificial intelligence
JAMA
Integrated diagnostics the computational revolution catalyzing cross-disciplinary practices in radiology, pathology, and genomics
Radiology
Lung nodule detection using 3D convolutional neural networks trained on weakly labeled data
SPIE Med. Imaging
Cited by (157)
Technical/Algorithm, Stakeholder, and Society (TASS) barriers to the application of artificial intelligence in medicine: A systematic review
2023, Journal of Biomedical InformaticsThe top 100 most cited articles on artificial intelligence in radiology: a bibliometric analysis
2023, Clinical Radiology