A recent editorial, published in European Radiology, has highlighted the challenges and solutions for introducing artificial intelligence (AI) into daily clinical workflow.
AI is ubiquitous
The applications of AI are increasing. A particular area of application that has shown great promise is radiology. In 2019, the number of AI-related abstracts submitted for radiology reached 25%. Many of these publications focused on AI tools, particularly deep learning and its ability to identify patterns within medical image data. While there is a lot of hype around AI, some major bottlenecks still exist in the introduction of these algorithms into diagnostic and routine clinical practice. These include:
- The availability of high-quality annotated datasets required for training.
- An algorithm’s ‘black box’ nature makes it difficult to prove algorithm robustness and reliability once it has been trained.
- Lack of standards for data sharing between digital systems.
AI and radiology
Good examples of where AI algorithms would most support radiologists include the detection and follow-up of lung nodules in CT scans. These are typically repetitive and time-consuming tasks for radiologists. However, for radiology departments, it is very difficult to negotiate contracts for a variety of AI systems and to implement them into all IT environments. This becomes even more difficult when hospitals want to use their own algorithms.
The team believe that non-interpretative AI applications, such as patient scheduling, will be introduced into clinical practice before diagnostic AI algorithms. Some of these applications are already commercially available.
In a lot of non-scientific media, the idea of AI replacing workers like radiologists has circulated. Unlike algorithms, radiologists are experts in interpreting images from a wide variety of diseases and making decisions based on the whole patient. They are also able to integrate information from a variety of sources. Therefore, their role is indispensable in finding the best therapeutic option for their patients. The authors of this article believe that AI will complement radiologists and help enhance their performance, rather than replace them.
Implementing AI into healthcare raises many ethical questions. Radiologists must be actively engaged in the development of ethical and regulatory guidelines for the use of AI tools in radiology. Experts must consider the potential built-in bias of algorithms as they may cause unforeseen harm to patients. To detect such significant errors, the overall integration of AI systems will require standardised and regulated monitoring of outcomes. The team believe that AI should respect human rights and freedom. They suggest that it should be designed for maximum transparency and dependability. Ultimately, responsibility and accountability of AI will lie with those who use or have designed it.
In the next few decades, more systems will become available that allow us to observe new phenomena. Despite concerns of threatened progression, the younger generation seem aware that AI will actually improve radiology rather than replace humans. The team emphasise that education on AI for radiologists is important and should include both technical and ethical aspects of AI. The next generation of radiologists should be equipped with basic knowledge underlying these techniques. The team believe that AI will help strengthen radiologists’ function and give them an indispensable role in personalised medical care. As a result, it is key that radiologists become active in this ongoing transformation.
Image credit: By Stephanie Noiret – canva.com