Blog
23 August 2024
Navigating the Intersection of Radiology and AI: Insights from Dr Stavroula Kyriazi
The integration of artificial intelligence (AI) into medical practices has become an increasingly important topic. As technology advances, healthcare professionals face new opportunities. At Unilabs, AI solutions are enhancing the detection of over 100 different conditions, including brain aneurysms, tumours, and hemorrhages.
In radiology, the use of AI is set to revolutionise diagnostic processes, enhance productivity, and ultimately improve patient outcomes. However, this promise brings considerations such as managing expectations, evolving attitudes towards AI, and balancing human expertise with technological assistance.
Unilabs’ Dr Stavroula Kyriazi, a consultant radiologist specialising in emergency radiology and gynaecological imaging, shares her insights and perspectives on digital health technology and AI applications. Since April 2023, she has been the clinical lead for the emergency radiology section within the Unilabs AI Centre of Excellence, exploring and implementing AI solutions to improve radiologists' performance and facilitate their workflow.
Could you elaborate on the importance of managing expectations regarding AI in radiology and its implementation?
There was a big hype when AI was first implemented in radiology, leading to doomsday predictions of imminent replacement of human radiologists by algorithms. Nowadays, most voices in the field agree that AI will likely become radiologists' "amiable apprentice" rather than their "awful adversary". The prevailing attitude is one of synergy (Does the combined performance of human plus machine outweigh each separate performance?) rather than competition. As radiologists, we want AI to enhance the accuracy of our reports and reduce repetitive, time-consuming tasks that burden our workflows. We need solutions that add value to our accuracy and efficiency.
Has your attitude towards AI changed since you started working with AI in radiology? Do you also notice this in your colleagues?
Attitudes tend to change as hands-on experience with AI increases. Personally, in my initial exposure to AI I tended to consider it as almost infallible. As I became more familiar with each AI solution, I gained confidence in critically interacting with AI and using it to its full potential. On the other hand, colleagues who were initially overly sceptical of AI have come to embrace its positive impact on their individual practice.
In what ways do you see AI complementing human interpretation in radiology? Could you discuss how collaboration between AI and radiologists enhances patient outcomes?
You can imagine AI as a second pair of eyes. It behaves similarly to young radiology registrars, eager to assist by flagging any potential abnormalities. This functionality can enhance the ability of the human eye to detect subtle abnormalities, positively impacting the quality of reporting. A negative AI output can reassure the radiologist that the finding in question has not been inadvertently missed. However, current AI models target narrowly defined findings in a single modality, lacking the ability to synthesise these findings into complex diagnostic interpretations, a cognitive task which to date remains the prerogative of the trained human radiologist.
For the last AI solution, a special focus group was created and you were highly involved. Could you walk us through the process and the outcomes of involving a broader group in testing and feedback before rollout?
At the AI Centre of Excellence, we have introduced a stepwise approach to implementing new AI solutions. Before group-wide implementation, priority access is granted to a focus group of five to six dedicated radiologists, who have the task of reporting system failures, providing early feedback, highlighting areas where fine-tuning would be desired, and helping colleagues navigate the solution. This practice has enabled us to increase radiologists' engagement with AI, which is highly important to ensure we explore its full potential.
Radiologists remain 100% responsible despite the assistance of AI. Can you elaborate on the role of human judgment and oversight in conjunction with AI technology?
Liability for all radiological reports lies entirely with the radiologist, regardless of any use of AI assistance. Therefore, it is imperative that radiologists remain aware of automation bias, or the tendency to over-rely on AI outputs and relinquish human oversight. In our practice, we prioritise training for radiologists before and after AI implementation to ensure they are empowered to use AI responsibly and thoughtfully.