This week, we were extremely delighted to dive into the topic of artificial intelligence in radiology with Brendan Kelly. With a passion for AI, Kelly is currently an AI and Paediatric Radiology Fellow at Great Ormond Street Hospital for Children NHS Foundation Trust, a 2024 NDTP Dr. Richard Steevens Fellow, and an ESOR Fellow in AI and Paediatric Radiology. In addition to these impressive accolades, he also serves as the Deputy Editor (Social Media) at the European Society of Radiology’s flagship journal, European Radiology. Now let’s talk AI!
What is your background/experience with artificial intelligence and what first attracted you to the topic?
I had the fortunate opportunity to pursue an integrated PhD alongside my radiology training, thanks to funding from the Wellcome Trust and the Health Research Board of Ireland through the ICAT program. When I embarked on my PhD in 2019, artificial intelligence was the hot topic in radiology research, making the choice clear. This period allowed me to learn coding and spend a semester at the Center for Artificial Intelligence in Medicine and Imaging at Stanford. Considering that less than 50% of the world has access to safe and expert radiology, I was drawn to AI not only because it is an exciting and cutting-edge field but also because it has the potential to democratize access to medical imaging and healthcare, extending these services to underserved populations.
What are the biggest challenges to AI adoption in clinical practice?
The challenges to AI implementation in clinical practice are multifaceted. A primary issue is demonstrating improved outcomes with AI. While in silico results are impressive, translating these successes to the clinic remains a hurdle. Another significant challenge is determining who bears the cost of AI and who is accountable for errors. Further research and stakeholder engagement are essential to address these issues. However, if AI can be shown to concretely improve patient outcomes in prospective clinical trials, I believe solutions for these other challenges will follow. Additionally, ethical considerations and patient acceptance are critical. Clinical radiologists must advocate for interventions that genuinely benefit patients while remaining vigilant against those with uncertain advantages.
Give us an example (or an educated guess) of what you think AI will be able to do in 3 years. What about in 10 years?
AI is already being used in some clinical scenarios. In the next three years, I foresee the adoption of AI for “upstream” tasks, such as faster MR image reconstruction, dose reduction in CT scans, and optimizing patient selection for personalized follow-ups or identifying redundant procedures. Predicting AI’s capabilities in 10 years is more challenging, given the rapid advancements we’ve seen over the past decade. While Geoffrey Hinton predicted the end of radiology within five years back in 2016, radiologists are now more in demand than ever. However, I believe Professor Curtis Langlotz’s prediction from Stanford will come true: AI won’t replace radiologists, but radiologists who use AI will replace those who do not.
As a radiologist, what is your advice for younger colleagues wanting to dive into the topic of Artificial Intelligence, Machine/Deep Learning, and/or Radiomics? What are your tricks for staying up to date in this fast-evolving field?
As a radiologist who codes, I don’t think it’s necessary or advisable for all radiologists to have an in-depth understanding of machine learning models. However, as AI becomes more prevalent in clinical environments, a basic understanding is essential. I view AI as our generation’s MRI—a new technology that can benefit our patients, but only if we learn to use it properly. Radiologists must learn enough to have meaningful conversations with AI developers or vendors and to discern hype from genuinely beneficial products. Staying up to date in this fast-evolving field requires continuous learning and engagement with the latest research and technological advancements.