The presenter of this year’s ESC Paul Hugenholtz Lecture in Innovation is Professor Charalambos Antoniades (University of Oxford - UK) who will discuss his work developing AI models that predict cardiovascular events years before they happen.
“We have the ability to interrogate cardiac CT images using AI, to predict not only heart attacks and cardiac death, but also stroke, heart failure and other conditions in a timely fashion, allowing us to intervene with preventive strategies. And it is worth noting that the first step on the journey was a basic science discovery. More than a decade ago, we observed that signals released from the inflamed coronary artery diffuse to the perivascular adipose tissue, inhibiting local adipogenesis [1]. This process changes the texture and composition of perivascular fat around inflamed arteries, shifting its attenuation on cardiac CT from the lipid to the aqueous phase. We developed an imaging biomarker of coronary inflammation, the perivascular fat attenuation index (FAI) score and were able to confirm it identifies high-risk patients from routine cardiac CT scans better than any other tool available in practice, as shown in the CRISP-CT study [2].
In the large ORFAN study, we showed that an AI-enhanced prognostic model – the AI-Risk algorithm – which incorporates FAI score, the extent of coronary atheroma (if any) and traditional risk factors, could powerfully predict cardiovascular mortality and MACE over 10 years, both in the presence and absence of coronary atherosclerosis [3]. Since the introduction of AI-Risk, we have developed more advanced models based on agentic AI, one of which enables clinicians to identify not only the presence of inflammation but also the type of inflammation. This tool, which is being investigated in randomised trials, may inform the type of medication a patient is most likely to respond to. Another exciting development is an AI-driven tool for the measurement of myocardial inflammation, which can facilitate identification of individuals likely to develop heart failure at least 5 years before symptoms appear [4].
The FAI score and AI-Risk technology are now part of a medical device that is being used within the UK’s NHS, changing the management in approximately half the individuals undergoing cardiac CT scans, while other AI tools in development are expected to completely transform the way we interpret medical images in the clinic. However, introducing AI into practice also brings several risks. These include clinician de-skilling due to overreliance on technology as well as the inability to evaluate how these models work as they are essentially operating as ‘black boxes‘. In addition, given the unprecedented rate of development, models must be updated regularly to ensure they remain reliable over time, and there is the need for a strict regulatory framework that will enable safe use and prospective surveillance, maximising their impact on routine clinical care.”
