Review of the scientific publication by Inria/Liryc researchers, in the journal Nature Reviews Cardiology, which discusses the clinical questions in cardiovascular imaging that AI can be used to address, the principal methodological AI approaches that have been developed to solve the related image analysis problems and the remaining limitations of AI approaches in cardiovascular imaging


Maxime Sermesant1, Hervé Delingette1, Hubert Cochet2, Pierre Jaïs2 and Nicholas Ayache1

1 Inria, Université Côte d’Azur, Sophia Antipolis, France
2 IHU Liryc, CHU Bordeaux, Université Bordeaux, Inserm 1045, Pessac, France
https://doi.org/10.1038/ s41569-021-00527-2REVIEWSNATURE REVIEWS | CARDIOLOGY

Research into the application of AI to medical imaging is now very active, especially in the field of cardiovascular imaging because of the challenges associated with acquiring and analysing images of this dynamic organ. Over the past decade, multiple proof- of- concept studies have outlined the extraordinary potential of AI to transform the way that cardiac images are prescribed, acquired, reconstructed, analysed and used to tailor patient care.


Cardiovascular imaging has several distinctive characteristics that can be viewed as both challenges and opportunities for AI: it combines structural and functional information about a patient’s physiology; it is redundant, agile and ever-changing given the wide variety of modalities and techniques to assess the heart and the very rapid developments in image acquisition and reconstruction; it is computationally demanding given the multiple scales on which the organ can be analysed; and it directly affects patient management given the broad spectrum of cardiac interventions developed over the past two decades, all of which require images for patient selection, preoperative planning or even direct intraprocedural guidance.


The implementation of AI in predictive medicine is a far more attractive prospect but will certainly require more time and validation efforts, particularly to ensure generalizability. The interactions between AI and pre-existing knowledge of anatomy, physiology, biophysics and experts opens up many possibilities. We are not there yet and, for this objective to be achieved, AI will have to meet the highest standards and, perhaps even more importantly, gain the trust of both patients and clinicians.