The Cardiovascular and Interventional Radiological Society of Europe (CIRSE) 2022 annual meeting (10–14 September, Barcelona, Spain) saw presenters explore artificial intelligence (AI), its current role in interventional radiology (IR), and how the two will evolve together to shape innovative practice in a session named “IR: The most tech of the medical specialties”. Among the key takeaways was reassurance that AI will not “replace” human interventional radiologists, rather that it is a question of understanding how AI can be useful in IR, learning how best to harness it as a tool for improving practices, and combining it with the knowhow and abilities of interventional radiologists to achieve ‘hybrid intelligence’.
Discussion of virtual reality (VR) and augmented reality (AR) also featured in the session, as well as wider commentary on what it takes to become a technological innovator in the IR space.
AI can lead to “truly patient-focused, individualised medicine”
The first speaker, Fernando Gómez Muñoz (Netherlands Cancer Institute, Amsterdam, The Netherlands; Hospital Clinic and Hospital Sant Joan de Déu, Barcelona, Spain), provided a comprehensive overview of how AI is currently being applied in IR and how he predicts this landscape will look in the coming years. Referring to the results for ‘artificial intelligence interventional radiology’ that a pubmed.gov search yields, the presenter demonstrated that AI in IR has been a topic of growing research focus—in 2016 there were 16 publications on the subject, and in 2021, 290.
Defining AI as “a combination of algorithms to provide machines with the capability to solve problems in a similar way to humans”, and machine learning as “a subtype of AI that uses computer algorithms to build predictive models,” Gómez Muñoz gave the example of how the latter can be used to predict type 2 endoleaks. The importance of this capability is, he shared with delegates, that “if you can predict [these], you can embolize the vessels that may cause the endoleak beforehand, avoiding a more complex procedure later, or an aneurysm rupture, which can happen if there is no adequate follow-up”.
Among the other applications of AI is in predicting local tumour progression post-thermal ablation, classifying peripheral arterial disease (PAD) in conjunction with computed tomography (CT) angiography, 3D planning for ablation procedures, as well as intraprocedural verification. The presenter then spoke to the advantages that AI offers for limiting unnecessary radiation exposure, such as during fluoroscopy-guided endoscopic procedures. Gómez Muñoz relayed his experience using verification software after ablation of liver tumours. It can “lead to early retreatment in some situations”, he explained, and can “dramatically reduce the recurrence rate of colorectal liver metastases”.
A further take-home message from Gómez Muñoz’s presentation was that he sees haptic perception in robots as a key area of future development with relevance to IR procedures. “Pressure, vibration, temperature and viscosity” are among the types of perception that the presenter hopes to see in robots used in IR settings, bringing the interventional radiologists who will make use of these innovations in a position to practise, increasingly, “truly patient-focused, individualised medicine”.
VR: the potential to reduce radiation exposure
Bradford Wood (National Institutes of Health Center for Interventional Oncology, Bethesda, USA) focused on VR in IR now, and its future directions. An important message he had for delegates was regarding the precise ways in which VR can benefit IR, which “can be poor” when reproducibility is concerned: by standardising practice and “equalising” the care patients receive. VR also has the potential to reduce interventional radiologists’ and patients’ radiation exposure through providing real-time mapping during cases, such as in conjunction with cone beam CT, to increase accuracy. This will also, in turn, reduce procedural complications and the time the given procedure takes. Related to this, VR can track patient movements so registration—the matching of images to the patient—remains accurate throughout the procedure.
However, Wood admitted that there are hurdles to adopting VR technologies in an IR practice. For example, goggles—a common means of accessing VR for use in a procedure—can bring on cybersickness, which discourages its use. However, handheld and projector-based tools can be a way around this ergonomic issue—Wood cited a smartphone app that can aid precise needle placement, even for interventional radiologists who are less experienced. Cost-effectiveness, Wood explained, is another barrier to healthcare providers incorporating VR into IR procedures, even though the benefits of increased accuracy and therefore reduced radiation exposure and intraprocedural complications, as Gómez Muñoz outlined, are highly valuable.
Technological innovations at lower cost
Luigi Solbiati (Humanitas University, Milan, Italy) gave his presentation on technological innovations in IR, what it takes to get them off the ground, and the obstacles that innovators can come up against in trying to do so. Solbiati also covered the benefits and pitfalls involved in developing these types of innovations and what he foresees the next disruptive IR technologies being. In terms of the latter, Solbiati cited histotripsy as already showing significant “world-changing” potential as a non-invasive, non-thermal, non-ionising procedure, where high-intensity focused ultrasound (HIFU) has been less successful.
Image fusion technology, which came into being around 20 years ago, was an example Solbiati gave of game-changing technology innovation in IR—“nowadays, we have wonderful rooms where you can do [ultrasound, angiography, CT and cone beam CT] together”.
However, he was keen to acknowledge the lack of accessibility of these types of advancements throughout the world, due to the cost involved. Therefore, Solbiati concluded, “we should find alternative solutions”. Among these, he listed “the fusion of real-time ultrasound (US), contrast-enhanced CT and positron emission tomography (PET) in the US room,” which he qualified by stating that it is “a simple and significantly lower-cost alternative to interventional PET-CT rooms.”
The presenter also opined that robots that facilitate remote needle placement will be a crucial future development in AI for IR, and that ablation confirmation software that should become mandatory after every ablation.
Solbiati himself, as highlighted by his co-presenters, has successfully created technology that can thermally ablate liver tumours using AR alone, and no radiation. One of the ideas behind this is to solve the aforementioned issue of the high cost associated with installing sophisticated multifunction CT suites.
To round off the session, moderator Laetitia Saccenti (Hôpital Paris Saint-Joseph, Paris, France) asked a question on behalf of those who may be concerned about AI capabilities exceeding those of interventional radiologists, and thus reducing or eliminating the scope of human practitioners’ roles in IR. “You cannot change the way things are going, you just have to surf the wave […] the more you learn, the more useful [a tool] it will be,” Gómez Muñoz answered.
Wood reiterated that “[AI and VR] will not replace” interventional radiologists. Certainly, a key takeaway from all three speakers was that technology will evolve to complement the work of interventional radiologists; enhancing what they are able to achieve and far from rendering their role obsolete.