ISC 2026 | Where AI and large language models add value in stroke care

I think the immediate value of these tools is as adjuncts, so helpers, either to the patient or to the clinician or provider. They are not ready for autonomous use. I sort of have to say that because there isn’t a standard for them yet. I think that might come in the future, but at the moment they are adjuncts. So what do I mean by an adjunct? I mean that let’s talk about the patient first. There’s a lot of lingo and jargon in the way we describe CT scan results, in the way that we describe how treatment should be prescribed to prevent stroke or in rehab after a stroke or in conditions that predispose to stroke like atrial fibrillation and others. Chatbots can be very useful for the patient to decode that and to suggest the next best question. Your doctor, your nurse says this, what would be a next best step? They recommend therapy X or Y. Is there anything else I should ask? So I think that would be an obvious thing, which is an adjunct. It’s firmly in the realm of interacting with the healthcare provider. From the provider’s point of view, I think it’s the same, but on a more literature-based level. So we’re all human. We all forget things. We sometimes don’t take into account what the patient’s specific circumstances are. And so it will be very useful and is already being used to check that we’re not missing anything. So on the CT scan, is there any evidence of new lacunar events? Is there any evidence of cortical atrophy? Are there any other worrying signs? Based on a CT of this type, what would you recommend treatment path should be? Now, the reason I want to add an asterisk there, a caveat, is that although that is a good adjunct already and people are doing that, there are specific areas where a general large language model would go wrong. So if you fed that into GPT, I think you’d probably get a lot of false positives, a lot of potential errors. So for the provider, we have to be very careful that we use carefully curated, specialized tools. Such tools already exist, particularly for image analysis of the brain, but also actually different parts of the body. So, for example, there’s a lot of focus on large language models and AI to interpret chest CT scans and even other modalities. We have to use specific tools.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.