Dr. Jennifer Soun, MD, neuroradiologist at UCI Health, talks about artificial intelligence helping to diagnose stroke quicker.
Interview conducted by Ivanhoe Broadcast News in September 2022.
Is stroke one of the leading causes of death?
SOUN: It’s the second most common cause of death worldwide. About 50% of patients who have stroke are left chronically disabled after having a stroke.
One of the key parts of treating a stroke is to treat it early, correct?
SOUN: Yes. The mantra in stroke treatment is time is brain. With every second that passes, several million neurons actually die. So, it is critical to treat as early as possible. And part of that is actually being able to diagnose the stroke promptly.
Are there other things other than a stroke that people may misdiagnose?
SOUN: Yes. Stroke can mimic other etiologies, other diagnoses. But every etiology has a very different treatment pathway. And stroke is certainly one of the main emphases of stroke treatment is treating promptly.
Are you starting to use AI to help you diagnose a stroke?
SOUN: Yes. Right now, we are using different AI tools that help with the acute stroke triage pathway. Our goal of this research is to hopefully improve door-to-treatment times by expediting the diagnoses and allowing our stroke treatment team to know whether or not a patient has a stroke and if they’re a candidate for treatment.
How does AI make that possible?
SOUN: AI uses different machine learning types of tools to do a variety of different tasks. In stroke, one of the main tasks is to first detect whether or not the patient has a stroke by looking at the imaging and figuring out whether it looks like a stroke or something else. But it also can help in other ways. It can help rule out other mimics such as hemorrhage or masses. And it can help diagnose something called a large vessel occlusion, which is actually a treatable type of stroke.
Do all of those things that you just mentioned take time to diagnose in the ER?
SOUN: Yes. The radiologist serves as part of the stroke treatment team, and what we do as radiologists is the diagnostic part of the stroke triage. It does take time because generally, a radiologist is reading from a busy work list. These AI tools actually help us by triaging the cases that might demand urgent attention or the positive cases that should go to stroke treatment.
How much quicker can AI help you do your job?
SOUN: The great thing about AI is that it’s pretty much automatic. It can deliver results within seconds. Many of these stroke AI tools are actually phone apps. We can get phone notifications whether there is a stroke or not, and as the radiologist, I can look at those images and verify whether the AI is correct and promptly alert the stroke team for intervention.
Do you even have to be on premise?
SOUN: You can be anywhere. And that’s one of the main impacts that AI can have in the stroke space is being able to reach a wide variety of people, a wide variety of clinical care settings – everything from rural areas to comprehensive stroke centers. And everyone can actually be connected through these AI tools.
Are there, like, 10 things that AI is looking at to diagnose whether someone is having a stroke or experiencing something else?
SOUN: AI tools, at least right now, usually, they answer a very specific question. Usually, it’s just one question for now. If the question is, “Is the patient having a stroke due to a large vessel occlusion?”, there are tools that can answer that question. There are separate tools that answer the question of, “Is the patient having a hemorrhage?”, or “Does the patient have some other disease process?” But right now, AI is very specific in terms of what it answers. It’s using a lot of data from the laboratory, from the testing sites to actually improve the algorithms for these tools so that they make more accurate diagnoses.
At UC Irvine, is every patient that comes in having a stroke, is their data being used for AI to learn?
SOUN: Yes. Actually, there are some commercially available AI stroke tools, and we are using a few of them in the clinical setting for patients who come in with acute suspected stroke. They will get imaging done routinely, and with that imaging, automatically the AI tool runs an algorithm to determine whether there’s a large vessel occlusion or not, whether there’s a hemorrhage or not, and then we are alerted about those results.
What kind of imaging are you talking about?
SOUN: For every suspected acute stroke patient, they typically get a CT scan as well as something called a CT angiography with contrast dye. That allows us to look for the stroke itself, or if the patient has hemorrhage, which may require a different treatment pathway, and then lastly, whether or not the patient has a clot in their brain that could be treated.
Is this currently a clinical trial then that’s happening here?
SOUN: We are doing clinical research that it’s more of a translational project where we’re implementing these tools not just in a laboratory setting but in a real world setting. And we’re comparing how these tools perform based on the different patient populations that we have and then also whether or not these tools actually make a difference in clinical outcomes.
How have you seen them make a difference?
SOUN: Right now, we’ve seen that these tools do improve the radiology workflow. We are able to diagnose strokes and hemorrhages faster than we were before implementing the tool. And the other thing is we are seeing that there are faster door-to-treatment times, and that’s really promising because every minute that passes it does affect the patient’s outcome.
Do you have any specifics on how much faster?
SOUN: Door-to-treatment times generally are on the scale of minutes. Right now, we are seeing a reduction in time within minutes compared to before the implementations.
When you’re talking minutes, what’s that mean for the patient’s brain?
SOUN: The profound impact of thrombectomy, which is the treatment for stroke due to a large vessel equation, is quite significant. The number needed to treat for a patient to benefit is less than three patients total, and that benefit comes and manifests as a better quality of life, being able to move, being able to speak. Those are often impairments that stroke patients have if they have an untreated stroke. These treatments allow for the potential to be able to regain those very important functions for quality of life. With faster diagnoses, that means there’s potential to save more brain.
We’re just seeing the beginnings of AI hit the medical world. But do you think this is just the start of diagnosis in the hospital across all fields?
SOUN: Yes. Actually, AI is not just radiology specific. There are several tools for other purposes, other specialties that are also using AI. So, yes. We are sort of at the beginning of the AI revolution in medicine. But I think in the near future, as these tools get better, people are more comfortable using them. They’re easier to access. I think with all of those things, we’ll see AI sort of permeate throughout the hospital system.
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
If you would like more information, please contact:
John Murray
Sign up for a free weekly e-mail on Medical Breakthroughs called First to Know by clicking here
