OSF HealthCare is expanding use of artificial intelligence [AI] to all its hospitals for potential stroke patients, including those at OSF St. Joseph Medical Center in Bloomington. The Rapid AI suite of tools had been in place at St. Francis Medical Center in Peoria and OSF St Anthony in Rockford. It’s now available at OSF St. Joseph Medical Center in Bloomington and at rural hospitals in the OSF system.
“Every community we serve should benefit from the same advanced technology, whether a patient enters one of our largest medical centers or a rural, critical access hospital,” said Dr. Arun Talkad, director of stroke care at OSF.
A spokesperson for Carle BroMenn Medical Center said that hospital began using Rapid AI in 2020.
Rapid AI also does automatic measurements of key indicators in the brain and offers other information a standard CT scan does not provide, what’s called CT perfusion, what the blood flow is really doing in the brain.
Talkad said it takes guesswork away with those more precise measurements instead of relying on doctor estimates of how wide an area a stroke affects.
“The AI actually says, ‘here's this area of stroke, say, 10 mL in volume. Here's the area that's at risk, which is 75 mL,” said Talkad.
'Time is brain'
In many cases neurologists look at their own image scans rather than wait for a radiologist to analyze them. With Rapid AI the resolution of the images is of high enough quality to read on a mobile phone, not just a computer at a major medical center, said Talkad.
This can speed up treatment.
“If I'm actually talking to the ER doctor, that information may be on my phone before the patient actually even even gets back to their room from the CT scanner. That's how efficient this is,” said Talkad.
And in strokes, physicians say, ‘time is brain.’
“It is the most important thing in the modern day of stroke,” said Talkad. “Every minute of stroke costs 1.9 million brain cells, and so if it takes an hour to process images, you've lost 20 million brain cells. That's the same amount of brain that we lose in 3.6 years of life, and so the cost measurement is really, frankly, incalculable.”
Treatment metrics used to be time-based. If a patient was more than three to four-and-a-half hours into a stroke, doctors thought it was past the useful window to administer a clot busting drug called TPA or TNK. Now, doctors can decide based on how much of the brain they see affected.
“Now, it doesn't matter that you presented at six hours or eight hours or 12 hours. If the CT images and the AI technology is saying that this is still an early stroke, then we can treat it,” said Talkad.
Between 85% and 87% of strokes are called ischemic strokes. Those involve blood clots. Talkad said Rapid AI is also useful in treating hemorrhagic strokes, or brain bleeds. Those actually have higher mortality rates than ischemic strokes. Hemorrhagic strokes may need surgery and if they do it needs to happen quickly.
“At a rural hospital where there is no surgery, there is no ICU capability and no neurologist, they have this hemorrhagic stroke,” said Talkad. “If I have this ability to use the rapid technology to look at that image right away, I can say that patient needs to be flown here… That patient needs to go first in line. Period.”
St. Francis Medical Center treated over 1,000 strokes last year. It has been a stroke center for 30 years covering the central part of the state from Indiana to Iowa. Talkad said they have historically been the busiest single hospital for stroke in three states.
“Over the last two years we've had more strokes than Barnes Jewish in St. Louis, and we've had more strokes than University of Iowa, which is the only comprehensive center in Iowa.”
Across the OSF system, Talkad said there were 2,000 strokes last year. Expanded use of the tech in rural hospitals and medium sized communities like Bloomington-Normal, essentially doubles the number of scans neurologists must review. Talkad said that may eventually force OSF to add specialists.
Limits of technology
AIs become dependable through large data learning models. The Food and Drug Administration approved Rapid AI in 2014. OSF St. Francis began using it in 2018. Rapid AI has 70% of the comprehensive stroke center market share in the U.S. reading the majority of an estimated 800,000 stroke brain scans per year in the U.S. said Talkad. That’s a lot of data to continue the AI learning curve. The technology is pretty reliable, he said.
“We're now up to greater than 95% when it comes to the accuracy of looking at whether it catches a hemorrhage or whether it catches early changes of a stroke. We're getting better almost than human eyes, and depending on which studies you look at, the rapid technology has been able to diagnose certain types of strokes, ischemic strokes and hemorrhages better than human eyes,” said Talkad.
The AI frontier
Talkad said there are developments yet to happen with AI and stroke treatment and in other areas of medicine. Accuracy will improve incrementally, though he’s not sure when that will plateau for stroke assessment. It will never be perfect. Some companies are expanding outside the brain.
“The brain was the easiest because you have one focused organ and stroke had no treatments. But blood flow goes to all these organs, right?” said Talkad.
Illinois State University is doing heart research on AI and the condition of atrial fibrillation.
He said Rapid AI is close to getting FDA approval for detecting pulmonary embolisms [clots in the lungs] and doing work on aortic aneurysms.
Nothing is perfect though, not humans, not AIs. Talkad said human and machine errors can be cumulative and offsetting or either, depending on the context. AI, he said, does not replace the clinical aspect of human physician interaction with the patient.
“But the human interaction can be enhanced because now we have technology that allows us to incorporate better what the human eyes and the human ears have seen…. When a neurologist looks at the patient and a neurologist looks at the CT scan or the perfusion, now we can put both halves of those together and say, okay, this is how we need to do it,” said Talkad.
