Hospitals Deploy AI Tools To Detect COVID-19 on Chest Scans

Deep learning algorithms can diagnose, triage, and monitor coronavirus cases from lung images. Next, can they predict who will need a ventilator? From a report: AI-powered analysis of chest scans has the potential to alleviate the growing burden on radiologists, who must review and prioritize a rising number of patient chest scans each day, experts say. And in the future, the technology might help predict which patients are most likely to need a ventilator or medication, and which can be sent home. "That's the brass ring," says Matthew Lungren, a pediatric radiologist at Stanford University Medical Center and co-director of the Stanford Center for Artificial Intelligence in Medicine and Imaging. "That would be the killer app for this." Some companies are selling their tools, others have released free online versions, and various groups are organizing large crowdsourced repositories of medical images to generate new algorithms. "The system we designed can process huge amounts of CT scans per day," says Hayit Greenspan, a professor at Tel-Aviv University and chief scientist of RADLogics, a healthcare software company that recently announced one such AI-based system. "The capability for quickly covering a huge population is there."

Uh huh

[Viol8]

Another AI is the solution to this problem article. Call be skeptical but I suspect its quite obvious to doctors which patients will soon need a ventilator based on dropping O2 saturation levels in their blood and general condition. But no doubt a number of AI companies would be only to happy to cream off some of the money heading the way of health organisations around the world with their snake oil, I mean stellar medical IT solutions.

Re:

[kot-begemot-uk]

Actually all that there is here is image recognition. That is something AI nowdays does quite well. It is the "idiot savant" to look at photos. Some of the tasks in CCTV analysis or image processing used by criminal police for crime scene analysis are way more difficult.

So, yes, for once, this may actually be useful.

In fact extremely useful because at the stage when there is lung damage a lot of patients now show ZERO virus on PCR. The immune system has killed it and is now eating itself. While you can

This is how the robot apocalypse will go:

[Rick Schumann]

(Patient has COVID-19): AI says "Perfectly okay! It's just the flu. Tell them to hit the bars and shake lots of hands, they'll be fine!
(Patient has COVID-19): AI says "This one has found Jesus, they can't get infected, they're just hot with the Holy Spirit! Send them to church, all is well."
(Patient is completely clear of any infections): AI says "Warning! Danger! COVID-19 carrier! They'll destroy you all! KILL THEM, KILL THEM WITH FIRE, KILL THEM NOW!!!11!!"

"That would be the killer app for this."

[TD29]

"That would be the killer app for this." is not an appropriate phrase - unless you do not get the ventilator

How is this even possible?

[Astramensis]

How is this even possible considering that Covid-19 has no gold standard (positive ID)?

Re:

[Randj Herdle]

Trial by jury is not a gold standard for diagnosing murderers. But it's used to medicate society via the death penalty.

Solutions looking for problems

[ixneme]

This "development" is irrelevant in a number of ways:
- The imaging findings of COVID (like most lung diseases) lag behind the clinical findings, not the other way around. If you're getting a chest radiograph it's because someone thinks you might be in trouble already (or possibly just placating the occasional anxious and/or demanding patient).
- Imaging is not terribly sensitive to COVID. Up to half of chest CTs are normal in early COVID positive patients who are symptomatic. They are not terribly speci

Re:

[gweihir]

Actually, this may be somewhat useful (or not) in the cleanup when epidemiologists reconstruct what actually happens in the next few years. For clinical use, this is pretty meaningless for the current crisis. They do not even have a good ground truth training set at this time and will not get one for the next few months at least.

Re:

[sixoh1]

Unless HIPA is waived in the US for research, it's highly unlikely that anyone will have access to enough chest x-rays across a spectrum of folks to do anything like this. I'm not sure I want to enter the Brave New World (tm) where all my medical data is given to any random researcher who says "trust me I'm just using this to look for stealth Covid infections"...

Re:

[gweihir]

I was more thinking of statistics, where you basically have a rough feature distribution for each time interval and geographic location and only work on that. The actual x-ray would there go through a classifier and the results for each large enough set of x-rays would be aggregated (small sets would be dropped or aggregated with other small sets to ensure patients cannot be identified). If that is cleanly isolated and the aggregation results in good anonymization of the data, this should be possible to do

Nice research idea

[gweihir]

May even have useful results in a year or two. Or more likely will find yet another thing "AI" is not really god at, which, to be fair, is also a useful result.

Next, can they predict who will need a ventilator?

[AbRASiON]

"Next, can they predict who will need a ventilator? From a report:"

Having followed a few doctors (yes, real doctors, I'm not being trolled by someone) on twitter, I can tell you. That prediction should read more like.

"Has 80-90% chance of death"
Those are the figures, if you're unwell enough to need the ventilator and you survived you beat 1/5 to 1/10 odds.