Previous: The Key to an Artificial Heart ... and Open-Heart Surgery
Next: Where Should I Put My Wi-Fi Router?



View count:124,293
Last sync:2019-12-04 07:50
This week the FDA approves the first ever blood test for diagnosing concussions, and a group of scientists develop a neural network that could save you a trip to the eye doctor.

Hosted by: Hank Green
Support SciShow by becoming a patron on Patreon:
Dooblydoo thanks go to the following Patreon supporters: Kelly Landrum Jones, Sam Lutfi, Kevin Knupp, Nicholas Smith, D.A. Noe, alexander wadsworth, سلطا الخليفي, Piya Shedden, KatieMarie Magnone, Scott Satovsky Jr, Charles Southerland, Bader AlGhamdi, James Harshaw, Patrick Merrithew, Patrick D. Ashmore, Candy, Tim Curwick, charles george, Saul, Mark Terrio-Cameron, Viraansh Bhanushali, Kevin Bealer, Philippe von Bergen, Chris Peters, Justin Lentz
Looking for SciShow elsewhere on the internet?

Concussions are just the worst. Among other things, they can cause massive headaches, memory loss, nausea, and make you lose consciousness.

Not to mention the potentially life-threatening complications, like bleeding in the brain. So when someone hits their head really hard, it’s important to know if they have a concussion. And now, the process of figuring that out will be a little easier.

Because last week, the U. S. Food and Drug Administration approved the first-ever blood test for diagnosing a concussion.

Concussions are notoriously hard to diagnose. Doctors mostly base it on symptoms, using a 15-point scale to check the severity based on things like whether you can open your eyes, whether you can speak coherently, and whether you can move. All important things.

Then, they follow-up with a computed tomography, or CT scan, which creates a 3D image using. X-rays and can show whether any part of your brain is bleeding, or if it’s swelling. The problem is, CT scans involve a lot of radiation exposure.

A head CT is like getting a hundred chest X-rays at once, and that can increase your chances of developing harmful DNA mutations that lead to cancer later in life. Not to mention, the scans only show damage in about 10% of concussion cases. So as if hitting your head wasn’t bad enough in the first place, you then have to expose yourself to a bunch of radiation that might not even tell you whether you have a concussion!

There’s a reason doctors do these CT scans, though. They may not be very good at diagnosing concussions in general, but they are super useful for detecting brain lesions — the damage that might mean there’s bleeding or swelling. And you definitely want to know if that’s happening, so surgeons can go in and try to fix the problem.

That’s where this new blood test can help, because it predicts the results of the CT scan. The test works by detecting two proteins that appear in the blood within an hour after a brain injury. One is an enzyme in neurons that breaks down proteins they don’t need anymore, called.

UCH-L1. The other, known as GFAP, is a structural protein in astrocytes — star-shaped brain cells that help support your neurons. Normally, if all is well, these proteins stay inside your brain cells where they belong.

But if you hit your head hard enough to break open some brain cells, the proteins will get into your bloodstream. That’s what this test looks for. Based on the blood levels of the two proteins, it can predict whether someone’s CT scan will come back positive or negative.

The FDA gave the blood test a green light because, in a clinical trial of around 2,000 people, it correctly predicted the brain lesions found by CT scans 97.5% of the time. And it was even better at predicting a negative result. So if you take this blood test and it’s negative, great.

I mean, you could still have a concussion, but at least you know your brain isn’t bleeding or anything. And you can skip the CT scan because it probably won’t give you new information. If it’s positive, on the other hand, doctors will know to do the scan to see what’s going on in your brain.

So the blood test won’t necessarily make diagnosing concussions more accurate. But at least it’ll make the process easier on patients. According to the FDA, the blood test should help doctors rule out CT scans in a third of people, if not more.

And that will add up, since about 2.8 million Americans head to the ER every year for concussions. So hurray for less radiation! —. If blood tests will soon be handy for concussions, how about something that sounds even more futuristic: using artificial intelligence as doctors!

This week, an international group of scientists unveiled their latest effort to use AI in medicine. Specifically, they programmed a system to diagnose eye diseases that can lead to blindness, like age-related macular degeneration, where the center of the retina starts to get damaged. The best way to diagnose these types of diseases is with an imaging technique called optical coherence tomography, or OCT, which uses light to create a 3D map of the retina.

If the scan shows deposits or fluid buildup, you can use that to diagnose the problem. In some cases, doctors can then start treatment right away. The researchers wanted to see if an AI could make the same calls.

So they programmed a type of artificial intelligence known as a neural network. A neural network is basically a bunch of interconnected processing units set up to analyze data in a way that lets the system learn for itself. To train it, you just give it a bunch of examples of whatever you want it to learn — in this case, 100,000 OCT scans.

The then AI comes up with its own rules for solving the problem — here, how to link what’s in the image to the correct diagnosis. Then, when you give it a new example, like an OCT scan it’s never seen before, it uses those rules to figure out the most likely answer. Neural networks are used in some of the most advanced AI in the world, and this isn’t the first time they’ve been trained to diagnose diseases.

What’s different about this system is that the researchers only needed 100,000 OCT scans to train it, whereas most neural networks would need millions of examples. They did that by using something called transfer learning to create a shortcut, basically borrowing some of the knowledge from another program and applying it to this new, more complicated problem — diagnosing diseases. In the end, this new system could diagnose the eye conditions in 30 seconds with 96.6% accuracy, on par with eye doctors.

The computer even outperformed 2 of the 6 experts in the study. While these researchers focused on eye diseases, they also trained the system to tell the difference between bacterial and viral pneumonia in chest X-rays. And the same basic approach could be used for pretty much any other image-based diagnosis.

It will probably take a while before AI like this becomes standard in medical treatment. But once it is, it’ll allow patients to be diagnosed more quickly and easily, especially when there’s a shortage of specialists. We have seen the future of medicine, and it is now.

Thanks for watching this episode of SciShow News, and a special thanks to President of. Space fSR Foxley! It's wonderful to have his support and the support of all of our patrons.

Thank you so much for helping people get this And for helping us make it [ OUTRO ].