DeepMind's New AI Diagnoses Eye Conditions

The paper "Clinically applicable deep learning for
diagnosis and referral in retinal disease" is available here:

We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
313V, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Brian Gilman, Christian Ahlin, Christoph Jadanowski, Dennis Abts, Emmanuel, Eric Haddad, Eric Martel, Esa Turkulainen, Evan Breznyik, Geronimo Moralez, John De Witt, Kjartan Olason, Lorin Atzberger, Marten Rauschenberg, Michael Albrecht, Michael Jensen, Milan Lajtoš, Morten Punnerud Engelstad, Nader Shakerin, Owen Skarpness, Raul Araújo da Silva, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Thomas Krcmar, Torsten Reil, Zach Boldyga.

Two Minute Papers Merch:

Károly Zsolnai-Fehér's links:

[MobyMotion]

Something I really like about this paper is the intermediate, segmented, interpretation. As you say, it allows you to classify different modalities than OCT, but it also helps with interpretability. This is something that is particularly sensitive in healthcare data, where patients want to know why decisions about their care are being made.

[mysticmarble]

Day gets better by at least a factor of 10 every time I recieve a new Two Minute Papers notification 😊

[thisisnonpractice]

I like the way you say, "What a time to be alive!" :)

[TheCobyRandal]

This channel is one of the best, if not THE best I've ever subscribed to. What an amazing thing: we're generating more and more data which is hard for humans but it is getting better and better for training machine learning. Wow. And this medical application will bring health care into a preventative consumer level that is actually self-responsible "health" care instead of sick care, using AI tools to identify the state of the body, make precise early detection of issues, and suggest lifestyle changes for optimal health before any serious conditions arise. Doctors will only get better and better as the general public uses tools like this to prevent disease. Amazing.

[PandoraMakesGames]

I love to see AI applied in a medical domain. Incredibly useful!

[rabbitpiet]

“If I Didn’t see these reports with my own eyes”

[kayrosis]

a two minute paper that doesnt cause existential dread! yay!

[bipulkalita]

This time too, I nearly fell off my chair.

[ymi_yugy]

One risk we should be aware of from the beginning is that some doctors will start to blindly trust the result due to high stress and little time.
This is especially worrisome as neural networks often make stupid mistakes that are obvious to humans even if their overall level is above that of a human.

[Armuotas]

At my clinic there's a new program to train young MRI students by sitting them next to the usual technicians so they could learn. So far so good, but, I've been told that they will be slowly replacing the staff (because it's cheaper). Mind you that there's a difference between MRI technician and MRI diagnostician. Later can "read" the images, the former only operate the device. And the salaries are quite different. So I think the medical imaging community bosses have a good sense of what's comming in the next several years.

[scichopath]

I'd love to see AI being used in public decision.

[Qman]

We're moving closer to a time where going to the doctor is just a notification you get from an AI enhanced suite of sensors that become increasingly better at catching the early warning signs of any chronic medical condition.

[atroxit]

Optometrist 4 was terrible, 24.1% error rate.

[scruff]

Amazing from such a relatively small set of training data. Like to know how this would do with a much larger data set.

[not_a_human_being]

such an amazing content, please do not stop!!! Diversify if you can into other stuff - I mean images are great, but there are a lot of advances in reinforcement learning too for instance. I feel you're coming strongly from video side, but with your talent of reading/interpreting papers you surely can do other applications of NNs!

[paulbloemen]

Interesting! I saw this video because of my interest in AlphaGo Zero and AlphaZero, thus, my interest in go and chess. My question is: has DeepMind improved the Artificial Intelligence algorithm where both go and chess engines can profit from it? Like, a higher maximum level, a quicker learning process, less resources needed to proceed?

[vidusha]

May I ask How do you create these videos and animations?

[artman]

Within a year or so, the thing is probably even more accurate.

[tostupidforname]

Does this mean we can use AI to detect cancer early?

[Intrevel]

hopefully some AI sorcery in the future can fix strabismus/astigmatism