Eye eye! DeepMind teams up with doctors to ogle eyeballs for illness

AI can help ophthalmologists diagnose more than 50 common eye diseases from retinal scans, according to a paper published in Nature Medicine on Monday. Researchers from DeepMind, University College London (UCL), and Moorfields Eye Hospital in London, have developed a system based on two neural networks that analyses 3D optical …

[Anonymous Coward]

It still needs someone to take the scan. Staff shortages at my local hospital have caused my OCT scan to be cancelled twice.

If they could automate that mechanical process as a sort of self-service kiosk then it would help. The opticians' shop doesn't use iris dilation drops for their retina pictures - so presumably those are not essential.

[charlieboywoof]

Maybe

But this will be done at the cash point in the future and notify you.

[ibmalone]

Not an ophthalmologist, but... high street opticians typically take a picture of your retina (called a fundus image), not an OCT, which can image the different layers, often taken as a cross-section, see https://www.opsweb.org/page/RetinalOCT

Unsure whether OCT requires iris dilation or not (have had it done a few times and you lose track of all the drops they're putting in).

[Fruit and Nutcase]

OCT

@AC & @ibmalone

Staff shortages at my local hospital have caused my OCT scan to be cancelled twice.

Check your local opticians/ask - some do have OCT machines. Mine offers OCT scan as part of their "enhanced" eye test - cost about £40, takes only a few minutes. No need for dilation.

[ibmalone]

Re: OCT

Yes, did a little digging afterwards and smallish studies on pupil dilation seem to suggest it doesn't have much impact on OCT quality, possibly unless you need to see the periphery of the retina.

Thanks, might look into that. Unfortunately if you're being monitored long-term you are relying on the care team to be able to look at the older images. I got surprise transferred to a different hospital between annual visits, as the one I had been going to shut down their eye department. (I was informed by a missed appointment letter for a fictitious appointment.) The new place was unable to access three years of retinal images. Almost certainly there were people with longer histories that were also lost in this transition.

[Anonymous Coward]

Oh Grandma What Big Eyes you have ...

... All the better to see you with my dear.

Jeremy Bentham must be spinning in his cabinet.

[GordyBUK]

So it's wrong 6% of the time

Whey are we expected to accept crap reliability rates when something is described as "AI"? If you car brakes failed 6% of the time when you use them people would be outraged. It would mean your brakes failing more than once in every 20 times you came to use them, which is probably at least once a day for people use their car on a daily basis.

Why is AI in medicine, in particular, not being held to the five nines standard expected of technology in critical systems?

[Anonymous Coward]

Re: So it's wrong 6% of the time ...

.. whereas us real-I humans have a record of being right that is /far/ superior... :-D

[Cuddles]

Re: So it's wrong 6% of the time

"Why is AI in medicine, in particular, not being held to the five nines standard expected of technology in critical systems?"

Because five nines has never been the standard expected for diagnosis of diseases. No human doctor has ever come close to that kind of accuracy, so it would be utterly insane to demand a computer must be that good before it can be used. If a machine learning system can assess scan as accurately as a human expert, what exactly is the problem? It frees up the humans to spend more time doing jobs the computers can't do, and brings essentially zero risk since no-one is going to start removing eyeballs based on a single scan without further tests regardless of who or what made the initial diagnosis.

[Anonymous Coward]

Re: So it's wrong 6% of the time

"Whey are we expected to accept crap reliability rates when something is described as "AI"?"

Because we accept that kind of rate when someone is described as a consultant ophthalmologist as well.

The overwhelming majority of the paper deals with the comparative performance of the model, which is markedly more complex than the 6% figure quoted in the headline. For example in urgent care decisions the rate falls to well below 1%, while when presented with scans from a machine the model had never seen before the performance initially collapsed.

The other aspect glossed over is that this isn't a binary output. The model produces a probability matrix over both its suspected diagnoses and recommended next action. Organisations will be able to set decision boundaries at the appropriate confidence levels, in line with their risk appetites, to account for the inaccuracies in the model.

There are millions of these scans performed worldwide every year, each one needing to be reviewed in detail by a skilled clinician. This is a technique that promises to perform just as well as that skilled clinician, allowing us to triage and prioritise our responses. The savings and the potential improvements in care, training and diagnosis are huge.

[Anonymous Coward]

Re: So it's wrong 6% of the time

"For example in urgent care decisions the rate falls to well below 1%,"

Or to translate, when it's obvious something is seriously wrong and needs urgent attention, the system still makes errors, would your opthalmologist make an error in such a case?

Does the NHS get to use this for free, being as "we" gave them the valuble data to train this. Why is it that one of the world's most profitable companies is given the highly valuble data for free**?

** Oh yes, Google, they expect all data the collect to be free. cnuts.

[Anonymous Coward]

Re: So it's wrong 6% of the time

"would your opthalmologist make an error in such a case?"

Yes. As I mentioned, the paper goes into a lot of detail on comparative performance:

"Performance of our framework matched our two best retina specialists and had a significantly higher performance than the other two retinal specialists and all four optometrists when they used only the OCT scans to make their referral suggestion (Fig. 3a, filled markers). When experts had access to the fundus image and patient summary notes to make their decision, their performance improved (Fig. 3a, empty markers) but our framework remained as good as the five best experts and continued to significantly outperform the other three "

I should have been clearer. The model is performing as well as the top performing specialists. Most of these millions upon millions of scans are not assessed by top performing specialists.

[The Nazz]

Love the diversity of the sample size.

If my arithmetic is correct, the sampling trainees included 359 less able people with only one eye (Ray Sawyer amongst them?).

Or 358 such people and one cyclops.

Biometrics : with the increasing use of biometric data, it's gonna be a bit of a devil's choice between keeping your identity intact or getting your diseased eye(s) fixed.

[Anonymous Coward]

DooM Awaits!

Of Course, like Watson, the machine intelligence will soon figure out ways to kill the patients.