Google's DeepMind AI To Use 1 Million NHS Eye Scans To Spot Diseases Earlier (arstechnica.com)

Google DeepMind has announced its second collaboration with the NHS, as part of which it will work with Moorfields Eye Hospital in east London to build a machine learning system which will eventually be able to recognise sight-threatening conditions from just a digital scan of the eye. The five-year research project will draw on one million anonymous eye scans which are held on Moorfields' patient database, reports Ars Technica, with the aim to speed up the complex and time-consuming process of analysing eye scans. From the report:The hope is that this will allow diagnoses of common causes of sight loss, like diabetic retinopathy and age-related macular degeneration, to be spotted more rapidly and hence be treated more effectively. For example, Google says that up to 98 percent of sight loss resulting from diabetes can be prevented by early detection and treatment. Two million people are already living with sight loss in the UK, of whom around 360,000 are registered as blind or partially-sighted. Google quotes estimates that the number of people suffering from sight loss in the UK will double by 2050. Improvements in detection and treatment would therefore have a major impact on the quality of life for large numbers of people in the UK and around the world.

Comments

"anonymous eye scans"

[Anonymous Coward]

I think enough evidence exists that anonymous medical records don't exist - they effectively form a pseudonymous fingerprint and can be correlated with other databases to associate with names.

It's be interesting to know how many of these people actively opted in with sufficiently informed consent to know that Google might be processing their data for profit.

Re: "anonymous eye scans"

[Anonymous Coward]

No one opted in. The NHS digitised their records and were supposed to contact everyone to offer an opt out. No-one I know was contacted, and the plan was officially abandoned with consent assumed despite there high number of people that complained.

Re: "anonymous eye scans"

[Richard_at_work]

Ask your GP Surgery to add the following codes to your records, they opt you out of data sharing:

XaZ89 - stops non-anonymous data sharing
XaaVL - stops anonymous data sharing

Re: "anonymous eye scans"

[nospam007]

"There's a huge NHS ophthalmology centre opening nearby. I would not use it if I ran the risk that Google would be able to access my data in any sense."

You should have begun that sentence with "I'd rather go blind ..."

And the only way to treat this is by

[melted]

And the only way to treat this is by injecting a $1K+ per shot medication into your eyeball once a month (Lucentis). And the treatment is not permanent: macular degeneration returns again and again.

Re:And the only way to treat this is by

[jeffb (2.718)]

Imagine that -- a brand-new and effective treatment for a formerly-intractable disease is kind of expensive, and doesn't offer a one-shot cure. The evil doctors who came up with this diabolical scheme should be stripped of their medical credentials, if not summarily executed.

If you'd rather just stare at your feet and hope really hard that nothing's wrong, that's still free, and nobody's stopping you.

Re: And the only way to treat this is by

[melted]

It's still intractable unless you're willing to spend $10k/yr per eye for the rest of your life. And even then it doesn't fix the symptoms completely, and merely slows the progression of the disease. That is, if you don't have a reaction to this drug, in which case you may lose an eye, or both.

Re:And the only way to treat this is by

[melted]

It's not a "hospital" price. It's just the price of the drug, worldwide. US hospital prices are higher than that. And I think Margaret Thatcher explained many years ago that there's no free lunch, and eventually _someone_ pays for Lucentis. And that someone is not the government.

Interesting to look for disease markers

[afeeney]

While iridology is bunk , it would be interesting to see what disease markers could be found with eye exams. We already know about a few. Ankylosing spondylitis is often associated with eye inflammation and abnormalities in the retina can be associated with diabetes, hypertension, cardiac disease, and stroke, as well as a lot of systemic diseases.

Eye exams are generally non-invasive and the scans could be set up almost anywhere.

Some diseases stand out more than others.

[jeffb (2.718)]

There was a bit of news coverage a while back about childhood eye cancers being diagnosed from snapshots taken with on-camera flash. I have no doubt that detailed scans, processed against a very large dataset, could reveal other diseases that doctors currently don't catch early.

I know, I know, OMG GOOGLE BIG BROTHER, but I'd rather save my privacy outrage for proposals that don't offer a chance to substantially reduce human suffering.

Re:other uses for the data

[bigpat]

Why have some special deal with google only ?

If it's truly not personally identifiable, why don't they just release the data for any academic researchers to use in their machine learning datasets?
Perhaps they can use it to figure out other diseases and other information that is interesting...

Although I don't read into this any nefarious intent....
I think some public openness in the training data set so that different teams of researchers can try out different machine learning algorithms is a very good idea. All you need to do is remove identity information and keep the diagnosis tags for the training.

Anonymous?

[Macdude]

Anonymous eye scans? Isn't that like saying anonymous fingerprints? Aren't eye scans one of the biometric methods you can use to identify people?

Was this done on an opt-in basis or did the NHS just decide to give everyone's private medical information to Google.

Re:Anonymous?

[dave420]

Yes, it's anonymous. Just like fingerprints can be. No-one can look at a fingerprint and know whose it is without also having an identified fingerprint. Get it?

When are they going to use it to read films?

[hawkfish]

Consider: There is already a huge corpus of tagged digital data for training, knowing the results can be time-critical, and the specialists cost money. This job is already being outsourced remotely; the next step is outsourcing to software.

DeepMind alternatives

[Actually, I do RTFA]

Does anyone know what happened to all the rumored open-source clones of DeepMind? Did any of them get off the ground?

Whether built off of TensorFlow or not.

To Spot Diseases Earlier

[frovingslosh]

Next step, checking fingerprints to spot diseases earlier.

Re:other uses for the data

[ShanghaiBill]

I think some public openness in the training data set so that different teams of researchers can try out different machine learning algorithms is a very good idea. All you need to do is remove identity information and keep the diagnosis tags for the training.

It is not that simple. Even if identifying information is stripped out, it can be reconstructed by correlating with other databases. Retina scans can be used to diagnose some very sensitive conditions, including sexually transmitted disease. Specific researchers should be able to access the data, with appropriate safeguards and ethical oversight, but there is no way that this data should be public.

Doctors are going obsolete

[fph il quozientatore]

MDs are afraid of becoming obsolete, like horse farriers, photo and film shops, and soon also taxi drivers. Google's AI is probably already better than your non-specialist doctor at diagnosing diseases. So I expect they will be actively opposed to any attempt to step up machine learning for the health sciences.