back to article Ayyy-EYE! Google code 'predicts heart disease' by eyeballing retinas

AI researchers at Google have developed algorithms that can assess the risk of heart attacks by analyzing retinal scans. By looking for common patterns in images of retinal scans and matching them up with the data in the patients’ medical records, one algorithm could determine if someone was a smoker or non-smoker to an …

Anonymous Coward

'The level of accuracy is similar to drawing blood to measure cholesterol levels.'

To be clear, there isn't universal agreement in the medical community about Cholesterol levels. Some think monitoring LDL/HDL is unreliable. Others think HDL doesn't help offset LDL... Or the negative effects of LDL are actually overstated. So good luck putting this all together Google. But I'm sure you'll spin it every way you can / get every inch of AI-dominance PR!

5
0
Silver badge

Re: 'The level of accuracy is similar to drawing blood to measure cholesterol levels.'

I read that as "Eat all the bacon you like."

9
0
Silver badge

Re: 'The level of accuracy is similar to drawing blood to measure cholesterol levels.'

"To be clear, there isn't universal agreement in the medical community about Cholesterol levels."

So what? No-one here is claiming anything about cholesterol levels. It's a new tool that has similar accuracy to existing tools, but in non-invasive way and much faster. You're whining that Google are bad because the existing tools and medical understanding aren't great to start with, but that has absolutely nothing to do with anything Google have done here.

1
1
Silver badge
Thumb Up

Re: 'The level of accuracy is similar to drawing blood to measure cholesterol levels.'

I read that as "Eat all the bacon you like."

I love that the person who posted this chose the name "Crisp"!

2
0

Hmmm, perhaps econo-political diatribes could be avoided...

... for awhile in this case.

This appears to be decent research, and inasmuch as cardiovascular disease is a killer particularly in Western societies, I would be inclined to give the researchers some credit for possibly valuable findings. It's not like they're claiming a Holy Grail, just positing a new diagnostic tool.

"Lily Peng, a doctor and lead researcher on the project said that it was early and they were working with small data sets. In future large data sets could provide deeper insight. One of the problems with this study was that it could look at eye images at 45 degree views and this could miss out vital zones in the retina. Researchers are trying to correct this problem with new versions. Although more research is necessary, the team still calls this a major step towards 'non-invasive' diagnosis and predictor of cardiovascular health." Linky

I would prefer, for once, not to connect every damned thing with the sins of Google, or of Microsoft, or of Apple, or of Canonical, or whatever. Just let the research stand (or fall) on its merits. But that's just me.

6
0
Silver badge
Big Brother

The web will be like TV on a weekday afternoon

I suspect that Google will soon deprecate fingerprint scanners in favor of retina scanners. There's big money in pharmaceuticals.

1
0
Silver badge

I wonder (maybe just being paranoid) what's in this research for Google? They seem to be funding a nice chunk of research into retina work.

2
0
Anonymous Coward

Finding the perfect place to implant ad's directly into the eyes?

2
0
Silver badge

It can predict age or gender

Wow! That's so useful. I can't tell you the number of times I've been to the optician and they'd didn't know whether to address me as sir or madam until they'd scanned my retina. Oh, wait, I can tell you how many times: zero.

The number of times you're going to need a retinal scan from somebody whose sex is otherwise indeterminate (but not because that person has suffered so many injuries as to be incapable of surviving) is close to zero. So it's not a useful feature of the technology.

And if it's not a useful feature, why mention it when it is only around 75% accurate? That merely highlights the inaccuracy, which throws doubt on other predictions it makes. Hint: I can do better than 75% at guessing the sex of people without submitting them to a retinal scan.

3
0
Silver badge

The system compared retina images with medical records.

Which makes me think the condition was already noted in the patient before the retina image was taken?

The trick is to identify the condition from the retina before the doctor notices it, not after. Which would require photographing a large number of volunteers, seeing which ones the system assessed as possibly having CV issues and then checking their BP etc and comparing.

Having looked at the paper, I see that predicted vs actual mmHg looks like quite a bad correlation (ie a fuzzy cloud not on the slope) That major adverse cardiac events is AUC of 0.70 which is "fair" not "good" match

3
0
Bronze badge

Exactly, this is the ML meaning of "predicted" which means "we could guess one sample attribute based on the others", rather than "we can predict what happens in five years time" which requires follow-up. If they really want to predict something from fundus images they'd be going after risk of macular degeneration or (my request) glaucoma.

1
0
Anonymous Coward

Hell no

This is NOT Medical Science it's Probability using Normed Stats and utterly useless.

If you rely on this you will have a bad time with so many false positives it will bring the whole circus down.

Have fun.

0
0

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Forums

Biting the hand that feeds IT © 1998–2018