AI in healthcare.

[kindai]

So this popped up in my news feed earlier.

NHS AI test spots tiny cancers missed by doctors

An AI tool called Mia found missed breast cancer symptoms in the scans of 11 women during an NHS test.

www.bbc.co.uk

I must say damn impressive, and the accuracy of it is astounding.

Looking at the picture at the top with the 2 circular marks however, can anyone see what its picked up on?! Because to my untrained eye I cant see anything that doesnt look like another part of the image in some way.

 

[413x]

So this popped up in my news feed earlier.

NHS AI test spots tiny cancers missed by doctors

An AI tool called Mia found missed breast cancer symptoms in the scans of 11 women during an NHS test.

www.bbc.co.uk

I must say damn impressive, and the accuracy of it is astounding.

Looking at the picture at the top with the 2 circular marks however, can anyone see what its picked up on?! Because to my untrained eye I cant see anything that doesnt look like another part of the image in some way.

This is where AI could be a game changer.
Medical diagnostics.
You could have an mri and the results instantly. No waiting for weeks.

It would Truely be a game changer. Only takes half an hour for most MRIs. You could have scans basically 24/7 and only needing staff at the point of the scan.

Faster it happens the better.


Can't believe (well I can) some people didn't want thier scans looked at by the AI.
Maybe after reading the results that it picked up cancers that radiologists missed they'll regret that decision.

 

[radderfire]

No I can't.

You can see though that AI would be well suited to this as essentially it's a pattern recognition task, and very tiring for a human to do.

 

[PlacidCasual]

This is a real opportunity for the UK and the NHS. With such a large patient record database training "AI" models for such tasks is something the NHS could make quite some income from. Just need to put the appropriate legislation and safeguards in place regarding records so the "AI" is being taught blind and the data never really leaves NHS control.

 

[jeffk]

Can't believe (well I can) some people didn't want thier scans looked at by the AI.
Maybe after reading the results that it picked up cancers that radiologists missed they'll regret that decision.

I’ve worked in this field and, as others have pointed out, AI is great for pattern-matching over large volumes of data.

However, it’s not perfect and never will be. I’d always want a second opinion from a human being.

 

[Freakbro]

However, it’s not perfect and never will be. I’d always want a second opinion from a human being.

Does it need to be? As neither are doctors, so I'd just prefer the least fallible

 

[413x]

I’ve worked in this field and, as others have pointed out, AI is great for pattern-matching over large volumes of data.

However, it’s not perfect and never will be. I’d always want a second opinion from a human being.

It looks like it's at the point of being better now if studies like that are coming out.
If it was me.. I'd rather have the AI results now and a doctors when available.

Especially with cancer when weeks can make a difference.

 

[jeffk]

Does it need to be? As neither are doctors, so I'd just prefer the least fallible

If you have two fallible systems, it’s best not to rely exclusively on either.

Being the victim of a false negative is life-threatening but being the victim of a false positive can also be mentally damaging.

 

[arknor]

This is where AI could be a game changer.

but image recognition models to train them don't they basically play spot the difference between sets of images.

you tell is what in the image to identify and feed in wrong images on purpose so it learns the difference, kinda like pattern recognition

so surely this cancer spotting AI can only spot known examples of cancers as would be detectable by humans in the first place? because a humans literally had to train it on what cancer looks like.

you could argue it spots cancers and flags possible cancer cells, that a human might not always 100% notice, like maybe 7 in 10 would spot them etc depending on how good they are at their job

hopefully they use AI to check xrays too, AI could do a humans 9-5 in a few seconds

 

[Steeps]

This shouldn't just be for diagnosis but across the entire NHS. The two most common excuses we get from medical professionals when considering treatments are "I wasn't made aware" or "I don't believe", where the professional either hasn't had time to read up on new studies and treatments, or in their opinion there's no benefit because the existing treatment system is all they want to follow. Then you have the committees who allow and remove treatments purely on a cost to purchase instead of the holistic approach of the benefit to the NHS and the benefit to the patient, if they factored in treatments which needed less NHS intervention and repeat visits but at a slightly higher cost and with a greater success rate, it would save more of the resources the NHS is struggling with, which is the time of a professional. More AI analysis is needed as the wealth of resources available is much greater, the risk is making sure it's secure and hard to influence.

 

[Mr Jack]

I wonder how certain we can be that these "tiny cancers" are really medically important? Cancer has a history of early diagnosis not achieving anywhere near as much as expected. Still, treated sensible computer-assisted diagnosis seems like a good thing but what's the false positive rate? We already carry out a large number of needless and harmful operations due to false positives in cancer screening.

 

[arknor]

This shouldn't just be for diagnosis but across the entire NHS. The two most common excuses we get from medical professionals when considering treatments are "I wasn't made aware" or "I don't believe", where the professional either hasn't had time to read up on new studies and treatments, or in their opinion there's no benefit because the existing treatment system is all they want to follow. Then you have the committees who allow and remove treatments purely on a cost to purchase instead of the holistic approach of the benefit to the NHS and the benefit to the patient, if they factored in treatments which needed less NHS intervention and repeat visits but at a slightly higher cost and with a greater success rate, it would save more of the resources the NHS is struggling with, which is the time of a professional. More AI analysis is needed as the wealth of resources available is much greater, the risk is making sure it's secure and hard to influence.

loads of people cry and kick up a fuss everytime Palantir gets an NHS contract, and they are a company that can massively improve the efficiency op the NHS

have you seen this clown website

No Palantir in our NHS

Add your name now to call for US data company Palantir to be dropped from the NHS – and for the future of health data to keep faith with patient trust.

nopalantir.org.uk

Palantir is a US tech and security corporation with a terrible track record. They help governments, intelligence agencies, and border forces to spy on innocent citizens and target minorities and the poor.

We don’t trust them with our health data, and we don’t trust them to respect the values of our NHS.

erm.... NO they don't.......

Terrible track record? whos funding this website China or Russia?


the "border forces to spy on innocent citizens" I don't think they are citizens if they are crossing a border like that.
AFAIK no government is using them for border security because no country is securing their border... so why would anyone pay Palantir for intel they wouldn't action anyway....


wow the about Palantir page on that site is a full on KAREN, mentioning trump and stuff...

is the real problem with Palantir that is pro west, names countries like China etc as enemies of the west so refuses to do any business with those countries or its companies.

and Palantir is outspokenly pro israel to the point the CEO has said they have lost employees over it and expect to lose more.

god forbid a company draws a line in the sand and refuses to help our enemies and openly calls them that...

 

[kindai]

If it was me.. I'd rather have the AI results now and a doctors when available.

This, it would be good if it could be blindly split and recorded some way to ensure the doctors dont fall back on complacency that the AI will do the work for them.

So you go for a scan.

AI checks it.

If it finds something, immediately sent to the front of the queue for doctor A review without information on what it has found, if the doctor signs it off as clean, it then sends it to doctor B to look at the bit the AI picked up.

If nothing is found, then the scans get placed into the queue for manual review regardless.

 

[MatsyLR]

I'd say it would need to be verified by a human.
Scan -> AI checks -> goes to multiple doctors if needed to check the AI is correct.

 

[413x]

I'd say it would need to be verified by a human.
Scan -> AI checks -> goes to multiple doctors if needed to check the AI is correct.

Negate a lot of the benefits here.
There's a backlog that needs to be reduced.

 

[arknor]

Negate a lot of the benefits here.
There's a backlog that needs to be reduced.

Wouldn't the AI be ruling out all the negatives and leaving just the positive and possible cancer results for actual doctors to check.


probably eliminates like 90% of the workload?

Don't see the point of verifying by human for negatives, their would be enough testing done to see the accuracy of the model anyway.

and with patterns/images these models are insanely accurate, yet somehow can't win versus a captcha test :O

 

[413x]

Wouldn't the AI be ruling out all the negatives and leaving just the positive and possible cancer results for actual doctors to check.


probably eliminates like 90% of the workload?

Probably. After enough trials I, myself, would be happy with that to get quicker gp to scan to results.

 

[kindai]

Heh!

So I have meant to get a mole checked on my leg recently as its started itching. Funny enough ive been referred to a clinic where it will be examined by AI under heavy magnification.

Will be interesting.

 

[adam cool dude]

This is a real opportunity for the UK and the NHS. With such a large patient record database training "AI" models for such tasks is something the NHS could make quite some income from. Just need to put the appropriate legislation and safeguards in place regarding records so the "AI" is being taught blind and the data never really leaves NHS control.

The government will just privatise after brown envelopes are given to them by some big cheese living in the Bahama's. They will reap the financial rewards charging £££'s a scan whilst the NHS continues to crumble. No doubt the tax payer will fund all the infrastructure too.

 

[aVdub]

Low subscription vs teams that miss things....

I'm kind of guessing the NHS is throwing bank notes at AI and then after a few years will not carry on using it.