Hacker News Reader

Comment by Gatsky | Hacker News Reader

I don't know anywhere that has a surfeit of radiologists just sitting around chewing gum. The types of tools described in the paper will enter the market as radiologist assisting devices, if for no other reason that the regulatory burden for this is dramatically less than a total radiologist replacement. Their introduction will make radiologist's jobs easier, and improve scan throughput and reduce waiting times for results. This will free up radiologists to deal with more complex cases. But I doubt it will materially reduce the number of radiology jobs, at least in the next generation.

View on HN

> The types of tools described in the paper will enter the market as radiologist assisting devices, if for no other reason that the regulatory burden for this is dramatically less than a total radiologist replacement.

100% agreed.

A significant dimension of the regulatory dynamics is accountability: Who will sign off on and ultimately be responsible for findings from radiologic studies?

Yes, taking responsibility is primarily the function of a doctor. This is actually a very unique and rare thing. Everywhere else we see people minimising their responsibility to others, rather than actively taking it on board.

Ceding this responsibility to corporations is a terrible idea. After all, one of the things about a corporate entity is that there isn't really anyone responsible. The GFC and Boeing are recent perfect examples of this. Automating medicine will result in making healthcare more like trying to get tech support. Yes doctors are imperfect, they make mistakes, some definitely shouldn't be working, they are territorial and monopolistic etc etc, but when the system is working you walk into a room with another person who wants to listen to you and help you, and we shouldn't ever try to take that away.

If you could reduce the cost by increasing the productivity of radiologists, we'll be doing lots more and be getting better outcomes. This should be a win-win.

I think there's a huge potential for patient benefit if this is done correctly, though I worry that we'll just create mostly adequate classifiers based on existing data.

I also think we need to watch out for the human-attention issues illustrated in almost self driving cars. If a radiologist gets used to the computer being right 9/10 times, they could miss the 10th which would usually have been caught.

Overall we need more CV/ML/AI (choose your acronym) in this space, but it definitely requires some care.

There's a ton of deep learning startups working in this space already. Hardest part for most of them is getting access to training data. Because of that we'll probably see most of the commercial innovation coming from China and other countries that don't care about privacy.

Doctors make mistakes all the time and people in the medical profession work odd hours so there's already a ton of room for errors. Having a machine check their work or provide a second opinion will help them a lot.

Having a system that can do the primary screening and prioritize patients before a radiologist is available will save a ton of lives.

Part of the problem is that inadequate training data may not be a barrier to successfully marketing and implementing these systems - the tech is sexy and people are rightly excited.

> Doctors make mistakes all the time and people in the medical profession work odd hours so there's already a ton of room for errors.

That's my point - this is the training data. Unless we're careful, we're just going to approximate what we're already doing.

These systems can be trained on more data than any doctor will see in their lifetime. This means that it can pick up on discrepancies between doctors and has room to outperform most doctors.

I'm working on a similar system but for dietaty feedback based on images and it's amazing to see the model outperform all of the dietitians because it's able to see how all of the coaches respond to similar items.

Is this different than any other illness that became easily testable? Wouldn't this eventually move into something a technician would perform, like a blood test?

Technicans perform blood tests but doctors are the ones that interpret the results.

Right, but GPs can interpret the results and raise red flags as needed. My point is not every test needs a specialist.

The problem is that unlike a blood test, neural networks are not transparent at all. So when an imaging test comes back with "possible malignancy" with red circle around the highlighted area, a GP doesn't have the experience to agree or disagree with the result. So every time that happens the next step will be to send the image to a specialist.

And in the case where the machine learning algorithms don't find anything suspicious, the GP again won't have the training or experience to confirm those results. Now if the person was otherwise healthy and this was just a screening that might be enough, but if the GP was suspicious enough to order the test in the first place, it won't be.

What will probably happens is that this kind of technology increases productivity for radiologists, and maybe increases the number of screenings done on healthy people. But it's not going to reduce the demand for radiologists.

Basically the problem is that to be able to interpret the output of a neural network you need to be an expert. What we need is AI that can present a fully formed argument that is easy for a non expert to follow and validate, but we are nowhere near that in most cases.

Technicians don’t interpret blood tests, and radiologists don’t perform most imaging procedures, so I don’t understand the comparison.

Well my thinking is more that the results would be read by your GP and not a specialist.

Reply on news.ycombinator.com