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?
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.
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.
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.
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.
> 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.
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.
This stuff will be used to do initial screening to prioritize cases and provide an initial analysis for the radiologist to confirm. Once they're deployed it won't be too long before they're as good as the top practitioners.
It will be a long time before they completely replace radiologists in America but we'll probably see them on autopilot in third world countries where there's a shortage of doctors and data privacy laws are not as stringent. I've met a Chinese guy doing medicine in the states who claimed to have access to all medical data for a bunch of hospitals back in China.
Radiologists are likely to see a lower demand as a result of these technologies and will either A) Spend more time on complicated cases or B) Be let go. Nobody is saying ALL radiologists are going to be out a job. Look at dosimetry as a recent example of how software improved, and the time to contour per patient decreased, causing many health systems to shrink their dosimetrist staff or offload the responsibilities to the physician office.
This isn't the first time technology has been applied to healthcare, change will come slowly and eventually people will have to find new fields to work within healthcare.
They've been successfully pushing back against scope of practice changes for RAs, PAs, ANPs, PTs and other midlevels, let alone allowing foreign doctors in without an expensive residency medallion. And on the data side, hospitals are doing all they can to make sure that is as locked down in their silos as possible, because all these AI papers have been posing a fundamental threat to their bottom line (Medicare and employer reimbursement for physician and related services).
Basically every single decision-making regulatory body on state and Federal levels is full of MDs, with inherent conflict of interest.
Now new FDA chief is an old-school MD again, and his first order of business was to call a conference studying the potential dangers AI can pose to patient safety.
As such changes continue to be signed into law, I assume you mean, overall, unsuccessfully doing so.
I had some coworkers who went to RSNA this year. The AI companies are still desperate for data. There was direct discussion of the disappointment in AI in more than one talk.
It'll happen, but like anything else it'll take a lot longer than people were predicting.
I don't think taking the human all the way out of the loop is a great idea given the state of models we've seen to date (in medicine, anyway).
I think a better direction would be looking at how to make these systems more complementary with human operators, surfacing interesting features or distant connections humans tend to miss and guarding against the big errors anyone might make staring at grayscale images on the night shift.
Yes, I agree it should be another, yet increasingly important tool for a while. Perhaps not until a near-AGI is invented that machines can completely automate away doctor jobs.
In some cases they also spend a lot of time annotating and measuring small areas of the images and having a model generate suggestions for them would save a ton of time.
If the NNS for a screening test is 5,000, those who advocate screening must make the ethical argument that the large benefits to 1 individual justify the sum of the harms to which 4,999 people are exposed. Whether this holds up to moral scrutiny depends on the nature of the harms.
The problem that I think the medical community misses is that the unwashed masses are generally completely unable to access procedures that aren't recommended by a doctor. So in the above assertion, that someone 'advocating for screening' must take on the ethical burden of harms coming from that screening, in my estimation, is complete and utter bullshit. What I would prefer to see is that screening comes with patient education so they understand the potential risks and inaccuracies and then let 'er rip. If something goes south, it's on them.
Gerd Gigerenzer has done a lot of work on this, so his books are useful. Reckoning With Risk or Risk Savvy are good.
> What I would prefer to see is that screening comes with patient education so they understand the potential risks and inaccuracies and then let 'er rip. If something goes south, it's on them.
Informed choice should already be built into all healthcare systems because it's a feature of international human rights laws and patients are usually allowed to decline to have a test done. And there's usually a doctor somewhere who'll perform an unnecessary test if you're prepared to pay.
Communicating risk is difficult.
We know that merely giving people information and letting them take full responsibility won't work. We know it won't work because we already study whether people understand the risks of testing and treatment, and we find a disturbingly large number of people don't, and that includes the HCPs recommending the testing and treatments.
Most people struggle with this question: "A machine has been invented to scan a population for a disease. The machine is good but not perfect. If you have the disease there is a 90% chance it will return positive. If you do not have the disease there is a 1% chance it will return positive. About 1% of the population have the disease. Mr Smith is tested, and the test comes back positive. What's the chance Mr Smith actually has the disease?"
But the problem of lack of numeracy is more severe: only 20% - 25% of people understand that 0.1% is 1 in 1,000. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3310025/
People do not understand the difference between "absolute" vs "relative" risk increases, so we see newspapers reporting "100% increase in risk from eating X" when the numbers are an increase from 1:100,000 to 2:100,000 deaths.
https://bestpractice.bmj.com/info/toolkit/practise-ebm/under...
https://www.eufic.org/en/understanding-science/article/absol...