[roentgen]

I'm not sure if you've ever seen a CT stroke study, which typically includes 4,000 images. If you think a computer can accurately interpret one of these any time soon, I would say you are poorly informed.

See my response below, and look into "CAD" and mammography.

[Locke1689]

Hmm, how exactly does the number of images affect predictive algorithms? If anything, more data is better. No offense, but as a radiologist you're not exactly familiar with the state of the art in A.I. image processing. I'd say that both of us are unqualified to state the degree of effectiveness of computational geometry and computer vision in radiology.

[adammichaelc]

You might not see it because it's such an inconvenient thing to see; afterall your income is tied to it. The main factor keeping your income so high is the AMA and its regulatory allies.

With rising costs and increasing pressure to save money in healthcare, you better believe that one day a computer will do your job.

PS You might want to re-read The Innovator's Dilemma and The Innovator's Solution.

[arn]

My income's not tied to it, so I'll say it.

Computers reliably interpreting films as the final word is not going to happen in the near future (within this guy's career). Reading films is as much an art as it is a science. I'm sure a lot of advances can be made on it, and we might see nice proofs of concept. But to the point that it actually replaces radiologists?

Even if you assume the technology can be perfected, there are too many non-technological hurdles for that to happen. Liability, trust, etc...

Machine EKG interpretation has been around for a while, but it's not even close to perfect and no one relies on it, and it's a much much much simpler problem.

[roentgen]

This guy gets it. Many replies in this thread are asking about computers interpreting scans, assuming I know nothing about the underlying technology, or am blinded by some form of bias.

I have been programming computers since I was 5 years old. I have a MS in neuroscience, and I am a board certified radiologist, so I think I'm qualified to understand the problem.

Believe it or not, nothing would make me happier than a magical black box that could spit out accurate radiology reports. Someday I'm going to get sick, and I would benefit from the technology.

If my job was replaced tomorrow I would be OK. I'm smart and hard working, and I'm good at almost everything I try, eventually. Also, I'm saving every last penny I earn, so I can keep things up for a few more years I should be financially secure.

Having said all that, I still think the problem is not solvable. On any given day I read xrays, CT scans, MRIs, ultrasounds, PET scans, mammograms, nuclear medicine studies, or live flouroscopic studies, and using CT or ultrasound guidance I can get a needle into just about any part of your body to take a biopsy. Doctors talk to me and our discussion influences the differential diagnosis, and the interventions planned. I am not just matching patterns, I am thinking and using my hard worn judgement.

Wishful thinking aside, computers cannot do this now, if ever. And if / when we reach the point that computers can do this, my guess is every other job will have fallen, with the exception of plumbing.

[krschultz]

The day computers can do your job, they can do every job.

But never underestimate the ability of programmers to oversimplify every other job while proclaiming a computer can do it.

Computers are tools to aid doctors, they are not doctors. It's like a blacksmiths claiming the horseshoes can get you somewhere without a horse. It's just not going to happen.

[jfoutz]

First, a radiologist made a judgement call and saved my mom's life. I've got a lot of respect for what you do.

Second, there must some set of scans that are easy. It's not hard to imagine a device that says, "yes" or "see a real radiologist". Perhaps later revisions can even say "No". an example might be a mammogram analyzer. I think even 10% getting an immediate answer would save a lot of money.

I think it's the normal progression of technology. Generally, you don't need a Phd in math or physics to program computers like was required in the 60's. The net effect will be the average case you look at is much more challenging.

[roentgen]

> an example might be a mammogram analyzer.

These exist, google for "mammogram computer aided detection". While the data on their efficacy is equivocal at best, I well tell you that they are useless. I do get to bill more for reading a mammogram if I run it through a CAD machine, which my group owns, so of course I do it.

Strangely enough, patients are reassured when the learn that the computer didn't detect any problems. And, more importantly, ignorant juries can be swayed by this piece of information. "The computer didn't detect anything? Then there is no way the radiologist should be held liable for missing that little tumor!" I'm not joking.

Little do they know.

[adrianwaj]

Do you use Thermography, I've read it to be safer?

[Robin_Message]

It's naturally for humans, especially ones who've invested years of effort into something and get paid a lot for it to continue, to claim that something is "more of an art than a science." They are wrong. Do they collect data on how they make decisions and what the outcomes are, and then check that they are improving?

For contrast, look at http://www.lifeclinic.com/fullpage.aspx?prid=508121&type... There they decided to stop gathering lots of information and making a judgement and instead use a few simple rules to make a decision on whether someone was having a heart attack and how bad. It was more accurate.

[thedufer]

I don't know much about reading films, but I do know something about AI, and this kind of direct data->solution problem with an enormous existing data set is just about as easy as AI problems get. I wouldn't be surprised at all if we are already at the point where its a trust/liability issue rather than a technological one.