computers have the potential to outperform physicians in every aspect- clinical - efficiency - surgery- , but similar to a giant mecha, they need a driver--- look at the davinci machine and robotic radiology. the era of man+machine is already here in all fields- medicine no different.

Vince: You want to let a robot operate on me?

Dr. Cameron: The technology is amazing. It magnifies everything ten times, it’s ten times the accuracy.

Patient: No way, I want a person!

Dr. Cameron: A person will be controlling the –

Dr. House: People suck. People have turned you from a guy with a swollen tongue into a guy with one eye, one ball, and a stapled-on face. If you want someone to hold you while you cry yourself to sleep at night, choose warm and soft. If you want someone to write you a poem, pick the sensitive loner. If all you care about is that something’s done right, pick the guy with the metal head.

As I mentioned in the other thread, as a future radiologist, I'd love to own a Watson and just "check" its findings. The technical fee that would be associated with it, and the volume of professional fees at the current levels would make me more money.

And they do have current image recognition systems, but they aren't clinically good enough at this point.

PS. vhawk, swing by The SDN Lounge for some WW.

I guess you havent worked with R2CAD for mammo

Any published papers on accuracy that you can link to readily?

I thought radiologists always complain about that program giving so many false positives that its not helpful, or have things gotten better?

you are a radiology resident. look it up yourself or ask a senior resident.

After following the other thread, and then seeing this one, i really have come to the conclusion that alot of the posters have very little idea of how doctors think or what they do.
Patients are generally quite poor at describing their symptoms without multiple follow up questions or prompting. They are also unable to weight these symptoms as far as their importance in figuring out what is wrong with them. I had a guy a few hours ago whose main complaint was ear pain and he wound up having bells palsy. No computer would have figured this out.

In the er we often have to make decisions and diagnoses with very little information and varying degrees of cooperation from the patients. Most of the people posting in these threads seem to be coming from the perspective of someone with a sprained ankle or seasonal allergies. Stuff that requires about 1% of our diagnostic ability. Of course this stuff is amenable to computer diagnosis. Try having someone come in with altered mental status, or shock, or a cardiac rhythm problem or whatever and see how a computer does. Not to mention the computer can't intubate the three people I did tonight or do all the procedures.

I find this stuff interesting from a theoretical perspective but I find a huge disconnect here between what people are posting here and the reality of practicing medicine.

As time moves forward the role of the human will be diminished with regards to diagnosis. As we further our understanding of the human body, less "intuition" will be required. It's actually quite clear from my point of view because I don't lose anything if computers replace humans. What the heck is the worth in determining limits of current technology when we are really talking about what the future holds?

Also a quick googling tells me Bells Palsy is a diagnosis of exclusion.. what leads you to believe that a computer would be unable to deal with this situation? Ten years ago did you believe an iPhone would exist in ten years? Did you think that the Watson AI would do so well on Jeopardy in year 2011?

Make claim. Asked to see evidence of claim. Refuse to provide evidence. Profit????

I would take a doctor over a computer. Doctors can register cues and fill in gaps of language and supply interpretations that computers can't. Just having one of those voice recognition systems that misinterpret statements when people mumble or lisp or mispronouce when they call into a business for customer service is enough to suggest to me there are things computers aren't proficient at doing.

Besides there is no trust/human factor. Medicine is highly concerned with compassion and a computer can't convey that with near the success rate of a human being.

http://www.auntminnie.com/index.asp?...s&ItemID=92216

Its not a "claim" anymore than claiming the sky is blue-

CAD already used by virtually all mammographers.

look up efficacy studies on google if you have the time - i dont - 2008 british study showed it is as good as a double read.

Thanks, that was all I was asking for. I'll look up the study.

i feel like a jerk -
here

Cadet II trial
http://www.asco.org/ASCOv2/Meetings/...stractID=33114

cost effectiveness still up in the air
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3032650/

thanks for the good educational excercise-
fyi - being used more and more for catching small lung nodules you can miss on Chest CT

ah internet guilt

Often you can't expect a computer to think about a problem the same way that a human would. That doesn't necessarily mean that it can't do a better job. An example would be a program that plays chess. Computers can do that arguably better than any human, certainly better than all but a small number of humans. However, computers don't think in terms of strategy the way that humans do. They operate largely on brute force, but in this case brute force wins. AI is replete with other examples where this approach fails, and what humans do cannot be well duplicated at all. The question is whether diagnostics is more like chess or more like these other examples.

diagnostics is a lot less like chess- its simpler, particularly reading x-rays, CTs, EKGs, pattern recog, med errors, and lab evaluation.

Yes a computer can outperform us on all this stuff, and we should take advantage of this.

Pattern recognition is an area that I know something about. What's easy for you can be extremely difficult for a computer. It all depends on the problem. There are some very successful programs in some application areas, and there are other areas that are the subject of intense research.

Looking trough my AI textbooks for anything on medical diagnosis, I find that it is based on abductive inference. Abduction is not a valid technique of logical inference, so it must be used with great care, but expert systems do it all the time. A stupid example would be "John feels bad, people with cancer feel bad, therefore John has cancer". Actual diagnostic abduction is fundamentally like this, just not as stupid (hopefully).

Tonight I called the cath lab team and a cardiologist in emergently within five minutes of a guy arriving by ambulace with chest pain for what the computer read as a stone cold normal EKG. The EKG actually had some diffuse concave st elevation in all the anterior leads which could have been a normal variant called early repolorization but based purely on gestalt of looking at the patient I thought it was an acute myocardial infarction.
Guy had a 100% LAD lesion successfully stented by the cardiologist. His initial troponin was normal since he arrived twenty minutes after the pain started.
So we got a "normal" EKG and normal labs.
Computer would have killed him or at least left him a cardiac cripple when his second set of labs it ordered four hours later came back with a troponin of 200 and half his heart is dead.

I don't doubt all of you pointing out how sophisticated AI, Watson, etc is at this point.
My only contribution here is trying to point out that there seems to be quite a
bit of underestimation if how monumental the task is because most of the posters only experience in medicine seems to be the minor straightforward problems that the relatively young and healthy population here have.

Nice catch on that MI.
Similarly, we've all seen stone cold negative EKGs, but the guy "looks" like an MI and is saying the right things - straight to cath is always great in that situation..Most places, that pt of yours would have been possibly stress tested..
Yes, Ive seen this exact situation several times myself.
Yes CP is vague and tricky.

Im obv not too worried to lose my job anytime soon..like I said, the robot will always needs a driver. Clinical Desicion Support , and other "robotic tools" will come to hospitals and EDs soon - (hospitals generally lag 3-4 years in IT infrastructure).

The trick is not to become lazy from it (ordering scans on everyone, and going only by the computer reading)

I built my career on strong ED communication - cheers to one of the real docs ITT.

Yeah, congrats on saving a life. What did I do today, watch tv, post some nonsense...is this something that many doctors would not have caught?

Are you sure the EKG in the ambulance was properly calibrated? I've heard that is a common problem. Ok, I heard it on CSI, but I heard it.

Guaranteed part of the ACS/MI protocol is a 12-lead EKG, regardless of what was done in the field

12 Lead will be negative in an acute coronary syndrome patient (99% blockage) who got nitro already, and is Chestpain free at the moment.. unfortunately the EKG, and the pt is going to get ugly really quick.

The problem is that it is tough to spot this guy just going on labs/ekg- he needs heavy blood thinners and a cardiologist...

unfortunately, many times this man will just say "im nauseated", my shoulder hurts, or even worse "chest-painless" ACS.

Thats where history taking (turbo style) and general feel come into play.

I think it may be safe to say, based on the volumes (#2 diagnosis for us), and variablibility, Chest Pain is the trickest of all. Usually straightforward, and occasionally a doc (and patient's) worst nightmare..

I mean the fact that his chest pain went away with nitro should tell you something imo - its probably not esophageal spasm when he's all sweaty and looks scared to death.