Comment by nkassis | Hacker News Reader

Comment by nkassis | Hacker News Reader

This article was extremely interesting. I remeber wired reporting on a system which would allow doctors to give a list of symptoms and have the most likely diagnostics listed by probability. Not sure of the name but technology like that has incredible potential. I also remember the article mentioning the reluctance of doctors to use it.

This kind of problem seems more widespread than only in the medical field. This stuff just reminds me of what Doug Englebart was trying to do with his NLS system. Augmenting Human intelligence. Were reaching the limits of what single humans are cable of retaining and processing in their daily jobs. I think that it's time for stuff Englebart did in the 60s to be rediscovered. While the internet is great, it's potential hasn't been maximised in area like this. Englebart's vision went much further than just google. Wolfram alpha is getting close to this in a way but still not far enough. I'm sure someone is already starting to develop this stuff (post some links if you have any).

I can just imagine an iPad application (preferably android table app ;p) for doctors which implements the system I talked about at the beginning of this post. Giving doctors on the go access to a database of diseases and medical records of the patients and a way for the computer to anylisze the data and return the info right there (powered by a huge hadoop cluster). Typing a few symptoms, getting the list of possible diagnostics and viewing some checklist of things to do. This would in no way replace doctors, I don't believe that we have reached a point where you don't need a human person to judge the validity of the computer guesses. In fact, assessing the symptoms is probably going to require a trained professional for ages but, at the same time I think there is massive potential for this kind of AI in these information intensive fields.

There is definitively lots of startup potential in all this.

I see a lot of startups solving easy problems (for good reasons) but this is worth a 10 twitters in value.

If I wasn't doing stuff in the medical imaging field (Brain imaging specifically) I'd go out and build this.

a system which would allow doctors to give a list of symptoms and have the most likely diagnostics listed by probability

Oh, good. Seems every episode of House I watched would contain a scene that began with, "Differential diagnosis for shoulder pain, go!" And then they'd spend ten minutes listing off all the diseases that could cause it. And I'd spend ten minutes thinking "You people really need a database."

I'm not sure that House is an entirely accurate depiction of how diagnostic medicine works. For instance, I'm pretty sure real diagnosticians have to look something up in a book every now and then.

Indeed. I hoped that something like what I imagined either existed or was actively being built. Until this comment, though, my faith in the medical industry was insufficient to justify the belief.

This article was extremely interesting. I remeber wired reporting on a system which would allow doctors to give a list of symptoms and have the most likely diagnostics listed by probability. Not sure of the name but technology like that has incredible potential. I also remember the article mentioning the reluctance of doctors to use it.

I keep reading comments like this and am wondering if anyone has been to an HMO in the past 5 years. This is part of any GP's office with a full EPIC install. I guess many probably don't use it, unless they are forced to. Just wanted to point out this tech already exists in a commercial product. It's not wonderful, it actually kind of sucks. I also wonder why hackers think software like this is going to have "incredible potential," or even be "good."

Why wouldn't it be "good"? If you can only remember the symptoms of 200 diseases, but the machine can remember 18000, why wouldn't it be net-beneficial to always input the symptoms into the machine, as well as performing whatever deductive diagnosis you do normally? Is it just because the UI of existing implementations is horrible?

I thinks this is the whole point, were not just talking about making a wiki of disease but doing some analysis of the case and narrowing down the range of disease possible to something the doctor can analyse.

To the ../../post can you remember the name of one of these systems? it be interesting to see what it does and how it works. I don't believe what's out there is what I'm imagining.

EDIT: reread your post, I just found the side for that EPIC system (terrible name). http://www.epic.com/ The page doesn't inspire confidence in this system being of any quality or does what I am talking about.

It does exactly what you're talking about. The quality is not good, but I don't know why anyone thinks the situation will improve. Countless careers have been wasted on expert systems, recommendation systems, and the like. The best minds doing this sort of thing will always be poached by wall street, where they still fail most of the time. However if it makes people feel better, most HMOs and managed care systems will start requiring their doctors to use such a system, if they don't already.

I understand what you are saying but I'm an optimist at heart. I don't think careers were wasted because most of what's out there is not very good. This stuff is hard to solve but after seeing things like wolfram alpha I can't help but wonder if these systems will improve soon. As for startup attempting this, it's a long shot but those are the ones that make money.

The problem is of expectations. We don't expect computers to write books (creativity is a Hard problem), but we do expect writers to use word processors.

Likewise, we shouldn't expect computers to deduce your ailment and prescribe treatment (critical thinking is also a Hard problem), but we should expect doctors to use computers to remember, search, and sort things for them: the things computers are good at.

We don't need the computer to think—we need it to help the doctor think. People should be working on that, not on systems to replace the doctors themselves.

One such system was Pathfinder, which (I believe) used a relatively simple Bayes net to diagnose within certain categories of disease.

It was able to outperform pathologists in the area it was designed for because (or so I've been told) it didn't face the cognitive biases that human doctors have (like the inability to reason with probabilities, the accessibility bias, etc).

Mycin? http://en.wikipedia.org/wiki/Mycin

Reply on news.ycombinator.com