Is anybody familiar with this type of research? I know that nothing can replace human judgement, but imagine... if you had supersmart programs preview your slides for you, wouldn't it make your life at least a bit easier?
Can computers read slides?
- Thread starter Nilf
- Start date
- Joined
- Jul 25, 2001
- Messages
- 9,319
- Reaction score
- 125
Someday it'll all happen. Someday computers will read films and slides with unparalleled precision. Someday robots will perform all surgeries. I think we're so far from achieving that sort of AI that probably our great-great-grandchildren will still be asking this question.
- Joined
- Dec 11, 2003
- Messages
- 6,617
- Reaction score
- 14
There isn't even a cure to the common cold yet.
- Joined
- Feb 22, 2003
- Messages
- 106
- Reaction score
- 0
I think all those things will be possible in the near future, but will patients and physicians trust the clinical judgement of a machine? who will be liable for "computer error"? Similarly, the technology exists today for airliners to fly "by themeselves" using GPS, but who the hell will board a plane with no pilot 😱 .
- Joined
- May 31, 2003
- Messages
- 599
- Reaction score
- 0
Our cytology laboratory in Rochester participated in a study with just such a machine. The company was called NeoPath I think (they've gone through a couple of name changes, so I'm not sure if they still go by that name or not. The name of the machine was the autopap). A couple of our cytopathologists were involved with the company. Anyhow, I'll try to explain how the machine works. Basically, the computer "learns" what a normal pap should look like by you, the end user, running a litany of "negatives" through. There are special algorithms in the machine that do this. I think you're supposed to put in a bunch of abnormal slides, too. Again, not sure of the exact numbers of each.
The autopap then "screens" each slide. The slides that it deems as negative are given a "No Further Review" classification, meaning that the cytotechnologist or Pathologist does not need to review the slide. Others are flagged for Review, or given an error code of some reason (i.e. there are large air bubbles, the coverslip isn't square on the slide, etc.). It doesn't "diagnose" the pap (so it cannot differentiate a Low Grade Squamous Intraepithelial Lesion from an ASCUS, for instance). If you use a special program with the machine, it will read the barcode on the sample, and take you to each area of the slide that was "flagged" as abnormal (somewhat similar to the now defunct Pap Net machines of a few years ago), or the cytotechnologist can rescreen it.
The machine was surprisingly good at detecting slides for "No Further Review", and indeed, it is FDA approved to do primary screening. So, you can technically have the machine call the slide negative and never look at it (but I'm pretty sure you still need to have at least 10% of those negatives "quality control" rescreened by another cytotech, just like the others).
While the machine was pretty good, I'm not sure when or if they will catch on. It may be the perfect addition to a small lab in an underserved area (like here in the Caribbean, for example, where cytotechnologists are virtually non-existent), but in our lab, it was more of a pain. Lots of error codes. If the slides aren't close to perfect (as far as coverslipping and staining), the machine kicks them out. I doubt a machine will ever be able to make a definitive diagnosis--especially an abnormal one. There is just no substitute for the human eye or human judgement. Even the coulter counters in hematology can only flag something as "Abnormal WBC (or PLT or RBC) Population". In the end, I think humans will always have to say just how "abnormal" something really is.
The autopap then "screens" each slide. The slides that it deems as negative are given a "No Further Review" classification, meaning that the cytotechnologist or Pathologist does not need to review the slide. Others are flagged for Review, or given an error code of some reason (i.e. there are large air bubbles, the coverslip isn't square on the slide, etc.). It doesn't "diagnose" the pap (so it cannot differentiate a Low Grade Squamous Intraepithelial Lesion from an ASCUS, for instance). If you use a special program with the machine, it will read the barcode on the sample, and take you to each area of the slide that was "flagged" as abnormal (somewhat similar to the now defunct Pap Net machines of a few years ago), or the cytotechnologist can rescreen it.
The machine was surprisingly good at detecting slides for "No Further Review", and indeed, it is FDA approved to do primary screening. So, you can technically have the machine call the slide negative and never look at it (but I'm pretty sure you still need to have at least 10% of those negatives "quality control" rescreened by another cytotech, just like the others).
While the machine was pretty good, I'm not sure when or if they will catch on. It may be the perfect addition to a small lab in an underserved area (like here in the Caribbean, for example, where cytotechnologists are virtually non-existent), but in our lab, it was more of a pain. Lots of error codes. If the slides aren't close to perfect (as far as coverslipping and staining), the machine kicks them out. I doubt a machine will ever be able to make a definitive diagnosis--especially an abnormal one. There is just no substitute for the human eye or human judgement. Even the coulter counters in hematology can only flag something as "Abnormal WBC (or PLT or RBC) Population". In the end, I think humans will always have to say just how "abnormal" something really is.
- Joined
- Jul 25, 2001
- Messages
- 9,319
- Reaction score
- 125
Brian Pavlovitz said:
Well, "always" is a long time. Someday AI will have advanced to the point where it can pass the Turing Test (i.e., if you have a conversation with it, it will be indistinguishable from another human). We won't see this in our lifetimes, but it'll happen eventually. A century from now (or two, whatever...) the technology will have progressed to where the machine makes fewer errors than even the most expert human. And then there will be no substitute for the computer eye and the computer judgement, since the computer won't be tired, bored, distracted, etc.
There once was a time when the notion of a computer chess program beating a grand master was absurd, because it was believed that no computer could grasp the transcendent logic of the game. Whether or not any "grasping" is going on will remain up to debate, but the fact remains that sheer processing power will carry the day.
- Joined
- Aug 15, 2003
- Messages
- 28,059
- Reaction score
- 441
Yup. 2 constants in medicine: Never say never, and never say always. I agree that it probably isn't something we have to worry about in the near future. I heard a discussion once about the potential future of computers reading histology slides (or at least, screening them) and the problems as of now is that the computer has a difficult time distinguishing normal variants from true abnormal pathology. Severe inflammation can be read as tumor, and vice versa, which as we can see in an example such as pancreatic cancer vs pancreatitis, is often the differential anyway. A computer is going to have to get very good at this and require a lot of programming in of algorithms, pattern variations, etc.
What I can see in the future is a computer being used to screen tumor cells via flow cytometry or molecular diagnostics, looking for expression of numerous markers, proteins, etc, so that exact classifications can be made. Thus, the histology may potentially be rendered worthless. Again, I think any changes like this would be slow to come, and of course there will always (whoops! Shouldn't say that word!) be people who insist on looking at slides under a scope. These individuals will hold out, fight the coming of the robots at all costs, and probably eventually sabotage them so that they fail and humans are kept employed. Of course, by this point the robots will be smart enough to fight back, and James Cameron's visions of the future will become horribly true.
Who knows what will happen. Perhaps one day pathologists will be obsolete, but I would imagine by that time nearly every job will be obsolete, and then I don't know what the politicians are going to fight about. We can be fairly certain that politicians will never be replaced by robots!
If you live long enough to see this, remember to stay away from the Soylent Green. It's made out of people. People, I tell you.
What I can see in the future is a computer being used to screen tumor cells via flow cytometry or molecular diagnostics, looking for expression of numerous markers, proteins, etc, so that exact classifications can be made. Thus, the histology may potentially be rendered worthless. Again, I think any changes like this would be slow to come, and of course there will always (whoops! Shouldn't say that word!) be people who insist on looking at slides under a scope. These individuals will hold out, fight the coming of the robots at all costs, and probably eventually sabotage them so that they fail and humans are kept employed. Of course, by this point the robots will be smart enough to fight back, and James Cameron's visions of the future will become horribly true.
Who knows what will happen. Perhaps one day pathologists will be obsolete, but I would imagine by that time nearly every job will be obsolete, and then I don't know what the politicians are going to fight about. We can be fairly certain that politicians will never be replaced by robots!
If you live long enough to see this, remember to stay away from the Soylent Green. It's made out of people. People, I tell you.
- Joined
- Mar 9, 2004
- Messages
- 160
- Reaction score
- 0
yaah said:
I am pretty sure that most of the politicians are already robots. You can pretty much predict what they are going to say before they say it just by looking to see which side of the aisle that sit on. However, John Kerry does flip flop a lot...maybe his circuitry is faulty or maybe he needs to reboot...
