This could lead to excellence (or serious injury)

I was at the dentist last week for my bi-annual cleaning and checkup. Ultimately everything was fine, but during the exam the hygienist pulled out a new method of taking X-ray images of my teeth. It was digital (and it was the good kind of digital). I still needed the heavy lead sheet on my chest, the X-ray gun was still there, and I still needed to clench the image receiver between my teeth (but with a wire trailing out to a computer somewhere else in the building). However, there was no film to change after each exposure. So, the hygienist would take a picture, then just reposition the receiver in my mouth before taking the next one.

On my way out, I caught a glimpse of the images from my teeth on a laptop that the dentist had brought in with him. They looked like any other film-based X-ray I've had of my teeth. This got me to thinking: If there is now a digital image, why couldn't a program be written to analyze the image and make recommendations when warranted?

Two months ago I had an opthamologist appointment. During that examination, a digital picture of the inner parts of my eyes was taken and analyzed by the doctor. Here again, why couldn't a program analyze the image and point out potential problem areas that are seen? As a corollary, would people be willing to trust an algorithm with diagnosing or detecting cavities, degenerative eye diseases, or something more monumental like cancer?

I suspect that biggest criticism against using computers for diagnostics through digital images would be a question of trust. With all the known problems with software and bugs that can be found or that aren't found until they cause a catastrophic failure (see Fatal Defect by Ivars Peterson for some scary stories), would you ever trust a software diagnosis? Doctors can get things wrong, too. For something serious, would you trust two or three or four different doctors that could reach a consensus after separately examining the evidence? I would.

Imagine if there were two or three or four different diagnostic algorithms available to "look over" the digital images and put forth a diagnosis. Would that make such a diagnosis any more trustworthy? This would be something more than just redundant systems running the same code on separate processors and comparing the answers. Different algorithms, created by different programmers with appropriate medical consultation (also different from the medical consultants used by any of the other development teams), will take different approaches to diagnosis of the same data. If all four or three of four yield the same diagnosis, would you trust that outcome?

The above scenario of multiple diagnostic programs running at the same time in analysis of the same data set is an obvious parallel execution. (You knew I'd get around to it eventually, right?) With more cores, why not envision 8 or 16 different algorithms all processing at the same time to come up with a diagnosis from digital images. I doubt you could afford to consult with 16 different specialists, especially in a timely manner. Automating analysis, certainly for more benign conditions, could be a boon to health care.

I don't think I even need to mention emergency situations where time-critical and correct diagnoses are the difference between life and death (e.g., automobile accidents, natural disasters, battlefields). 

Not to worry, there won't be computer programs taking on cases like you see on House, M.D. any time soon (though that would make a neat sub-plot for next season, especially with what we've seen of IBM's Watson project). If and whenever such pilot programs come to be, they will be monitored heavily by human doctors to oversee the process and check the results. Even so, I'd be comfortable if my doctors consulted computer-generated recommendations about what my X-rays or MRI scans might reveal about my health. On Parallel Programming Talk #102, we heard about parallel computations constrtucting a digital model of a patient's colon that doctors could examine. Why not add an analysis step to look for common problems and give the doctors a list of potential spots to focus in on?

Of course, the next step would be an even more automated healthcare experience. Would you be willing to let a robot handle the examination? Thinking about the appartus with a camera that was pushed up near my eye during the exam, I'm not sure if I'm ready to allow a robot make the positioning decisions and movements of something that close to such a vital organ. Also, even though I'm sure my head could be immobilized sufficiently, I wouldn't want a robot drilling out and filling a cavity. (And don't get me started about letting a robot handle the not-good kind of digital examination.)

Leaving aside Dr. "Bones" McCoy from Star Trek, I think we should be heading in the direction of automating health diagnostics. We'll still need doctors and specialists around forever. Medical research on new diseases and new techniques for treatment, as well as consulting on diagnostic software, will be critical to the industry and for instilling trust in the software. Parallel computation should be able to take over some of the more mundane and routine parts of a doctor's job and be able to do it quickly, efficiently, and extensively enough to become a trusted part of health care.
For more complete information about compiler optimizations, see our Optimization Notice.


sveyda's picture

This is why AI died. We could create a system that could diagnose a patient better than a collection of the finest specialists on earth. But who would accept a prognosis from a machine? If it were wrong, who would be liable? The same is true of almost any important computer-based decision- if the decision is incorrect (a who guarantees 100% in this world), who is responsible? That is what killed rules-base (knowledge-based) expert systems.

Clay B.'s picture

And even now, I see commercials for a portable blood sugar monitoring device that claims to test twice with the same drop of blood. That must be two different methods of analysis (in parallel!). Otherwise, why would the manufacturer claim that you get more accurate results or better reliability? If you take the same sample and run the same test on it twice, you should get the same results; if you take the same sample and run two different--yet equivalent--tests and get the same results, then you have good confirmation.

TomServo's picture

Since you brought up Star Trek, don't forget the holographic doctor in "Star Trek:Voyager". He must have been a parallel program back in there somewhere. It looked like a robot doctor fixing Luke's hand at the end of "Empire".

Clay B.'s picture

@Jérôme - I think you've hit upon an intermediate state, but since it doesn't involve parallel computing I didn't consider it.

You're right, cheaper imaging equipment and widespread networking can spread needed medical technology to parts of the world that don't have access to more expensive technolgy. With digital images, consultation with outside specialists is much easier and faster. Take it one more step and consider the case where the new imaging tech becomes more widespread than there is medical personnel to cover or consult with. At that point, I think some form of automated review of digital images would be economical and feasible. It's not computers taking over jobs, but using computers to handle a larger influx of data (patients) and to process that data in a timely manner.

Even so, this is far far away in the future.

jerome-muffat-meridol (Intel)'s picture

In a world where automated call centers can't recognize my voice, where face recognition is still a bit of a toy, where optical character recognition still can't deal with handwriting very well, you are looking at ideas that are quite faraway in the future.

Now, instead of trying to replace humans, why not make use of them instead? Now that your xray is digital, you can have a specialist from the other side of the world give advice. Wouldn't that be amazing for populations that are missing doctors when we have so many in "developed" countries? Isn't the case you mention more of an argument for smaller, leaner, cheaper devices that the whole world could afford, use and become a better place?

This doesn't preclude the need for processing power, as doctors collect xrays and tag them based on pathology, age group and other various metadata, it becomes a huge corpus that they could leverage as reference. I can see computer vision and image processing techniques bringing some powerful features to such a search engine. Of course, there is also the visualisation problem:

The key is to understand how the computer can help you do a better job, rather than how it might replace you.

Kathy F. (Intel)'s picture

I can't help but think of the high costs associated with being a physician, especially mal-practice insurance. That is putting a lot of doctors out of business/practice.What you suggest Clay could help to bridge a gap that is continuing to widen. I should send this blog's link to the President! :-0

Add a Comment

Have a technical question? Visit our forums. Have site or software product issues? Contact support.