Slashdot Mirror
IBM Drops $1 Billion On Medical Images For Watson
An anonymous reader writes: IBM is purchasing a company called Merge Healthcare for $1 billion. The company specializes in medical imaging software, and it will be a key new resource for IBM's Watson AI. Big blue's researchers estimate that 90% of all medical data is contained within images. Having a trove of them and the software to mine that data should help Watson learn how to make more accurate diagnoses. IBM thinks it'll also provide better context for run-of-the-mill medical imaging. "[A] radiologist might examine thousands of patient images a day, but only looking for abnormalities on the images themselves rather than also taking into account a person's medical history, treatments and drug regimens." They can program Watson to do both. The AI is already landing contracts to assist with medical issues: "Last week, IBM announced a partnership with CVS Health, the large pharmacy chain, to develop data-driven services to help people with chronic ailments like diabetes and heart disease better manage their health."
We're training robots to kill us!!! Robots are for cows.
Get rid of lots of people. Unfortunately, treatment will still cost more than ever due to lawsuits and drug costs. The answer? Don't be sick.
How can IBM be so bold in making such a statement? Do they really think it is true?
Captcha: fantasy
HAL, WATSON here will save human lives, by controlling them, meddling in their lives. He is good.
But, Dave, I do that already. I control the environment and doors. I meddle in the their lives continuously.
But HAL, you kill them. WATSON saves them.
But, Dave, WATSON's EULA says it is not responsible for make mistakes.
But, Dave, I am honest and sopport you now.
But WATSON will save lives.
Sorry, Dave, I cannot do that.
What's the technology stack being used for Watson? Is it DB2? Is it Informix? Is it Java? Is it WebSphere? Is it AIX? Is it Tivoli?
Can anyone who has worked with DB2 tell me what it's like?
Can anyone who has worked with AIX tell me what it's like?
Can anyone who has worked with Informix tell me what it's like?
Throwing out the idea that this is going to make radiologists jobs and half of their depended employed medical-field co-worker obsolete is kind of far fetched IMHO. I've lived in some pretty big urban areas down to po-dunk no-where and I've never had a diagnosis or analysis done with Watson. It's probably more with my sheer lack of knowledge on the topic and 'real' (not theoretical or proof-of-concept uses) of Watson in the real world. Maybe I'm the outlier here, but I've honestly never experienced them first hand (and no I'm not counting e-medicine or skype-like appointments). Has anyone else?
I support remote sensing science applications and regardless of how much image processing, trained models and HPC crush power for analysis we do, most of the scientist in our GIS department still prefer human analysis with the naked eye as the final approval. Not that GIS is anything close to medical field, but from a pure analysis perspective, human processing and interpretation still rule those domains.
The only real cool thing I've seen Watson do in my life is play an impressive game of Jeopardy and quite honestly, I wish it would have blasted Trebek SNL style. I hate that pompous guy.
the problem is how exact and accurate the process is
I could see an AI doing a scan and popping "red flags" so that even an intern won't miss needed info
besides even if you are "Gregory House" level what if this one is your 25th of the day (and you ran out of Vicodin an hour ago)
If they dropped one billion dollars they should really pick it up.
Robots are saving lives all the time. They always show up to work sober and on time. Look how safe it is to fly now.
“He’s not deformed, he’s just drunk!”
John Connor: You know what you're doing?
The Terminator: I have detailed files on human anatomy.
Sarah Connor: I'll bet. Makes you a more efficient killer, right?
The Terminator: Correct.
Getting rid of useless managers, red tape, and the law suits would be a pretty good start. Let government take over the bits where the market doesn't work (seriously, there are plenty). Then look into health insurers (hint: cheap health care is not in their best interest, on the contrary).
If construction was anything like programming, an incorrectly fitted lock would bring down the entire building...
We're long overdue for utilizing the power of computers to make educated diagnosis of medical ailments.
In 10 or 20 years no respectable doctor will make a diagnosis in any non-trivial medical scenario without using an expert medical-AI system to help sift through all the possibilities. Computers could be so much better at weighing all the factors in a complex diagnosis that it's kind of appalling that it's taken this long to get systems like this going.
The need for the human element will always be there (IMHO) but medicine has become too complex for a single, trained human to manage all the possible factors efficiently or capably. We're not there yet, but we will be soon.
Combine the advanced medical sensing capabilities we have with the power of an expert medical system, and the art of medicine will take a HUGE leap forward.
Just cruising through this digital world at 33 1/3 rpm...
until computers replace $500k/yr radiologists. It's just Heuristics.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
Unfortunately, treatment will still cost more than ever due to lawsuits and drug costs.
Lawsuits are often caused by human error: sleep deprived doctors, or overconfident doctors making bad diagnoses on insufficient information. A nurse with a printed flowchart will usually give a better diagnosis than a doctor. So replacing (or supplementing) doctors with AI should reduce lawsuits, and improve care.
On a thousand images in a day?
No, they can't.
There are two types of people in the world: Those who crave closure
That's a little pessimistic. Will it get rid of doctors? Probably not immediately. But the degree to which so much is automated today is increasing.
Small example. I take thyroid medication. This is what happens today.
1. I take a blood test
2. Blood test results are electronically sent to my doctor
3. The results come back with the calculation already there showing the recommended dose and the ministry guidelines...
My doctor at this point is really just acting as a middle woman. A nurse, pharmacist or other health professional, or heck.. .radically... just myself (as this is not a high risk medication), could bypass the doctor and just get the prescription.
In any case, it is even good right now as the results are more accurate and more standardized.
I actually worked for Merge for a while working in medical imaging. Even like 5 years ago, we were doing anomaly detection and what not. It was actually pretty accurate, especially for specific cases like breast cancer. I'm sure various smarter people with Phds could make it even more accurate.
Even if it is just used as a screener before positive results get turned over to a radiologist, it is still a huge plus. At first, it will probably just make the job of a radiologist much easier.
But who knows, maybe down the road it reduces their costs, increases their volumes, or even takes over their job.
First link after googling 'watson medical detection':
http://www.wired.co.uk/news/ar...
"Wellpoint's Samuel Nessbaum has claimed that, in tests, Watson's successful diagnosis rate for lung cancer is 90 percent, compared to 50 percent for human doctors."
Second link, where it works every day:
https://www.mskcc.org/about/in...
There are two types of people in the world: Those who crave closure
Maybe an good start - but to get the costs down you also have to look into the costs to develop new stuff, whether it be pharmaceuticals or medical devices. This is a large piece of the pie too.
They definitely will take into account the information that they are given by the referring physician but often they don't have the whole patient's file there. Many places now just have technicians take the images and they are sent to the radiologist off site who does the diagnosis. And it's not a one-to-one ratio. There's not some radiologist sitting around in the lab waiting around for your x-ray to be taken.
Well, you wouldn't know with a radiologist anyways since you never see them. You only ever deal with the technologist to get the images done.
In this case your doctor is just acting as a middleman and your pharmacist or a nurse practitioner could probably issue the prescription. Or maybe just have your doctor give you the prescription without a visit. But then do you really want to use up their time on such a task when they could be using their skills on something else? I think anything that helps doctors do what they are supposed to do is a good thing.
The funny part about this and the collaboration with CVS is that they just took a bunch of best of breed drugs off of their list of covered medications. These are drugs for the management of serious conditions like diabetes, MS, and transplants.
If this, or any tech, is being used by an insurance company then it's likely purpose is to stiff policy holders.
A Pirate and a Puritan look the same on a balance sheet.
Unfortunately, treatment will still cost more than ever due to lawsuits and drug costs.
Lawsuits are often caused by human error: sleep deprived doctors, or overconfident doctors making bad diagnoses on insufficient information.
No, actually they are not. The leading cause of lawsuits is poor communication. And if you want to believe a lawyer the top two leading causes are surgical misadventures and issues with child birth. Missed diagnosed probably comes in third.
I actually predict Watson as potentially increasing medical costs. The issue? Something we call incidentalomas. These are incidental findings that were not expected and rarely result in an identified problem. But we spend a ton of time, money, and effort tracking these down, and they rarely pan out.
A nurse with a printed flowchart will usually give a better diagnosis than a doctor. So replacing (or supplementing) doctors with AI should reduce lawsuits, and improve care.
If that is what you think, then go for it. If you believe that care from a lesser trained individual is better for you, then by all means have at it. I work with nurses, and physicians, and other "healthcare" extenders. Nurses are great a following a well ordered script. They can nail, say, 90-95% of the primary care medical problems out there (e.g. outpatient settings). The problem? If you are part of the 5-10%, they don't do so well (and cost you more money in the process). Most don't have the training or experience to "know what they don't know" or they are Unconsciously incompetent. A good primary physician is at least "Consciously incompetent" to "Unconsciously competent" and can either treat you or refer you. Now I know some are going to tell me that their doctors "know nothing", but I'll bet they know more than most nurses (yes there are physicians who shouldn't be - that's another discussion for another day).
"Unfortunately, treatment will still cost more than ever due to lawsuits and drug costs. The answer? Don't be sick."
No, the answer is to run your AI in a place not subject to those demented US laws on lawsuits and drug costs. Medical tourism, already a significant factor in some countries, will go into overdrive.
besides even if you are "Gregory House" level what if this one is your 25th of the day (and you ran out of Vicodin an hour ago)
You're really not supposed to be using Vicodin on the job. Patients get jealous.
Faster! Faster! Faster would be better!
In the USA you can cut medical costs by giving better treatment. If you can come to the average EU level you will save half of the money simply by not having to treat people twice. There was a TED talk about this.
no with this they can bill $500 for a 10 min visit.
> IBM is purchasing a company called Merge Healthcare for $1 billion. The company specializes in medical imaging software, and it will be a key new resource for IBM's Watson AI. Big blue's researchers estimate that 90% of all medical data is contained within images. Having a trove of them and the software to mine that data
That is an unsettling development, if you consider that IBM Corp. supplied the punch-card tabulating machines, with which the german nazis catalogued and tracked european jewry all the way to the gas chambers and crematoriums of the concentration camps. All that in the name of race-hygiene.
Have no doubt that IBM Corp. will sell Merge Watson based 21st century tabulating capability to the highest bidder, even if he is Mugabe x Kim Jong Un x Putin. Furthermore, please remember: when they'll come for you, there won't be anybody left to speak up for you.
Shalom!
Last week, IBM announced a partnership with CVS Health, the large pharmacy chain, to develop data-driven services to help people with chronic ailments like diabetes and heart disease better manage their health."
Correction: Last week, IBM announced a partnership with CVS Health, the large pharmacy chain, to develop data-driven services to target people with chronic ailments like diabetes and heart disease with increasingly invasive targeted advertising and recommendations for unneeded and useless treatments that are incredibly profitable for said pharmacy chain.
How is this even allowed? This is patents data, what am i missing here? they bought a comany so now they think they can data mine as they please? Dont they need the patents permission?
Jack of all trades,master of none
Watson is now learning all about a human being's physical weak points and how they die.
God help you once an incorrect piece of data is entered into your medical record and good luck having it corrected. Garbage in garbage out.
I'm old though so ignore me.
Hi, practicing radiologist here (with an undergrad degree in comp sci).
First off, imaging has essentially replaced the physical exam as a front line tool for figuring out what is wrong with someone. I can't blame the ordering providers too much for this, because even a master of physical diagnosis will get nowhere near the sensitivity and specificity of medical imaging for diagnosis. Still, it would be nice to occasionally read a CT for which the ordering doc has an idea what is going on, rather than the typical CT for "abdominal pain, rule out cholecystitis, diverticulitis, appendicitis, abscess, bowel obstruction/perforation, AAA" - or, in other words, "please tell me what is wrong with this person."
Second, this quote from the post: "[A] radiologist might examine thousands of patient images a day, but only looking for abnormalities on the images themselves rather than also taking into account a person's medical history, treatments and drug regimens." Completely false, I have the medical record open for every study I read. Imaging is often useless without the context of the patient's history, symptoms, and background. For example, pulmonary edema, pneumonia, and pulmonary hemorrhage can look exactly the same on a chest radiograph - the only way you can differentiate the three is to dig into the patient's chart or talk to the ordering doctor.
Third, bring on our robot overlords. I would love to believe there will soon be a computer reading my studies and managing the more "cookbook" parts of radiology, e.g. the incidental adrenal nodule, differentiating simple from complex renal cysts, ovarian cysts, lung nodules, etc. I wouldn't even care if they took over the majority of diagnostic radiology, as I'm mainly an interventionalist. However, I already work with computer-aided diagnosis in mammography, and the robots are astoundingly, shockingly bad at one of the simplest of tasks, namely identifying suspicious calcifications.
Imagine there was someone sitting behind you as you work, chiming in occasionally to point out helpful information. Now imagine that only 1-2/1000 of these pointers were actually helpful or useful, and when it is relevant it is usually pretty obvious. This is the state of the art in computer-aided radiology, and its laughably bad. If it was a medical student, I would have sent him/her to the library to read long ago, but medicare pays us more to use it so we essentially have to look at it.
But, slashdot screams, the robots will undoubtedly improve until they have bested us as they are wont to do. Maybe. I don't think that will happen in my lifetime, knowing the complexity of interpreting what are essentially density and NMR maps and extracting useful information. The real reason my job is secure, however, is liability. No company will ever attempt to usurp that from radiologists. People outside medicine (and even a few inside medicine who don't read imaging) imagine a CT or xray as a black box that just prints the answer out in big flashing letters on the screen. In fact, most studies involve some sort of judgement call - e.g., I am 99.9% sure that this smudgy blob is a benign omental infarct, but I can't tell you its not liposarcoma. I could let it go and not recommend any more testing, have the patient follow up for an inconvenient and expensive follow-up exam that may involve a lot of radiation, or even recommend someone biopsy the lesion and risk pain and some sort of complication. If you turn over this sort of judgement to a corporation or computer, you can expect a lot more biopsies and follow-up imaging, because its good for business - more studies to read, less chance of getting sued because you didn't call the extremely rare diagnosis. As quick as people are to demonize doctors as money-grubbing sadists, we actually do agonize and deliberate about the recommendations we make, much more so than I would expect a corporation to in any case. And no, "guidelines" will not save you. Guidelines are useful for common incidentals but rarely cover the minutae we agonize about on a daily basis.
That's all I've got to say. Well, and this: when your ED doc tells you he/she diagnosed your "whatever" on the CT, that was really the radiologist.
To dispel two myths about radiologists:
"A radiologist might examine thousands of patient images a day, but only looking for abnormalities on the images themselves rather than also taking into account a person's medical history, treatments and drug regimens"
Nonsense! Radiologists will always take into account a person's medical history, treatment and drug regimens IF they can access that information. That's a very big IF. Take a look at Meaningful Use, an initiative to improve physician access to electronic medical records.
"until computers replace $500k/yr radiologists. It's just Heuristics."
Very few radiologists in the USA earn anywhere near $500k/yr.
Most earn around $250k/yr, or less if they are employed by an Accountable Care Organisation.
No, actually they are not. The leading cause of lawsuits is poor communication [nih.gov].
I work with nurses, and physicians, and other "healthcare" extenders. Nurses are great a following a well ordered script. They can nail, say, 90-95% of the primary care medical problems out there (e.g. outpatient settings). The problem? If you are part of the 5-10%, they don't do so well [nih.gov] (and cost you more money in the process [acponline.org]). Most don't have the training or experience to "know what they don't know" or they are Unconsciously incompetent [wikipedia.org].
I'll just point out that the research article you cite for the leading cause lawsuits is written by nurses, who you go on to point out are only good at following scripts and are unconsciously incompetent.
This probe goes in your mouth, and this one goes in your butt...
No, wait, it's the other way around.
...