Slashdot Mirror
Facebook Wants To Use Machine Learning To Make MRIs Faster
Facebook believes they can use machine learning to speed up magnetic resonance imaging (MRI) scans. Computer scientists from the social networking site are working with New York University's medical school on the project. CNNMoney reports: NYU is providing an anonymous dataset of 10,000 MRI exams, a trove that will include as many as three million images of knees, brains and livers. Researchers will use the data to train an algorithm, using a method called deep learning, to recognize the arrangement of bones, muscles, ligaments, and other things that make up the human body. Building this knowledge into the software that powers an MRI machine will allow the AI to create a portion of the image, saving time. Making the tests faster would allow radiologists to perform a wider variety of tests.
It will probably misdiagnose the conservative ones...
While the use of a "legacy system" is a bad thing, the NFC tags inside the badges could be easily read. One vendor offered a free RFID wallet for passersby, who yes, had their badges scanned.
Unusually, the name tag didn't have embedded NFC, rather, an additional tag was used. Remove the tag, and no NFC read.
But the UBM contractor who screwed up.... is a Black Spot on their event.
---- Teach Peace. It's Cheaper Than War.
Am I the only one reading this as "It will speed up the imaging by using CGI to fake part of the image"?
I've abandoned my search for truth; now I'm just looking for some useful delusions.
Wrong story?
Vagina, pussy, snatch, etc are overly gender specific. Please use the term "front hole".
Sincerely,
State of California
You want faster scans? Write normal software that is better or use faster hardware. AI is SHIT and should not be trusted. An MRI should show the ACTUAL SCAN DATA not some made-up crap by some shitty fake-ass excuse for AI.
hipaa says no!
By using the Facebook MRI Scanning Technology, you agree that your MRI scan will be posted to your Timeline once the scan is complete. You also agree that Facebook may retain a copy of the scan for future use and we may share it with our business partners and affiliated companies for educational and marketing purposes. Or you can opt-out, and possibly die. Do you agree?
Wrong story.
---- Teach Peace. It's Cheaper Than War.
I did RTFA and I understand their motivation, but is anyone else annoyed that a medical university has to go to a company like facebook to find a critical mass of machine learning experts to help advance medical technology? That's not facebook's core business. Bless them for planning to open source the results, but... I also can't help feeling like the only reason this article is on /. is because facebook is in the headline. Would it be news if NYU was using their own CS department for this project?
How is Facebook making money off of this
They will not do ANYTHING out of the goodness of their (lack) of heart.
The millennial that doesn't like most of the stuff designed for millennials.
moron :-)
Only I can judge you.
Anytime anyone wants to train an AI system its always medical images. I did this as my undergrad final year project almost 20 years back. Sure the algos are probably better now but there is something about Medical Images which makes it satisfying for young idealist students to use for their project. Once the algo is perfected it can also be used for Face detection in kegger pictures.
Facebook has a big problem. The govt is asking them to police offensive images. They cannot hire enough humans to do it so they need AI.
**Life is too short to be serious**
... and it didn't turn out well.
I worked for Mobil Oil.
They made so much money, they had a cash store (ca. 1986) that was obscene and the shareholders wanted them to do something with it that would make more money.
Mobil bought out an insurance company, went self-insured, and sold policies to any and all.
They also went into the land-grabbing business and built Reston, Va. from the ground up.
They bought Montgomery Ward, too.
They folded shortly after I retired from there.
--
When companies step away from their core competencies, it's an indicator that the shit's fixin' to hit the fan.
--
Facebook is wandering all over the place. The recent scandals including scamming shareholders and advertisers with false info and not giving a single solitary shit about ethics while diversifying like this will be part of their demise.
That's not going to happen any time soon, though.
It little behooves the best of us to comment on the rest of us.
What do MRI algorithms have to do with social media?
Don't worry. We've all done it. It's all these damn browser tabs the kids have these days. Confusing as hell.
You are welcome on my lawn.
Stick to what you're good at, Zuck - being a creepy fucking cunt.
I do not want your cheap brainburning drugs. They are useless for work. And I am a working man today.
Every MR protocol is an engineering protocol that balances resolution (voxel size), signal-to-noise, field of view, and scanning time. If you increase resolution, you generally have to decrease your field of view, decrease SNR, or increase your scanning time. It's easy to make a 6 second scan - in fact, we do it every time we scan with what's called a localizer - a wide FOV, low resolution image that the techs use to orient the diagnostic images to be acquired. The localizer is not diagnostic, but I've caught large tumors on these images that weren't included in the more limited FOV of the diagnostic images.
Part of scanning time is "baked in" due to physical constraints - the protons take a certain amount of time to flip back after you've nudged them with an RF pulse. This time is reduced with higher-field magnets (at the cost of decreased SNR), but obviously AI can't alter physics. I imagine this "speed up" happens on the reconstruction side, where the sampled frequency domain data is converted to a spatial domain image. There are some protocols already that "undersample" k-space to speed up acquisition. In this case you're taking advantage of some of the natural "symmetry" of k-space imaging to interpolate "holes" in your sample data. Almost all clinical scans that can use this technique do, because time is money in the MR and because trade-offs in image quality using these methods are generally outweighed by those caused by patient movement, a phenomenon which tends to be more of a problem the longer people have been in the scanner.
But can a software solution improve on this by an order of magnitude? I highly doubt it. Besides the physical constraints, there is ALWAYS a trade-off. These trade-offs are very well understood and finely tuned to the application on hand. I'm guessing some 20-something Facebook engineer saw a k-space image for the first time and figured he/she figured they were the first to notice the apparent symmetry and potential for interpolation.
This idea isn't new - compressed sensing was pioneered in MRI a decade ago - sounds like these guys are amateurs...
Comment removed based on user account deletion