Slashdot Mirror

← Back to Stories (view on slashdot.org)

Nvidia Researchers Generate Synthetic Brain MRI Images For AI Research (zdnet.com)

Posted by msmash on from the pushing-the-limits dept.

AI holds a great deal of promise for medical professionals who want to get the most out of medical imaging. However, when it comes to studying brain tumors, there's an inherent problem with the data: abnormal brain images are, by definition, uncommon. New research from Nvidia aims to solve that. From a report: A group of researchers from Nvidia, the Mayo Clinic, and the MGH & BWH Center for Clinical Data Science this weekend are presenting a paper on their work using generative adversarial networks (GANs) to create synthetic brain MRI images. GANs are effectively two AI systems that are pitted against each other -- one that creates synthetic results within a category, and one that identifies the fake results. Working against each other, they both improve. GANs could help expand the data sets that doctors and researchers have to work with, especially when it comes to particularly rare brain diseases.

48 comments

  1. Link to Paper by Anonymous Coward · · Score: 2, Interesting

    On GAN's generally, since no actual research is linked to, here: https://arxiv.org/pdf/1406.2661.pdf

    1. Re:Link to Paper by ShanghaiBill · · Score: 4, Informative

      On GAN's generally, since no actual research is linked to, here: https://arxiv.org/pdf/1406.266...

      That was a profoundly influential paper. A must read for anyone interested in modern AI.

      Ian Goodfellow, the primary author, also co-authored Deep Learning, the best book available for learning about deep neural nets.

  2. GANs for data augmentation? by tempmpi · · Score: 4, Insightful

    I wonder if this can actually work. If you don't have enough images to train a classifier, why would training a GAN work? And even if training a GAN works, those images won't contain any information about tumors that were not already contained in the original images.

    --
    Jan
    1. Re:GANs for data augmentation? by Anonymous Coward · · Score: 0, Informative

      This is the most useless comment I have ever seen. You don't address the concern,
      which I think is perfectly valid, then give some useless facts and comparison. What
      is this? I want my 10 seconds back.

    2. Re:GANs for data augmentation? by sg_oneill · · Score: 1

      Using a GAN to generate fakes won't produce novel data that wasn't in the originals, but what it can do is pull out features common to the particular original dataset to throw into new falsified pseudo scans which can be used for further training or study. It kind of seems tit-about-ass, but this is a job GANs have proven adept at in other fields.

      Think of it this way;-

      Algorithm 1 says "Hey, I'm having trouble understanding these diagrams. I dont have enough examples to study."

      Algorithm 2 says: "No worry little buddy, I'll draw you up some more diagrams!!!!"

      Human says: "Well why not just get algorithm 2 to understand the diagram?"

      Algorithm 2 says: "Sorry pal.All I know how to do is draw diagrams for Algorithm 1!!!!!!"

      --
      Excuse the Unicode crap in my posts. That's an apostrophe, and slashdot is busted.
    3. Re: GANs for data augmentation? by Anonymous Coward · · Score: 0

      Isnâ(TM)t synthetic data produced by a GAN the thing that screwed up IBM Waysonâ(TM)s foray into cancer treatment , and led to it producing potentially fatal recommendations ?

    4. Re:GANs for data augmentation? by Hodr · · Score: 1

      My thought is similar. This is fine if you want to train an AI to be "as good" as a classically trained person, or to produce lots of images for helping some trained in the normal process do so more easily.

      What it doesn't do is take advantage of the primary benefit of applying AI to these sorts of problems, recognizing patterns that have not been recognized before (and therefore wouldn't be identified in the "rules for identifying tumors" or likely included in producing synthetic imagery).

    5. Re:GANs for data augmentation? by Anonymous Coward · · Score: 0

      Have you seen creimer's children band video [youtu.be]? Holy shit! That video got hundreds of view [twitter.com] with 95% coming from outside of the United States and the top three nations are well known for sex tourism. It doesn't surprise me that Slashdot has so many pedobears.

      So basically, you are bragging about providing video material to pedophiles and sex tourists and you do not see any problems with it as long as it brings views to your youtube channel.

      Poor Chris, sad, very sad...

      How long will it be before you do the right thing and take that video off line?

    6. Re:GANs for data augmentation? by Anonymous Coward · · Score: 0

      About learning how to pronounce yourself first before trying to teach us how to pronounce Chris?

      Maybe you should first learn to speak English too, it sure helps with pronunciation.

    7. Re:GANs for data augmentation? by Anonymous Coward · · Score: 0

      Creimer speaks well for an African-American who is a registered Republican and makes less than $10K per year in Silicon Valley.

    8. Re:GANs for data augmentation? by RandCraw · · Score: 2

      Yes, and a GAN can easily combine features in unnatural ways that never existed in nature.

      Unless the natural probability distribution of each clinically important feature can be accurately modeled by the deep net, combinatoric perversity is all too likely. (Features combos will arise synthetically that do not occur together in nature, but these images won't be excluded due to the lack of counterexamples in the training set).

      I work in medical image analysis professionally, and I'm certain no physician or biologist would approve of faking images to learn about abnormalities / disease, no matter how 'plausible' they look. (However, GANs might be warranted if the objective is only to model the range of natural anatomic variation, rather than unnatural variations.)

      There are better ways to augment pathology image data that avoid fakery.

    9. Re:GANs for data augmentation? by ceoyoyo · · Score: 1

      It doesn't really. GANs are cool, and having something that can generate fake MRIs is cool, but how do you use it? You can't write a paper anymore saying "look at this cool thing we did", you have to invent some reason why it might be useful.

      The "we can supplement datasets for rare conditions using synthetic data" idea pops up quite a bit. It doesn't really make any sense.

    10. Re:GANs for data augmentation? by mikael · · Score: 1

      If you look at diffusion tensor MRI volume scans, you get a set of voxel cubes that have direction gradients in them. With all those bundles of nerve connections, things can go wrong - disconnections, tumours, abnormal blood vessels. You could get an artist and some whizzy volume cube editing software to cut and paste some abnormality from one image to another in a Photoshop way. But it's quicker to use a pair of GAN's to generate these images.

      https://www.alamy.com/stock-ph...

      --
      Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
    11. Re: GANs for data augmentation? by Anonymous Coward · · Score: 0

      As youâ(TM)ve pointed out, the Emperorâ(TM)s New Clothes nature of this research screams, âoehammer, make me a nail!â

    12. Re:GANs for data augmentation? by mhesd · · Score: 1

      but these images won't be excluded due to the lack of counterexamples in the training set.

      They will be excluded because the discriminator network of the GAN will learn to distinguish between them and real images.

    13. Re:GANs for data augmentation? by tempmpi · · Score: 1

      They will be excluded because the discriminator network of the GAN will learn to distinguish between them and real images.

      The discriminator can't do magic. It can only decide based on the information contained in the training images. If the set of training images is rather small, the performance of the discriminator won't be very good. The discriminator will either overfit the training set and generate only images very similar to the ones already contained in the training set or won't be very accurate and the GAN will generate images are not plausible.

      Potentially a GAN could be a way of extracting hard to describe information out of a human oncologist:
      Let the GAN generate lots of images and let the oncologist rate them. The plausible images could then be used to improve the GAN.

      --
      Jan
    14. Re:GANs for data augmentation? by Anonymous Coward · · Score: 0

      Yeah, I was about to say that this would get avoided like the plague.
      This is a stones-throw away from falsified data, which has knowingly killed people in the past.

      GANs aren't magical. They'll spit out exactly what you know already, and in worse ways than ever before!
      Neural structure isn't the only thing that defines brain tumors, waste build-up like plaques and such also have a huge role in formation too, as well as sheer access to nutrients at the time.
      You simply cannot model that properly with limited brain-scans.
      It won't work.
      This will create horrific not-brains akin to someone opening a scan in Photoshop, doing a freeform selection around a tumor and pasting it around using context-aware mode.
      It might be accurate to structure, might being the keyword, but it won't be accurate to biology.

      If your base dataset is too small, you simply cannot go anywhere significant.
      Evolution itself cannot do it. It leads to genetic dead-ends and extinction.
      We're simply doing this same process but billions of times faster.
      And that's even with our intelligence behind it.
      Once we start adding in our own biases on what we know, you start getting in to the area where you generate things you already know AGAIN. You're doubly creating the same data! ARGH!
      This is fruitless. It seems like such a waste.

      Why can't they put some of this effort in to virtual cell modelling instead?
      Give virtual brains cancer at the cellular level!
      Or just ask more people to make brain scans public. People are stupid. Just word it properly.
      "What's the worst that could happen anyway? Someone clones you in 60 years? You'll probably not eve-..."

  3. late to the party, and you can call me Al by Anonymous Coward · · Score: 0

    https://regmedia.co.uk/2015/07...

    And THIS is why.

  4. Boring by Anonymous Coward · · Score: 0

    Call me when they create synthetic fecal matter.

  5. Great for when I get a simulated tumor by Anonymous Coward · · Score: 0

    This is great news... think about all the children out there suffering from this pesky simulated brain tumours... now we can feed those brains into the AI and it will be able to identify them.

    I just wonder if testing/training with fake data,will that only allow you to identity fake tumours?

  6. What a Great Day For A.I. by Anonymous Coward · · Score: 0

    will someone think of all the children that are suffering with these pesky incurable simulated brain tumors.. now we can feed those brains into the AI and it will be able to tell us 100% if they have the simulated brain tumors... this is a great day for AI..

  7. Will just regurgitate what was already known by Crashmarik · · Score: 1

    You'll just have convergence to what was already known about the characteristics of any abnormality, with near zero possibility of providing new insight.

    Seems that the way to characterize this, is that the networks are being used to generate plausible fakes.

    1. Re:Will just regurgitate what was already known by religionofpeas · · Score: 3, Insightful

      I'm glad that for anything that scientist have thought of, there's always a slashdot expert who knows better.

    2. Re:Will just regurgitate what was already known by Antique+Geekmeister · · Score: 1

      Some of us actually know about neural networks and the dangers of doing meta-analysis on top of meta-analysis, without more original data.

    3. Re:Will just regurgitate what was already known by religionofpeas · · Score: 2

      I'm very glad that "knowing about neural networks" allows some of you to dismiss the results of senior research scientists who actually did the work. Have some of you actually read the paper ?

    4. Re:Will just regurgitate what was already known by Anonymous Coward · · Score: 0

      Now when my abnormal brain actually gets scanned, nobody will try to understand the results since I'll already be classified as "fake".

    5. Re:Will just regurgitate what was already known by ceoyoyo · · Score: 1

      Hi, I'm an actual scientist who does medical AI research, in actual hospitals. The OP is correct. The claim that you can usefully supplement datasets with synthetic data is either journalistic hyperbole, a throwaway "look this is practical and useful" line from a paper or a misunderstanding by some researchers who don't have much experience in the medical field.

    6. Re:Will just regurgitate what was already known by Crashmarik · · Score: 1

      I am always glad that my fellow humans never cease to astound me.

      You engage in magical thinking while worshiping science. Do you use an atomic as your religious symbol ?

      Anyway the essence of science is validation against reality, so tell me how do these networks link back to reality or expand the knowledge of the disease ?

    7. Re:Will just regurgitate what was already known by Antique+Geekmeister · · Score: 1

      Pardon me, but what results? The analyses using the forged MRI images has not yet been done as best I can read the original article. There are no results yet to analyze or dismiss.

  8. Link to original paper by Anonymous Coward · · Score: 0

    https://arxiv.org/abs/1807.10225

  9. On the nose by nospam007 · · Score: 3, Insightful

    Artificial Intelligence finds artificial brain damage.

    1. Re:On the nose by PolygamousRanchKid+ · · Score: 2

      Artificial Intelligence finds artificial brain damage.

      This is a synthetic brain, not an artificial brain, Jim.

      So the really interesting question, for the IgNobel, is . . . "Do synthetic brains dream of polyester sheep . . . ?"

      --
      Schroedinger's Brexit: The UK is both in and out of the EU at the same time!
    2. Re:On the nose by shubhamtaxila · · Score: 0

      when there's existense of something, there have to be problems . top 10 mba college in jaipur

    3. Re:On the nose by shubhamtaxila · · Score: 0

      when there's existense of something, there have to be problems . best bank po coaching in jaipur

    4. Re:On the nose by RespekMyAthorati · · Score: 1

      syn-thet-ic (adjective):
      - made by chemical synthesis, especially to imitate a natural product.

      So, no, it is artificial and not synthetic.

    5. Re:On the nose by Anonymous Coward · · Score: 0

      I wonder if it dreams of artificially brain damaged sheep.

  10. This seems kinda pointless by Anonymous Coward · · Score: 0

    There are some things that even AI cannot do.
    They might as well take their few examples and generate every rotation and translation to create a synthetic data set.
    I hope the pollsters don't catch wind of this technique.

  11. so what? by Anonymous Coward · · Score: 0

    Give a monkey a box of crayons and they can create an image. Might not be as accurate but no one will claim that this is either under risk of being sued.

  12. Comment removed by account_deleted · · Score: 0

    Comment removed based on user account deletion

  13. Similar to BDTs by Roger+W+Moore · · Score: 1

    This seems to be a similar, but more complex, approach to boosted decision trees where the training samples that the algorithm initially mis-classifies are fed back through with a higher weight to make the algorithm pay more attention.