Slashdot Mirror


Chinese Startup Infervision Emerges From Stealth With An AI Tool For Diagnosing Lung Cancer (techcrunch.com)

Jonathan Shieber from TechCrunch writes of a Chinese company called Infervision that aims to help lower the number of people in China who die from lung cancer ever year. The company has created a tool that uses machine learning and computer vision to help diagnose cancers. From the report: The company is taking advantage of a digital infrastructure that's been in place in Chinese hospitals since the SARS outbreak in 2003. It is using training data from images stored in digital health records in China and coupling them with data the company's technology is collecting in real time from its deployment in 20 hospitals around China (including Peking Union Medical College Hospital and Shanghai Changzheng Hospital). Infervision is also working with GE Healthcare, Cisco and Nvidia to develop and refine its technology. The company has processed roughly 100,000 CT scans and 100,000 x-rays since its initial installation last year. Infervision installs its software on-premise at hospitals and updates its image recognition and diagnostics tools based on the data coming in from its training hospitals, Chen Kuan, founder and CEO of Infervision, said. Training procedures are divided into two separate components, according to Kuan. The first is the the actual training system, where annotated data is collected from radiologists and incorporated into the company's training data. Then an updated version of the software (including the latest training data) is distributed to the network of hospitals.

45 comments

  1. STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 1

    Stupid headline. Statistical models are not AI. Really, who is supposed to be impressed by this?

    1. Re:STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      Thank you. Bayesian statistics and cluster analysis methods are 100 years old. What is new is the sped at which they may be applied.

    2. Re:STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      And that's not at all what this is. Neural networks rely on summation and threshold functions across 1 or more hidden layers with any number of collecting nodes on each layer.

    3. Re:STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      Stupid headline. Statistical models are not AI. Really, who is supposed to be impressed by this?

      I know what you mean but... NN are statistical models.

    4. Re: STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      So are people

    5. Re:STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      Calling everything "AI" is the 2017 "3D printer". It will die too.

    6. Re:STATISTICAL MODELS ARE NOT AI by lorinc · · Score: 2

      I'm fed up with those recurrent comments that statistical learning is not AI. Please base your opinion on evidences and not on your feelings.

      For instance, go to the page of the major conference on AI: http://www.aaai.org/Conference...
      Look at the topics, they include machine learning. There you have it, the professionals in AI consider ML to be AI.

      We can now safely have that useless comment die and never appear again.

    7. Re:STATISTICAL MODELS ARE NOT AI by Anonymous Coward · · Score: 0

      People with undiagnosed lung cancer...thanks for playing asshole.

    8. Re:STATISTICAL MODELS ARE NOT AI by retchdog · · Score: 1

      that's just an advanced methodology for doing cluster analysis.

      --
      "They were pure niggers." – Noam Chomsky
    9. Re:STATISTICAL MODELS ARE NOT AI by Jzanu · · Score: 1

      No, that's not what it is. Learn the vocabulary of the items you are making claims about first, or you just show your idiocy. A Neural Network is a supervised (with outcome known and targeted) classification method using a specific mathematical approach, while cluster analysis is unsupervised and used for exploratory data analysis not final classification, and is done using a variety of distance measures.

    10. Re:STATISTICAL MODELS ARE NOT AI by retchdog · · Score: 1

      supervised learning is a special case of cluster analysis where you have two (or more) pre-labelled clusters for training purposes. supervised learning is an easier version of cluster analysis; so much easier, in fact, that some people mistakenly think it's fundamentally different.

      also, https://en.wikipedia.org/wiki/...

      if you only knew how much i know about what you're pretending to know, you'd be laughing as hard as i am right now.

      --
      "They were pure niggers." – Noam Chomsky
    11. Re:STATISTICAL MODELS ARE NOT AI by Jzanu · · Score: 1

      No, you're completely wrong. Seriously at least google some shit from an EDU website you are ignorant. Now you are mixing up cluster analysis and discriminant analysis.

    12. Re:STATISTICAL MODELS ARE NOT AI by Jzanu · · Score: 1

      Over decades I've taken more than 100 graduate hours in math, statistics, computer science, economics, operations research, and even some of the newer data science labeled courses. I know very clearly that you have no fucking idea what you are talking about. Here's a fucking hint: supervised learning applies to a LOT more than just neural networks; look up LDA/QDA and see if you can at least understand the equations.

    13. Re:STATISTICAL MODELS ARE NOT AI by retchdog · · Score: 1

      What the fuck did you just fucking say about me, you little bitch? I’ll have you know I graduated top of my class in MIT, and I’ve been involved in numerous secret modeling operations on Al-Quaeda, and I have over 300 confirmed predictions. I am trained in gorilla networks and I’m the top statistician in the entire US armed forces. You are nothing to me but just another target. I will wipe you the fuck out with precision the likes of which has never been seen before on this Earth, mark my fucking words. You think you can get away with saying that shit to me over the Internet? Think again, fucker. As we speak I am contacting my secret network of spies across the USA and your IP is being traced right now so you better prepare for the storm, maggot. The storm that wipes out the pathetic little thing you call your life. You’re fucking dead, kid. I can be anywhere, anytime, and I can kill you in over seven hundred ways, and that’s just with my bare hands. Not only am I extensively trained in data science, but I have access to the entire arsenal of the United States Marine Corps and I will use it to its full extent to wipe your miserable ass off the face of the continent, you little shit. If only you could have known what unholy retribution your little “clever” comment was about to bring down upon you, maybe you would have held your fucking tongue. But you couldn’t, you didn’t, and now you’re paying the price, you goddamn idiot. I will shit fury all over you and you will drown in it. You’re fucking dead, kiddo.

      --
      "They were pure niggers." – Noam Chomsky
    14. Re:STATISTICAL MODELS ARE NOT AI by Jzanu · · Score: 1

      God damn you are fucking kid, go to bed. Leave the militancy fantasy aside. I am a veteran of the Bundeswehr so that shit doesn't scare me in the slightest. Also, since you're going to fake military service at least don't try to imitate an officer, that's a crime in every country in the world.

    15. Re:STATISTICAL MODELS ARE NOT AI by ooloorie · · Score: 1

      I am a veteran of the Bundeswehr

      Calling yourself a "veteran" for playing soldier in the Bundeswehr: that's pretty funny!

    16. Re:STATISTICAL MODELS ARE NOT AI by Jzanu · · Score: 1

      God damn I knew you were stupid but this is a new height. Go fuck yourself you little weak coward! What have you done with your life you little faggot?

  2. Amazing! by Jzanu · · Score: 2, Insightful

    THIS is true innovation in applied technology! Building on their infrastructure developed for one epidemic, this firm has enhanced its service offering by applying new technologies in a way that treats an ongoing global problem and saves lives. These kinds of technological insights are why more legitimate scientific research papers were published by Chinese authors last year than the rest of the world combined.

    1. Re:Amazing! by Jzanu · · Score: 1

      As reference see here.

    2. Re:Amazing! by Anonymous Coward · · Score: 1

      Great, now that the propaganda is over, how about reducing the cause, errr ... like ... pollution?

    3. Re:Amazing! by toonces33 · · Score: 1

      That and getting people to quit smoking.

    4. Re:Amazing! by Zocalo · · Score: 2

      They're working on it. China is currently the world's largest producer of renewable energy, with more than double the capacity of the US in second place, but due to their huge energy consumption that was still just over 20% of their total production in 2013, increasing to 23% in 2014 due to their rate of deployment. Sure, there's a long way to go, but that's going to change fast as they've made a huge commitment to renewables and, unlike many other countries, are actually delivering on it; China dominates renewable deployments for multiple clean energy technologies over the last few years - although not without a fair share of controversy.

      --
      UNIX? They're not even circumcised! Savages!
    5. Re:Amazing! by Dunbal · · Score: 1

      Smoking is not the only cause of lung cancer, by the way. I don't know how many times I've heard patients say "but I never smoked" on receiving their diagnosis. While convincing people to give up smoking is important because it removes a preventable risk factor; early detection is the most important factor, as usual.

      --
      Seven puppies were harmed during the making of this post.
    6. Re:Amazing! by Anonymous Coward · · Score: 0

      But will the average Chan be able to see across the street in Peking?

    7. Re:Amazing! by Anonymous Coward · · Score: 0

      My next door (literally) neighbor died of lung cancer 4 months ago. She was not a smoker, had no risk factors (work in pollution, radon, etc) and lives in one of the best air quality cities in the US her whole life.

    8. Re:Amazing! by Anonymous Coward · · Score: 0

      Good for her I suppose

  3. deep neural nets are as close as currently possibl by DrYak · · Score: 4, Informative

    Statistical models are not AI.

    The article is poor on the details of the methods.
    But given the size of the data set, and given the kind of hardware used (massive GPU cluster),
    it's possible that the system relies on deep neural nets (probably of the convnet type, given that it processes visual information)
    and these things are as close as we can currently get to an actual AI.
    (i.e.: something that learns on its own, forms its own way to interpret the data, and vaguely works in the same way as visual cortex in animals and human)

    it's simplistic AI (it's closer to a collection to various visual cortex columns than a full-sized sentient brain. and even if there are some of the insects crawling in your basement with more brain power than this), but it's still AI.

    --
    "Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
  4. NOT going to happen in USA by Anonymous Coward · · Score: 1

    Privacy concerns already stopped such efforts in the USA, and are unlikely to be resolved any time soon. No real privacy, or privacy concerns in China.

    1. Re:NOT going to happen in USA by clodney · · Score: 1

      Privacy concerns already stopped such efforts in the USA, and are unlikely to be resolved any time soon. No real privacy, or privacy concerns in China.

      Actually, something very similar has already happened in the USA. Enlitic (http://www.enlitic.com/) is a company that uses neural networks and machine vision to do lung cancer detection in CT images, with accuracy rates that already compare favorably to human read images.

      Do any reading in the radiology space, and you see headlines about machine learning every day. It is definitely coming.

    2. Re:NOT going to happen in USA by Anonymous Coward · · Score: 0

      This is entirely incorrect. Why would you share such misinformation? The data being analyzed is entirely de-identified, do you have a portable CT or X-Ray scanner to run around, gather imaging data from people on the street & compare against this data to re-identify to a specific person? This isn't like face recognition for crying out loud.

      The 'training' data need NEVER be 're-identified' (e.g. no need even to have a 'unique ID number' attached to the data where that ID number maps to a specific patient within the hospital system) since its used only for analysis of new images. Heck, you could do the analysis on new images in the cloud itself with de-identified data attaching only a 'hospital specific identifier' & leave the mapping to a real patient to the on-site system and that's still not a real 'privacy concern'. A hacker would have to compromise the on-sight hospital system & at that point you have bigger issues then having access to de-identified imaging data. If you want to get really pedantic & 'secure' about it, once the data has been analyzed, confirmed to have been sent back (so there's a copy of the analyzed results at site) you can then just remove the hospital specific identifier from the images in the cloud...bang, there would be no way to re-identify that data short of having the original data AND the patient information about the patient it belongs to.

      You do nobody a service by spreading such misinformation. Patient privacy is a greater concern in 'the west' but those concerns can be easily mitigated for systems 'in the cloud', and 99% of people actually do not care (they'd rather get a good treatment than worry about if their data is exposed). The real issue isn't PHI data but the billing data that can be used to scam Medicare/Medicaid & other social health program payment systems.

      I have more than a bit of experience in this area.

  5. Fire and Brimstone falling down from the sky by Anonymous Coward · · Score: 0

    You'd want these guys on your side when the dogs and cats start living together

  6. They're in the right place for it... by MiniMike · · Score: 1

    a Chinese company called Infervision that aims to help lower the number of people in China who die from lung cancer ever year

    So they're trying to improve air quality and reduce smoking? Those seem like the best methods, and they're huge problems in China. Still any effort to detect cancer earlier will help people survive longer. Hopefully this works and can be expanded to other types of cancer soon.

  7. So which Company from the West Got Hacked? by Anonymous Coward · · Score: 0

    And trade secrets swiped? This is like an all-whites basketball team every happening in the pros. No way. No how.

  8. I can help too by Anonymous Coward · · Score: 0

    Stop smoking, stop spitting everywhere.

  9. Much needed by nospam007 · · Score: 1

    If you see how much the Chinese are smoking, this comes handy.

  10. Aim is to lower the number of people dying from LC by Ronin+Developer · · Score: 1

    Why not tackle the problem at it's source by cleaning the air and outlaw cigarettes? Anyplace where you have to wear a HEPA filter mask on a daily basis has bigger problems that need solving fast. I am hypothesizing that lung cancer deaths will fall significantly by fixing those two items.

    Still, it's a pretty interesting tool.

  11. Also are going to happen in Brazil by Anonymous Coward · · Score: 0

    Actually I think that something very similar are going to happen in Brazil so soon.

    Gabriel Lorenzi - Brazil
    My Website

  12. Re:Aim is to lower the number of people dying from by Anonymous Coward · · Score: 0

    I presume you think this is a 'zero-sum game' where you only do 1 thing at a time, that's a rather bizarre way to go through life.

    Besides which this research is applicable World-Wide, at the level of your 'innards' everyone is the same. Looking at a CT there's no way to tell what race/nationality someone is, the true 'equalizer'.

  13. Code of Chinese Startup Infervision already leaked by Anonymous Coward · · Score: 0

    if (patient->country == COUNTRY_CHINA)
        return true;

  14. Re:deep neural nets are as close as currently poss by Anonymous Coward · · Score: 0

    if (coughing_blood && is_smoker)

        printf( "You have lung cancer\n" );

    else

        printf( "You don't have lung cancer. Yet.\n" );

    printf( "Thank you for using the lung cancer detector AI.\n" );

    printf( "Ver 0.00a. All rights reserved 2017.\n" );

    It's an early version. but we expect to hold an IPO in 14 months. Pleas donate to our kickstarter project. kthanxbai

  15. Re:Aim is to lower the number of people dying from by Anonymous Coward · · Score: 0

    Let's outlaw alcohol while we're at it to reduce liver disease and cut down on DUIs.

  16. Re:Aim is to lower the number of people dying from by losfromla · · Score: 1

    I'm game for this.

    --
    Only I can judge you.
  17. I wonder if the smoking does it? by Anonymous Coward · · Score: 0

    One has to wonder if the fact that China smokes the 9th Highest amount of cigarettes per capita of any country in the world (2249.79 per adult per year) has anything to do with it.