Articles, Blog

Case Study: TensorFlow in Medicine – Retinal Imaging (TensorFlow Dev Summit 2017)

February 12, 2020


15 Comments

  • Reply AJ Usog February 16, 2017 at 2:55 am

    Are the images using the same type/quality imaging machine? if not, you've introduce image quality as a noncontrolled variable which is the one control that needs to be tight (exact same tolerances for precise validation) with this development.

  • Reply Nikhil Verma February 16, 2017 at 5:58 am

    This is really great!

  • Reply Dinker Chaudhary February 16, 2017 at 8:58 am

    this girl is sweet

  • Reply Carlos Chau February 16, 2017 at 9:32 am

    09:23 Nikon is not a little company, it's a huge DSLR maker LOL (I'm only saying this for a laugh)
    I work on healthcare data and this is exciting news and inspiration to me

  • Reply Katsufumi Wang February 16, 2017 at 12:12 pm

    one of the biggest obstacles of any product commercialisation is economics, in addition to politics of industry. having overcome the technical aspects (which in itself is another major obstacle), I hope this gets out to better the lives of many. thanks

  • Reply Arun Das February 19, 2017 at 12:35 am

    Awesome work ! One of the main points that Lily Peng pointed out was about data and its validity in general. As far as I know, DRIVE, STARE and Kaggle datasets are the best ones out there for deep-learning enabled Diabetic Retinopathy research. I am also amazed by how versatile Inception architecture is. She also pointed out that the network is pre-trained on ImageNet and thus got better results. Impressive work by the team ! Kudos guys !

  • Reply Gregor Samsa February 19, 2017 at 3:16 pm

    I really like all related to vision or image science, but it is hard to get into bioinformatics (or computer vision related) field even with computer science degree. Any advice if someone here has the answer on question 'How other people do this?' is valuable for me, so please share your thought!

  • Reply bruce bagnoli May 18, 2017 at 9:33 pm

    Exciting work with the potential to prevent so much suffering. The integration of the hardware, ML, and clinical experience is powerful. Thanks for sharing!

  • Reply mayur newase August 11, 2017 at 6:05 am

    Is data available to public??

  • Reply Escape Felicity December 3, 2017 at 2:20 pm

    That idiotic rising tone makes me sick

  • Reply saurabh dasgupta March 5, 2018 at 11:31 am

    Excellent. Keep up the good work

  • Reply Ashutosh Kushwaha May 13, 2018 at 8:51 am

    The best statement " In the previous life I was a doctor and have been repurposed as Google Product manager" ๐Ÿ™‚

  • Reply Nikhil S July 23, 2018 at 7:19 pm

    If you are so concerned about diabetic retinopathy, make your dataset open-source.

  • Reply MB February 4, 2019 at 4:45 pm

    03:33 she says that for the two images all grades are classified. This is not correct. For the lower image there is no blue (no DR). Also no orange (severe) in the upper image?!

  • Reply Unboxing Science June 7, 2019 at 10:45 pm

    What kind of coding do you use in tensor flow to detect these DR

  • Leave a Reply