• [contact-form-7 id=”68688″ title=”Contact form 1″]
    January 1, 2016

    Ask doctors and they’ll tell you: In children, there’s chronological age and there’s bone age. When the two don’t match, there’s a problem. Bones that mature too quickly or too slowly may impair a child’s growth.

    Radiologists measure bone age, or skeletal maturity, by comparing x-rays of children’s hands to the standard for their age. Aspects of this technique haven’t changed in more than 75 years. GPU-accelerated deep learning is poised to change that.

    Researchers at Massachusetts General Hospital’s new Clinical Data Science Center are testing an automated bone-age analyzer they’ve created. It speeds diagnosis of children’s growth problems and is nearly as accurate as human radiologists.

    What Bone Age Says About Health

    Set to begin clinical trials soon, the automated analyzer would assist radiologists rather than replace them, said Synho Do, an assistant professor at Harvard Medical School who leads the research team.

    “Growth can be a general metric for a child’s health,” said Mark Michalski, director of Mass General’s data science center. (NVIDIA is a founding partner of the center.)

    When a child’s growth is delayed or accelerated, the bone age test can tell doctors whether the cause is simply a family pattern or an indication of a chronic disease, endocrine disorder or genetic problem.

    Bone Age Measurement Grows Up

    But calculating bone age is a cumbersome, time-consuming process that requires radiologists to match digitized x-rays with images in a textbook published in the 1950s.

    “You look at beautiful digital images with modern technology and you flip through this old book trying to find the page with the right image,” Michalski said.

    Radiologists calculating children's bone age must use a textbook published 75 years ago to compare with digitized x-rays of a child's hand.
    Calculating children’s bone age now requires radiologists to match digitized x-rays with images in a 75-year-old textbook. Image courtesy of Massachusetts General Hospital.

    For Do, the problem was ripe for deep learning. He and the research team trained the neural network on about 7,400 x-rays and radiologist reports from Mass General’s records. Do said he was able to reduce training time by using the cuDNN version of the Caffe deep learning framework and the NVIDIA DIGITS DevBox deep learning appliance, equipped with four TITAN X GPUs and DIGITS deep learning training software.

    “Without GPUs, I wouldn’t have been able to get the performance I needed, and I wouldn’t have been able to develop such an accurate algorithm,” said Do, who is assistant medical director for advanced health technology, research and development for the Massachusetts General Hospital physicians’ group.

    Do expects even better performance in the future. In early September, the Mass General center became one of the first research institutions to receive our powerful DGX-1 deep learning system, the world’s first AI supercomputer in a box.

    Faster, More Accurate Diagnosis

    If the hospital can put Do’s automated bone age analyzer into practice, parents could see test results sooner.

    “Parents often have to wait a day or more for a test result,” Do said. “This algorithm can give results in seconds.”

    Deep learning could also boost accuracy by eliminating differences among radiologists with different training and experience, Do said. His bone age AI would process x-rays and provide suggested matches. The radiologist would choose from among these, and the system would generate a report on the results. The algorithm is 99 percent accurate, with an average error rate of plus or minus one year of age.

    “This is a small step that takes us toward a more effective healthcare practice for all patients,” Michalski said.

    Do and his research team presented an abstract describing their work on kids’ growth problems at the Conference on Machine Intelligence in Medical Imaging in Alexandria, VA, and plan to submit a paper soon.


    Comments are closed

    Sorry, but you cannot leave a comment for this post.


    Latest Posts

    Latest Video



    In less than one hour, WFNS Neuroanatomy Series Wednesday 12 pm GMT, 8 am EST

    WFNS Neuroanatomy Series Wednesday 12 pm GMT, 8 am EST TO REGISTER https://us02web.zoom.us/webinar/register/WN_tCjkjEphTrayRzDAfo5FbQ YOU TUBE LINK: https://www.youtube.com/channel/UCvMRufu7rEnrIFlbWHScE5A

    Join Mailing List


    “Neuroanatomy through Clinical Cases” , Second Edition by Hal Blumenfeld, Yale University

    SOURCE Printer Friendly Hal Blumenfeld, Yale University School of Medicine The book can be ordered through the Sinauer Associates website. Sample content is also available on that page…. Read More →

    Request a Webcast


    Request Panel Spot

    desertedbeach@hotmail.com please tell us the date, time, studio of Neuroanatomist P P


    MAILING LIST desertedbeach@hotmail.com