• Pediatric Bone Challenge: Boning Up • MedicalExpo e-Magazine
    The Smart Magazine About Medical Technology Innovations

    #32 - MEDICA & RSNA Special Issue

    Pediatric Bone Challenge: Boning Up

    /

    Develop an algorithm for determining bone age from pediatric hand radiographs (Courtesy of RSNA)

    The RSNA‘s most recent challenge highlights the potential of machine learning in radiology. The goal? To develop an algorithm for accurately determining bone age from pediatric hand radiographs.

     

    This year’s annual RSNA meeting focused on how artificial intelligence and machine learning (ML) can aid radiologists and other imaging professionals. In this context, the society’s high-profile Pediatric Bone Age Challenge ran from August to October under the auspices of the group’s Radiology Informatics Committee (RIC). Keenly contested, the challenge called on participants to develop an algorithm for accurately determining bone age using X-rays of children’s hands.

    The bone age of a child indicates the level of biological and skeletal maturity, and is typically used in the evaluation of endocrine and metabolic disorders.

    “The participating teams were judged by how well their algorithm-derived bone age evaluations matched the evaluations of expert human observers,” explained leading radiologist Dr. Adam Flanders, chairman of the RIC. “The results were both surprising and exciting.”

    Will Radiologists Follow the Dodo?

    The challenge’s top 20 algorithmic results surpassed the accuracy of all previous evaluations of this type.

    “We’re now talking accuracy to within one hundredth of one percent of the human evaluations.”

    “These algorithms are becoming more and more sophisticated,” said Flanders. “We’re now talking accuracy to within one hundredth of one percent of the human evaluations.” Such results make it clear that computer-aided radiological diagnosis will soon incorporate hugely complex and incredibly accurate ML algorithms. Some fear this could make radiologists obsolete.

    But Dr. Flanders believes such fears are unfounded. “This technology is not about replacing humans, but helping radiologists do their job more efficiently. ML-empowered devices could be hugely effective tools for precision medicine. By saving time and by allowing radiologists to focus on doing other things better, they could raise the bar for the entire discipline.”

    What You Can’t See…

    In fact, radiological algorithms simply compare image pixels. By doing so logically and mathematically, they can see things that are imperceptible to humans. “They can see trends, relationships that we might miss,” says Flanders.

    Algorithms may also help diagnose problems in parts of the world where there is no access to a radiologist. “Take the villages of sub-Saharan Africa,” says Flanders. “It is here that a reliable, portable, networked algorithm-based device could be used as a life-changing first step in diagnosis.”

    The challenge’s top 20 algorithmic results surpassed the accuracy of all previous evaluations of this type (Courtesy of RSNA)


    About the Author

    Daniel Allen is a writer and a photographer. His work has been featured in numerous publications, including CNN, BBC, The Sunday Times, The Guardian, National Geographic Traveller, Discovery Channel.

    Related Posts

    Medical images can help identify cases of intimate partner violence or sexual assault, according to...

    Smartphone addiction creates an imbalance in the brain that can result in a number of side effects,...

    French echotherapy specialist Theraclion presented its flagship Echopulse device at this year’s...

    Style Switcher

    Highlight Color:

                   

    Backgrounds:

                        

    You can also set your own colors or background from the Admin Panel.