Deep learning helps Korean researchers analyse x-rays
January 4, 2018
South Korean researchers have developed a deep learning-based technology for analysing medical images such as x-rays that dramatically lowers the rate of misdiagnosis.
X-rays are a used to detect and diagnose various injuries and diseases, such as bone fractures, lung cancer and breast cancer. Compared with computed tomography (CT) and magnetic resonance imaging (MRI) scans, x-rays are simpler and relatively less expensive. However, since the images must still be interpreted by doctors, the technique still has one critical weakness: occasionally, doctors fail to notice the signs of disease.
Lunit, a member company of the K-ICT Born2Global Centre start-up body, is trying to change this using AI technology in joint research with major medical institutions in South Korea, including Seoul National University Hospital, Severance Hospital of Yonsei University Health System, Samsung Medical Center and Asan Medical Center.
"The data-driven imaging biomarker (DIB) technology that Lunit proposed for the first time ever in 2015 is an AI system that has learned abnormal and clinically significant image patterns from big data," said Anthony Paek, CEO of Lunit. "Currently, DIB technology has achieved an accuracy level comparable to that of human experts. In the future, however, we will have new DIB technologies capable of outperforming humans."
The DIB technology, called Lunit Insight, uses AI technology to analyse existing x-ray images to help doctors make more accurate diagnoses. So far, the results of clinical trials have shown that using this technology increases diagnostic accuracy by 14 percentage points. Based on deep learning, an AI skill-building practice that makes use of big data and artificial neural network technologies, it is capable of accurately analysing images.
"Most medical imaging AIs have been directly and indirectly influenced by DIB technology,” Paek explained. “Since we now have access to the big data of hospitals and are capable of processing DIB learning from various perspectives, the future potential of our company seems quite high."
With a medical AI research team consisting of 12 deep learning researchers and three medical doctors, Lunit has developed the deep learning and computer vision technologies. At the 2015 ImageNet Large Scale Visual Recognition Competition, it ranked fifth in terms of object detection and localisation, surpassing the teams of Google and Tencent.
The technology has gained recognition around the world through the company's participation in international competitions, ranking first in the Tumour Proliferation Assessment Challenge in 2016 and Camelyon in December 2017. Also in 2017, Lunit was the only Korean company included on CB Insights' AI 100, which ranks the 100 most promising AI companies in the world, and secured an investment from K Cube Ventures, SoftBank Ventures, Formation 8 and Mirae Asset Venture Investment.
Lunit unveiled Insight at last year’s Radiological Society of North America event in Chicago. Able to detect accurately the signs of major lung diseases, such as lung cancer, pneumonia, pneumothorax and tuberculosis, from x-ray images, it rapidly processes medical images to provide results in only seconds after the images are uploaded and boasts an accuracy of 98 per cent. The software is expected to be used for clinical purposes in hospitals and medical centres starting in the second half of 2018.
"Lunit Insight is the first AI-based medical image analysis programme in the world that has been released in a final, usable state,” said Paek. “Through this release, we plan to expand the basis for the industry so that more medical professionals can gain easy access to such image analysis software."
In the first half of this year, Lunit is expected to unveil a mammography system that allows doctors to use Insight for the early diagnosis of breast cancer.
K-ICT Born2Global Centre is a full-cycle service platform for global expansion. Since inception in 2013, it has been setting the standard for a successful start-up ecosystem as the main Korean government agency under the Ministry of Science & ICT.