Stanford's deep-learning algorithm outperforms radiologists in diagnosis of pneumonia

Source: Xinhua| 2017-11-16 16:17:21|Editor: pengying
Video PlayerClose

SAN FRANCISCO, Nov. 15 (Xinhua) -- Researchers of the U.S. Stanford University have developed a deep-learning algorithm that is more competent than expert radiologists working alone in diagnosing pneumonia, said a study released Wednesday by the university in a press release.

The latest state-of-the-art method involves an extraordinary algorithm called CheXNet, which is a 121-layer convolutional neural network trained on ChestX-ray14, currently the largest publicly available chest X-ray dataset that contains over 100,000 frontal-view X-ray images with 14 diseases.

The algorithm, developed by the Stanford Machine Learning Group, is capable of diagnosing 14 kinds of diseases, but it performs best on pneumonia.

"Interpreting X-ray images to diagnose pathologies like pneumonia is very challenging," said Pranav Rajpurkar, a member of the machine learning group and co-lead author of the research, but he explained that the algorithm, based on deep learning technology, can learn from hundreds of thousands of chest X-ray diagnoses and make accurate diagnoses.

In cooperation with Matthew Lungren, an assistant professor of pediatric radiology of Stanford, the researchers had four Stanford radiologists independently annotate 420 of the images for possible indications of pneumonia.

Meanwhile, they worked on developing an algorithm that could automatically diagnose the pathologies. One week later, the algorithm was able to diagnose 10 of the pathologies labeled in the X-rays more accurately than the results obtained via the most sophisticated means.

Within one month's time, the algorithm was improved to detect all 14 identified diseases and did better than the four Stanford radiologists in the diagnosis of pneumonia.

Normally, the treatment of chest diseases like pneumonia depends largely on doctors' ability to read and interpret radiological imaging, but even the most experienced radiologists may make misdiagnoses, said the press release.

The deep learning-powered algorithm was coded to overcome limits of human perception and avoid errors, it added.

Pneumonia is a common disease that forces about 1 million Americans to be treated in hospitals every year, statistics of the country's Centers for Disease Control and Prevention show.

TOP STORIES
EDITOR’S CHOICE
MOST VIEWED
EXPLORE XINHUANET
010020070750000000000000011100001367573021