China hospital to use AI to detect lung cancer

World
Europe

Doctors in China may get a new aid in diagnosis of lung cancer in the form of artificial intelligence (AI).

"In China, 70 percent of diagnoses are based on information from medical images. The number of medical images is increasing at 40 percent every year, but there is only three percent growth on radiologists." Liu Shiyuan, director of the radiology department of Shanghai Changzheng Hospital told ThePaper.cn.

The lack of radiologists may lead to the rise of error rate in reading images . Some early-stage symptoms could even be missed.

X-ray image of chest. /VCG Photo

Shanghai Changzheng Hospital turned to AI to solve this problem. An AI-trained machine has already been used to assist doctors to diagnose lesions and cancer on the lungs.

How does AI detect lung cancer?

There is a reason for choosing the lungs as the first testing area. Lung contains air, which provides the best condition for radiology. If there is a lesion on the organ, it could be discovered clearly like "a white dot on a black paper." Other organs such as brain and liver are considered as "grey paper" under the machine.

However, the machine can make mistakes, producing inaccurate "fake positive" results.

Human lung and bronchi. /VCG Photo

A machine can only learn what’s in an image if human marks useful information first. Liu cooperated with Infervision Co., an AI company that focuses on applying deep learning technology to assist medical image diagnosis to be accurate and efficient. Liu used data marked by himself to train and increase accuracy of the machine.

The recognition rate of lesions has been improved after six months of training and experimenting. Error among ten "suspects" is down to one or below. The machine is also able to catch nodules that could be missed by human eyes.

Two problems for practical use

"We are at an early-stage of solving medical problems with AI. Although many companies are dedicating in this area, not a lot of doctors and hospitals are willing to adopt this technology." Liu said.

Inputting data to the recognition system is also a big challenge.

"These images must be filtered one by one, and all the lesions must be marked out correctly. It takes at least 30 minutes for one experienced doctor to diagnose one patient, you can imagine how long it takes for doctors to mark 6,000," Liu said.

Medical image of lung. /VCG Photo

Although data-inputting is extremely time-consuming, a trained system can benefit the hospital significantly. By using machine as radiologist, more medical resources could be released, and doctors can spend more time on communicating with patients.

China has one of the highest lung cancer rates in the world with more than 700,000 new cases in 2015 alone. There are 80,000 radiologists in the country who diagnose around 1.4 billion radiology scans a year.