Coronavirus: AI tool developed to detect abnormal chest X-rays quickly

Rapid flagging of pneumonia in suspected Covid-19 cases allows for earlier intervention

Dr Ting Yong Han, consultant in diagnostic radiology (clinical) at TTSH, looking at a chest X-ray of a Covid-19 patient that was flagged by the AI tool, RadiLogic, which can analyse an image within three seconds.
Dr Ting Yong Han, consultant in diagnostic radiology (clinical) at TTSH, looking at a chest X-ray of a Covid-19 patient that was flagged by the AI tool, RadiLogic, which can analyse an image within three seconds.ST PHOTO: ONG WEE JIN

An artificial intelligence-powered tool has been developed to detect abnormal chest X-rays quickly during Covid-19 screening at the National Centre for Infectious Diseases (NCID).

With rapid flagging of pneumonia cases, doctors can intervene earlier as more severe findings on chest X-rays may mean an urgent need for supplemental oxygen therapy and ventilation for patients.

Suspected Covid-19 patients who have pneumonia must be admitted to the hospital.

Everyone who visits the NCID Screening Centre who is suspected of Covid-19 or who has respiratory symptoms must do a chest X-ray.

Previously, each X-ray image was reviewed by a radiologist within an hour. The radiologists had to analyse each X-ray result in sequence, which was time-consuming.

Now, the AI tool, called RadiLogic, can analyse each X-ray image within three seconds and highlight abnormal chest X-rays on a computer quickly, helping radiologists prioritise which images to review. This allows radiologists to report such cases in about 50 minutes, after each patient completes the procedure.

The tool has an accuracy of up to 96.1 per cent, and has been used in the screening centre since May.

The screening centre saw 200 to 300 X-ray images a day during the peak of the pandemic in April.

Currently, there are 100 images coming in each day.

"The AI tool is performing close to human capabilities and it helps us prioritise the workflow to enhance our efficiency and diagnostic confidence," said Adjunct Associate Professor Tan Cher Heng, senior consultant at Tan Tock Seng Hospital's (TTSH) department of diagnostic radiology. He is also part of the team that developed the tool.

RadiLogic is not the first AI system for chest X-rays, but the team believes that it is the first chest X-ray AI tool developed for Covid-19 in Singapore.

 
 

Programmed through deep learning, RadiLogic was fed with 1,000 anonymised abnormal chest X-rays from Covid-19 patients and 4,000 anonymised normal chest X-rays to train it to detect pneumonia accurately.

White patches or haziness over parts of the lungs in the X-ray image indicate lung infections and pneumonia. The bigger the patches, the more serious the infection.

The tool was developed by radiologists from TTSH and researchers from the Agency for Science, Technology and Research's Institute of High Performance Computing, and Institute for Infocomm Research.

The team is hoping to deploy the tool to Covid-19 screening sites such as community care facilities and border control checkpoints, where radiologists are not on-site, and other hospitals such as Singapore General Hospital.

The team is also working to enhance the tool's AI capabilities in a few ways.

One is to train the system to distinguish between Covid-19 pneumonia and other types of lung infection, while another one is to zero in on abnormal regions and indicate the probability of pneumonia.

 
A version of this article appeared in the print edition of The Straits Times on September 03, 2020, with the headline 'AI tool developed to detect abnormal chest X-rays quickly'. Print Edition | Subscribe