Chinese high-tech companies have been ramping up efforts to develop artificial intelligence (AI) systems to speed up processing of CT scans of suspected novel coronavirus disease (COVID-19) cases.
National guidelines recommend CT scans as the key marker for diagnosing COVID-19.
Doctors can tell the disease from features like shadows over a patient's lungs. One patient can have about 300 CT images, which will take a doctor 5 to 15 minutes to analyze with the naked eye.
For COVID-19 patients, radiologists often need to check earlier scans, putting more pressure on diagnosis.
On Monday, AI firm iFlytek announced it had jointly developed an AI-based COVID-19 diagnosis platform with the Chinese Academy of Sciences.
According to iFlytek, the system can read and analyze a patient's CT scans within three seconds. Deployed at a hospital in Hefei, Anhui Province, the system so far has identified all confirmed cases and has a recall rate of 90 percent in lesion detection.
In machine learning, the recall rate, also known as the true positive rate, is used to measure the percentage of positives that are correctly identified.
It can analyze key image features such as the morphology, range and density of the lesion and present a dynamic 4D contrast of whole lung lesions, providing accurate and efficient reference for COVID-19 diagnosis.
Other AI companies are also developing diagnosis systems.
On Feb. 16, an AI system was introduced at a main COVID-19 hospital in Zhengzhou, Henan Province. Developed by Alibaba DAMO Academy and Alibaba Cloud, the system can analyze CT images within 20 seconds with an accuracy rate of 96 percent.
Yitu, an AI firm based in Shanghai, rolled out its scan-reading system on Jan. 28 at Shanghai Public Heath Clinical Center. After receiving feedback, the system has been deployed at about 20 hospitals across the country, including hospitals in Wuhan, epicenter of the outbreak.
The company said it can realize the automatic detection of the lesion area and the quantitative analysis can be completed in two to three seconds.
With the development of computer vision, reading and analyzing scans has evolved into a relatively mature AI technology.
Trained with scans and their clinical reports, some AI systems have outperformed radiologists at predicting lung cancer risks and spotting breast cancer in mammograms.
As for the system for COVID-19 diagnosis, the developers say with more data feedback, the systems have daily updates, and are getting more sensitive and accurate every day.