Chinese scientists have developed an AI tool to read CAT scans and find Covid-19 cases missed by swab tests, potentially leading to earlier detection of the disease.
The team, based in the eastern province of Zhejiang and led by prominent infectious disease expert Li Lanjuan, said the tool could determine from a CAT scan with more than 86% accuracy whether a patient had Covid-19, influenza A viral pneumonia or neither condition.
The researchers said this was important because in the early stages of Covid-19, patients might have signs of the disease in their lungs but test negative to nose and throat swabs.
“These patients are not diagnosed as suspected or confirmed cases. Thus, they are not isolated or treated in a timely manner, making them potential sources of infection,” the team said in a peer-reviewed paper released on the website of academic publisher Elsevier late last month.
Li was among the specialists sent to Wuhan in central China with Zhong Nanshan, the country’s top respiratory expert, in January and played a key role in the government’s decision to lock down the city. She spent months in Wuhan, heading up a team treating patients during the city’s epidemic.
The researchers said they used 528 CT sample scans as the basis for the tool. In all, 189 samples of the scans were from patients with Covid-19, 194 had influenza A viral pneumonia and 145 samples were from healthy people.
It said the tool had an overall accuracy rate of 86.7% in classifying patients into the three categories.
While speeding up CAT scan diagnosis can help detect cases that may be missed by laboratory tests, confirmation of Covid-19 still requires a positive laboratory test, according to World Health Organisation and Chinese government guidelines. Cases identified through CAT scans are treated as probable cases.
A number of other projects are under way to try to speed up diagnosis using AI assessment of CAT scans.
The operator of the Chinese supercomputer the Tianhe-1 has offered researchers around the world free access to its AI technology for diagnosis to complement swab tests.
In March, the National Supercomputing Centre of Tianjin, which hosts the machine, said a diagnosis can be done within 10 seconds with 80% accuracy. It said the accuracy would rise as it added scans to the database.
And in Wuhan in February in the depths of the epidemic, the radiology department of the city’s Zhongnan hospital retooled software used to detect lung cancer to help with Covid-19 diagnosis, according to technology magazine Wired.
Some scientists have also started working on AI algorithms to integrate information derived from antibody tests, CAT scans and clinical history hoping to improve diagnosis accuracy.
In May, the European Commission said it would invest in a project to automate Covid-19 diagnosis by using AI screening of CAT scans. – South China Morning Post