As they battle against the novel COVID-19 virus, experts around the world are focusing on diagnostics, prevention, infection containment measures and contingency plans. But they have yet to find a fast, effective way to halt the spread of the pathogen because they have not yet developed or discovered any remedies or therapies that directly counter the respiratory disease. Identifying coronavirus symptoms at an early stage so immediate action can be taken could be a decisive turning point.
In an earlier blog post entitled “Über die Herkunft des Begriffes Artificial Intelligence” (= On the origins of the term Artificial Intelligence), we explored the current capabilities of artificial intelligence. By collating vast quantities of data, the technology is able to extrapolate missing information. Thus you can generate the information you need based on information you already have.
In China, the Alibaba Group, one of the world’s largest online retailers, used precisely this approach as a starting point – and has now presented an innovation in the early screening of the coronavirus. On February 5, the Chinese National Health Commission expanded the criteria for diagnosing new cases of infection by declaring computer tomography (CT) scan results to be an effective method of detecting the virus, alongside the existing nucleic acid test method. Following this announcement, Alibaba made use of exactly this information. The company reported that one of its research institutes, Damo Academy, has developed an algorithm which, by using CT scans, can tell whether or not the person being examined is infected with the coronavirus – with 96 percent accuracy. The online giant is relying on the sheer versatility of artificial intelligence to produce such applications.
According to Asian Review, the new diagnostic tool was first introduced at the Qiboshan Hospital in Zhengzhou, Henan province. According to Alibaba, more than 100 hospitals in Hubei, Guangdong and Anhui provinces are now using the AI technology for COVID-19 diagnoses. Given that the recognition process takes just 20 seconds – whereas a human doctor needs between five and 15 minutes to make a diagnosis – the new algorithm could enormously reduce the pressure on hospitals. Using artificial intelligence in this way is a textbook example of machine learning or “deep learning”. The system is trained by inputting an enormous number of CT images of the lungs of patients who are already known to be infected with COVID-19. The more images the algorithm can compare, the more easily it can recognise what an infected lung looks like on a CT scan. It can learn to distinguish symptoms with a degree of precision well beyond the ability of most human diagnosticians.