Machine Learning and the Novel Coronavirus — Lessons from China
In the beginning of 2020 news started to come out of China of a new Virus that would lead to coughing and flu-like symptoms. The virus, first known as novel coronavirus and later dubbed as SARS-CoV-2 (Severe Acute Respiratory Syndrome Coronavirus 2) wreaked havoc on Chinese cities and continues to disrupt life in countries around the world.
After initially reacting reluctantly to the virus outbreak, China moved on to react swift and with force. The city of Wuhan was put on lock-down on January 23 and later the whole province of Hubei, which encompasses about 11 million people.
While doctors on the front lines were fighting night and day to keep patients alive, big Chinese companies were scrambling to help in other ways.
Saving Lives with Machine Learning
With limited test kits it, diagnosing the disease is one of the biggest challenge facing health providers right now. Test kits are in very limited supply right now ”Even if the patient were around and exposed to someone coughing, sick, sneezing, I cannot give them a coronavirus test,” the doctor said. “We are being crippled by our department of public health and the CDC on our ability to combat this pandemic.”, according to CNN. The current test kits are not only limited in availability, but also have inherent problems. Especially the test kits used in the US, are being reported as problematic, “numerous states reported trouble verifying the tests because one component of the kit was flawed.”
China turned to AI for this issue. A number of solutions have popped up to rapidly diagnose Covid-19 from computer tomography (CT) imaging. Ping An Smart Healthcare is the technology arm of one of the larges insurances in the world Ping An. Their “Corona Virus Disease 2019 (COVID-19) smart image-reading system” was launched on February 19 and is already in use by 1,500 institutions. Ping An reported an accuracy of “above 90%” and delivers results in 15 seconds, where a radiologist needs about 15 minutes to examine an image. Also many facilities may lack radiologists, especially with an exhausted and overloaded health system.
Beijing based inferVISION retrained their InferRead software from recognizing pneumonia to recognizing the novel coronavirus. Used in 34 hospitals in China on pictures of over 32,000 patients it is being reported to have a sensitivity of 98% and an accuracy of 82%. This means 98% of actually positive cases are being identified by the software as positive and 82% of the results are accurate.
Beside the private sector Chinese scientists have made research available that can be used to detect Covid cases using machine learning. The results are promising
For 27 prospective patients, the model achieved a comparable performance to that of expert radiologist. With the assistance of the model, the reading time of radiologists was greatly decreased by 65%.
The scientists used a “ResNet with location-attention mechanism model.” to analyze the data and made the code and the data available for use by others.
Developers and scientists outside China have already started using this data. For example in this in this github repository that contains a plethora of links and info about detecting the virus in CT scans.
This usage of AI could potentially save lives by detecting patients earlier and with better accuracy, especially combined with a traditional test kit. For US and European customers there will likely be an unease of handing over medical data to Chinese companies so it can only be applauded that research is available openly that can be used to improve detection rates in other countries and reduce loads on an exhausted health system.
Keeping Businesses afloat
The impact of the novel Coronavirus on business has been severe. After initially tepid responses the Chinese government came down hard on businesses. Only the most esssential business would be allowed to stay open. People were holed up inside and small businesses suffered. The Purchase Manager Index (PMI) has almost halved to 26.5 points dropping below 50 points for the first time in 15 years.
“Stagnating consumption amid the coronavirus epidemic has had a great impact on the service sector,”
Zhengsheng Zhong, Director of Macroeconomic Analysis at CEBM Group
Especially small businesses that are financially sound and would be making a profit in normal times are now facing a liquidity problem. While fixed costs, like rent, are still piling up there is no money coming in. Banks are also reluctant to give out loans, as they are fearing for their own liquidity and the fear of a new credit crunch is high. So far China has refrained from a massive stimulus program like in the one in 2008 that was almost $600 billion in size.
With limited options for loans, a wave of bankruptcies is looming. MYBank and WeBank are online banks that are integrated in their respecvive ecosystem Alibaba and WeChat respectively. They take their data from online e-commerce data in the case of MYBank and social media data in the case of WeBank. But what they have in common is the speed with which they can grant loans to SME’s MyBank pioneered the “310” model. 3 minutes to fill out the application, 1 second for the approval decision and 0 human intervention.
Furthermore Alipay just introduced “Like for Loans” where customers can give a like to extend the Small Businesses credit line by a small amount (100 yuan or about $15).
When public life shuts down it’s important to supply business with loans to keep them afloat while the crisis is going on, so there are businesses left when the crisis levels out. Also this needs to be done quickly and without friction to allow the economical engine to restart.
Identifying the Contagion Risk
Alipay Health Code is a color coded system that uses big data and machine learning to determine if a persons poses a contagion risk. Created by the same company that is giving cheap loans to suffering small business using machine learning. The traffic light coded system is based on travel history and medical conditions. “A green code enables its holder to move about unrestricted. Someone with a yellow code may be asked to stay at home for seven days. Red means a two-week quarantine.” according to the Guardian. The system has already been deployed in 200 cities in China. The system has faced criticism, however. Not only is the process opaque and “Neither the company nor Chinese officials have explained in detail how the system classifies people.” But also has an analysis by the New York Times indicated that the app is sharing data with the local police.
The Times’s analysis found that as soon as a user grants the software access to personal data, a piece of the program labeled “reportInfoAndLocationToPolice” sends the person’s location, city name and an identifying code number to a server.
In Coronavirus Fight, China Gives Citizens a Color Code, With Red Flags, New York Times
For Western democracies this would lead to an outcry and not only damage anyone who implemented that kind of surveillance but also the big data and machine learning disciplines itself. For the novel Corona virus, lessons must be learned from China, but not everything can and should be copied to avoid a backlash and harsher regulation.
