Mathematicians, software designers, epidemiologists and scientists are employing AI, machine learning, digital technology and cloud computing to detect infection, analyse transmission rates and distribution patterns of COVID-19, which is making so many millions of people around the world dangerously ill. The result, says Dr. Katarina Gospic, director of neuroscience at the VR/AR company Spinview Global: “Mathematics and artificial intelligence (AI) are two of the most effective weapons we have in the fight against the coronavirus outbreak."
Countering crisis with computer simulations
COVID-19 is a new virus. Whilst information and knowledge is accruing almost daily, very little is actually known about its means and rate of transmission, incubation period, fatality rate or what could happen after its peak. In the absence of such vital information, mathematicians are conducting computer simulations of this outbreak based on existing knowledge of past virus outbreaks such as SARS and Ebola.The simulation tools used by public health professionals to predict patterns of the virus' spread and whether border screening would be an effective means of detecting imported cases are similar to popular computer games like Cities Skylines or SimCity. For example, computer simulations based on airline passenger statistics and data on the numbers infected with COVID-19 outside of China provided an early estimate of the size of the outbreak in the city of Wuhan. A later simulation, led by UK epidemiologist Professor Neil Ferguson at Imperial College London, predicted the UK might face more than 500,000 deaths if the government took no action; estimates for the United States hovered around 2.2 million deaths. These theories formed the basis of government guidelines on social distancing in the UK and similar orders.
How interactive models work
A computer simulation is a set of equations and decision trees based on large quantities of data and various assumptions. The mathematics underpinning these interactive models is deceptively simple: Imagine there are four buckets. Bucket A contains all those deemed susceptible to the virus. Bucket B contains people who are not sick but might become sick. Bucket C contains all those who became ill and contagious. Lastly, bucket D contains people who have recovered, been isolated or have natural immunity. Mathematical modelling also requires information on the number of infections, hospital admission numbers, deaths and recoveries. Next come a series of assumptions. First, what is the probability of an infected person meeting a susceptible person? Second, what is the probability of transmission from the infected person to the susceptible person?
The answers to these questions are, at best, estimates. Added to that are inputs of the population's socio-demographic characteristics, such as age, gender, health status, employment, number of contacts and so on. Considered together with their weightings, this information can help form the basic ingredients of the computer simulations to provide “a sense of where you are, where you are going and how you are doing in the fight against the disease," explains Dr. Hannah Fry, associate professor in the University College London Department of Mathematics, in a BBC radio blog post. The results in such simulations are presented to government and public health officials, and used as an evidence base for social, economic, financial and health decisions designed to mitigate the impact of the virus. Dr. Fry calls this "bending the curve," suggesting that “math helps us to predict the future."
US employees use a fever-scanning thermal camera before entering the factory
Emerging technologies propel research forward
The mathematical tasks of processing such enormous amounts of data are increasingly being performed speedily and automatically by a combination of AI, machine learning and cloud computing. Public health bodies, universities and private sector companies such as BlueDot, Metabiota and Rockwell Automation are using health simulation forecasting software. For this particular outbreak, however, data inputs for the computer simulations are significantly weak. As John Edmunds, simulation modeller at The London School of Hygiene & Tropical Medicine, explains, “you can project forwards and then compare against what you get. But the problem is that our surveillance systems are pretty rubbish."
How AI works
Traditionally, a computer follows a set of instructions to complete a simple task in the same way. With artificial intelligence (AI) and machine learning, the computer software programme adjusts how it completes a task over time. In order to complete a task, the computer has to learn how to respond to certain actions, so it uses algorithms (a list of rules to follow in order to solve a problem) and historical data to create something called a propensity model. Propensity models will then start making predictions (similar to scoring leads). A simple example is a robot vacuum cleaner, in which the room's dimensions and furniture placement is contained in an algorithm, which guides the robot in its floor-cleaning task. A recent report from the World Health Organization concluded that AI and big data were key in China's response to COVID-19. A similar strategy is being employed in the west.
AI, machine learning, natural language processing (NLP) and semantic text analysis through machine vision have greatly enhanced traditional research methods of citation analysis and extraction of relevant figures, entities and venues from published research papers. The free, nonprofit Semantic Scholar project, based in Seattle, Washington, uses natural language processing to analyse scientific papers about coronavirus. In Canada, BlueDot uses NLP to skim the text of hundreds of thousands of sources for news and public statements about health of humans and animals. It sent an alert about the coronavirus to clients on December 31, 2019, several days before any official public health statements. Google subsidiary DeepMind is employing its AI-enabled AlphaFold simulation systems in the fight against COVID-19 as well. AlphaFold researchers are developing computational methods to predict a human cell's protein structure from the amino acid sequence in order to discover how to disrupt the COVID-19 virus from binding to human cells, as well as how different drugs can slow the virus's rate of reproduction. These genetic and therapeutic approaches to treating COVID -19 rely on artificial intelligence, software engineers and computing power.
Prevention and containment
In hospitals and elsewhere, AI-enabled mobile robots carry out a number of health and safety tasks. Danish robotics company UVD Robots is marketing an AI-enabled robot to disinfect a room with ultraviolet (UV) light. The company's website states that the use of its robots can prevent and reduce the spread of infectious diseases by breaking down their DNA-structure.
Screening populations using AI
Population screening for COVID-19 is a necessary but particularly challenging task. To screen America's adult population of 330 million at the rate of 100,000 tests a day, for example, would take at least seven years. More direct screening measures, such as those employed in China, raise fears of personal privacy in the west, but in a health emergency such as this one, may well be adopted. Health authorities in China are using AI-enabled drones to map areas, track objects and provide virtual feedback. Drones take images of a crowd in a particular area, AI-enabled facial recognition software is also being used to help identify individuals with a temperature who need to be isolated and given medical treatment. Furthermore, AI-enabled drones from Shenzhen MicroMultiCopter (MMC) are identifying people without mandated face masks and/or a fever in major cities, including Shanghai, Guangzhou and Foshan.
A German company, Dermalog, has developed an AI-enabled camera that uses facial recognition software to instantly calculate a person's temperature up to 1.5 meters away, with the capability to process multiple people simultaneously. This camera is also being integrated into the border control systems of many countries, including Thailand and Singapore. Between May 2019 and February 2020, it helped identify some 4,300 people on a national blacklist and approximately 127,000 travellers in breach of visa violations.
Chinese Police officers operate a drone during the COVID-19 outbreak
AI's role in PPE production
AI-enabled 3D-printing technologies are now helping to produce much of the personal protection equipment (PPE) and breathing devices that are in such short supply in the US and Europe. AI technology scans existing devices, and computer-aided design software then creates instructions for 3D printers to manufacture the requisite item. To counter the UK's well-publicised shortage of PPE, for example, 3D printers located in London's schools, are making personal protection equipment for health workers. In Brescia, Northern Italy, a hospital ran out of vital replacement valves for its “reanimation" machines used to ventilate COVID-19 patients. Manufacturing of these complex valves using traditional methods is a slow process. Instead, necessity made room for innovation and rapid production of valves by 3D printing.
Treating a patient's condition
There are many ways in which a combination of AI, machine learning and cloud computing are improving medical and treatment decisions. Changes in three vital signs—levels of the liver enzyme alanine aminotransferase (ALT), reported myalgia, and hemoglobin levels—are predictive of subsequent severe disease in COVID-19 patients. Combined with other relevant factors, these measures allow AI software to predict the risk of acute respiratory distress syndrome with up to 80% accuracy. This innovative approach dispenses with the need for a human expert to physically examine thousands of images a day. Instead, the combined knowledge of experts is embedded in smart computer software that can scan thousands of images for these relevant vital signs 24/7 and in record time. Outliers and unexpected results are subsequently brought to the attention of human experts. Ultrasound scans use high-frequency sound waves to display body organs, such as hearts, wombs, lungs, etc. A provider of AI-enabled ultrasound simulator training software, Intelligent Ultrasound Group, has updated its BodyWorks ultrasound simulator with a free COVID-19 training module. The update includes several examples of lung ultrasound appearances typical of COVID-19 infection to enable clinical staff to practice and train in using a lung ultrasound.Then there is Butterfly IQ, a handheld, AI-powered ultrasound device that sends images to a user's mobile phone for automated interpretation.
Here, AI comes into play in two ways. First, automated image analysis makes it easier for people with less training to use the scanner effectively. Second, as more medical professionals use Butterfly Network's scanner to check for symptoms of COVID-19, the company's cloud-based system gets more robust data about the virus and what it looks like. Because this outbreak has proven so rapid and widespread with more than 2.2 million deaths in at least 180 countries, governments, public health bodies and medical workers need virtual information on COVID-19. Understanding the transmission rate and incubation period in order to develop and adjust policies and treatments will be key to successfully overcome the biggest health epidemic since the Spanish Flu. A collaborative and multidisciplinary approach backed by AI, computing and digital technologies is making advances toward containing this human scourge.
The author: Nicholas Newman
Journalist who regularly writes about agriculture, aerospace, business, energy, engineering, rail, shipping, technology, transport for clients worldwide.
Read more on Covid-19
Selected content on this topic.