Advances in chatbots, redesigned AI algorithms and distributed computing are helping to solve the pandemic as fast as possible
In this two-part series written by University of Montana College of Business graduate students in the Master of Science in Business Analytics program, the authors explore the role of data science in solving the coronavirus pandemic. You can find the second article here, which focuses on the challenges of clearly communicating data.
COVID-19 has altered our world immensely, casting a net of uncertainty that holds us in a state of eerie permanence. The coronavirus’ rapid spread has put a strain on hospitals, drug companies and resources, causing an onslaught of issues along with lags in care for those critically ill. All the while, health care professionals and public policy experts are working overtime to lay out a tangible escape plan.
As a soon-to-be data scientist, I was intrigued when my father, the chairman of the department of psychiatry at Tulane University and the chief of staff at one of Louisiana’s state hospitals, texted me: “Data scientists are saving the world right now, Sarah.” From my point of view, health care professionals like him were saving the world, not the data people. His text prompted me to explore how data scientists were also involved in solving the crisis.
By leveraging data science, hospitals and health care professionals are automating time-intensive tasks to meet the increased demands for patient care. In addition, a global community of researchers and data scientists are working to scan chemical compounds for a cure to the virus. Chatbots, artificial intelligence x-ray scans and crowd-sourced computing power are just some examples of how AI is being tailored to address the pandemic, freeing up medical professionals and researchers to attend to the most pressing cases.
As COVID-19 spread, hospitals and virus-specific hotlines experienced surges in the number of calls and questions they were receiving about the illness and related symptoms. According to a press release from IBM, wait times for calls to be answered could often exceed two hours, causing many people to hang up before receiving the answers they sought. The human system was not working fast enough to provide the answers to anxious callers.
In response, IBM leveraged its Watson Assistant for Citizens natural language processing and AI technology to create a chatbot that answers common COVID-19 questions. The chatbots use the Watson Discovery search functionality to take in information from the Center for Disease Control and Prevention and local sources. Not only can the program disseminate valuable information, it also enables staff to help with more urgent tasks. For example, the University of Arkansas for Medical Sciences, in collaboration with IBM, deployed a chatbot agent that reduced patient registration time by 50%, freeing up staff and speeding up the triage process. Providence Hospitals created a similar chatbot system in collaboration with Microsoft, launching the platform in the beginning of March at St. Joseph health system in Seattle. In its first week, the Providence chatbot served 40,000 people in the first week, according the Harvard Business Review.
AI X-Ray Scans
At Zhongnan Hospital of Wuhan University in Wuhan, China, the staff in the radiology department altered an AI algorithm used for identifying cancer from chest X-rays into one capable of identifying pneumonia, an illness related to COVID-19. The head of the radiology department, Haibo Xu, emailed Wired magazine about his successful implementation of the altered AI algorithm, saying, “The software helps overworked staff screen patients and prioritize those most likely to have COVID-19 for further examination and testing.” While detecting pneumonia is not perfectly correlated with COVID-19, it does allow health care workers to isolate those suspected of having contracted the virus.
When Infervision, the creator of the cancer-detecting AI system, realized the potential use for COVID-19, the company worked through the Chinese Lunar New Year to retrain the algorithm with COVID-19 X-rays. The algorithm was trained on 2,000 images and is being used in 34 hospitals around China, having scanned more than 32,000 chest X-rays so far. However, there was no mention of the accuracy of the AI system, which can be worrisome. Algorithms with a high false negative rate could lead doctors to missing critical patients that need to be seen. A high false positive rate could also lead to patients with less critical needs receiving more hands-on care, which does not serve those who need it most.
In the U.S. and Canada, the skepticism about the accuracy of these AI X-ray scans prompted data scientists to use crowd sourcing to improve the current algorithms. Covid-NET was created to detect COVID-19 cases in chest X-ray scans. Designed by Linda Wang and Alexander Wong at the University of Waterloo, it relies on a convolution neural network structure, known for their ability to detect patterns in images. The algorithm, along with 5,941 chest images, were made available through the Covid-NET platform on March 24. The design architecture and images were shared so that other data scientists can tweak the structure and hopefully improve the algorithm. MIT Technology Review warns that until the accuracy has been confirmed, the information should not be used.
Borrowing Compute Time
Finding a cure for COVID-19 involves scanning the troves of known chemical compounds to identify any that may be suitable to treat the quickly spreading disease. To do this, organizations such as IBM’s World Community Grid are asking lay people to lend their computer’s idle time to science. Those wanting to get involved need only download the app on a computer connected to the internet and go about their day. The app waits until the computer is either idle or being lightly used and creates simulations and scans on known chemical compounds. All findings are sent to Scripps Research, a non-profit biomedical research facility spearheading the initiative with the help of IBM. Similar initiatives have been used in other studies, including those for cancer, Ebola and AIDS. All data amassed by the World Community Grid is made available to the public and, according to an IBM press release, more than 770,000 people and 450 organizations have contributed almost 2 million years of computing power to support 30 research projects.
Folding@home, an initiative created at Stanford University, is also borrowing compute time from citizens to scan potential chemicals for a COVID-19 cure. Folding@home has mapped disease proteins connected to Alzheimer’s and cancer for the past 20 years, according to an article in NS Tech. Toward the end of February, Folding@home pivoted to performing similar simulations for COVID-19, prompting an influx of users. As of March 31, 600,000 citizens were donating idle compute time to the scientific research. The network is generating about three times more computing power than the world’s leading supercomputers. In order to accommodate the distributed computing style of the network, large calculations can be broken up into small ones that run on thousands of computers at once. By breaking up the large computations and drawing on citizens’ computing power, researchers are able to generate insights into COVID-19 at a scale that has never been seen before.
Chatbots, redesigned AI algorithms and distributed computing only begin to scratch the surface of the countless applications of data science unleashed in the current global pandemic. Because they can scale up in ways that human beings cannot, these AI systems can facilitate in managing otherwise overwhelming situations. As a result, health care providers can either take a much-needed break or turn their attention to critical patients. The synergy of data science and health care has created unprecedented advances in AI, enabling both data scientists and health care professionals to collaborate in saving the word.
Photo by Ibrahim Boran on Unsplash.