The number of COVID-19 cases continues to grow at an alarming rate – with over 487,000 people being infected worldwide as of March 26, 2020 – and predictions paint a gloomy picture of the next weeks and potentially months to come. Members of our data science team joined forces with researchers from the World Health Organization (WHO) and the MILA- Quebec AI Institute to map the landscape of artificial intelligence (AI) applications that are being built to tackle the COVID-19 pandemic.
With the continued growth in the number of cases of the novel coronavirus, researchers worldwide are working around the clock to better understand, mitigate, and suppress its spread. Our paper compiles the mounting studies which have been published about the potential of AI applications to help manage global response. Specifically, we focused on three main areas: individual patient diagnosis and treatment, protein and drug discovery related research, and the socio-economic impact of the disease.
How AI can inform medical research against COVID-19
When it comes to medical imaging, an AI model may perform certain tasks, such as reading CT lung scans, faster and, given the right data to train on, even more accurately than a medical professional. With the current pandemic, quick diagnostics using machine learning (ML) approaches could save lives. In several promising studies, AI models were trained to identify potential COVID-19 cases; others are combining off-the-shelf software with bespoke machine learning approaches; others are using a human-in-the-loop approach to reduce the time required to label the disease. All of these efforts are in an incipient phase, however, the preliminary results are certainly encouraging.
In addition, there is ongoing clinical research to discover drugs to combat the disease. Scientists are working to identify existing drugs that may be repurposed to treat COVID-19. One such example is the case of the much debated, and heavily publicised, Chloroquine and Hydroxychloroquine – two drugs typically used to treat malaria that have shown some promising results. Alongside this, there are ongoing efforts to discover new drugs that can counter the disease.
As time is of the essence, AI systems, methods, and models can act as a compact form of knowledge sharing that can be used to train other specialists and can be deployed widely. In order to facilitate the sharing of such data, clinical protocols and data sharing mechanisms will need to be designed and data governance frameworks must be put in place.
How AI can shed light on societal implications of the pandemic
The strain that the COVID-19 pandemic has put on our society is being felt at every level: from closed businesses, to economic distress, to schools grappling with online classes, to people isolated at home. Advice at the national and local level is changing daily as new information and model forecasts become available. Given the rapid progression of infections, real-time short-term forecasting can be a vital source of information both for medical professionals, and public policy decision makers. In particular, models must be flexible in order to adapt to changing protocols and procedures. With a wide range of possible factors impacting on the dynamics of the disease, AI models could prove a vital resource for epidemiologists in approximating the underlying complex behaviour.
In addition, being able to quantify the spread of information surrounding the pandemic will help us curb the propagation of misinformation and inaccuracies, which are increasingly prevalent. Social media and online platforms have become key distribution channels for news surrounding the virus.
Although national and international organizations have used these platforms to constructively communicate with the public, we are also seeing an ‘infodemic’ that is overwhelming people with all sorts of details. In our mapping exercise, we highlight a number of efforts currently underway to curate specific news content related to the virus and perform both manual and automated fact-checking and relevance analysis.
International cooperation is a key ingredient
Ultimately, the success of the global effort to use AI techniques to address the COVID-19 pandemic hinges upon sufficient access to data. Machine Learning, and Deep Learning in particular, requires notoriously large amounts of data and computing power in order to develop and train new algorithms and neural network architectures.
Few of the systems we reviewed in this research are yet to have the operational maturity needed to combat the virus at this stage. However, they have moved the needle in the right direction and have much to teach us. In order to operationalize these efforts, we must work together to define a road map and a funnel for AI applications in order to understand how this technology can help today, and as the pandemic evolves. Even more, international cooperation based on multidisciplinary AI research and open science can help to prepare the regions of the world which have not yet experienced widespread outbreaks, and those where the majority of vulnerable populations live in.
It is our hope that this mapping exercise provides the community of practice with relevant information that they can use in their own research to turn the tide and get us closer to defeating this ‘invisible enemy.’
Download our paper “Mapping the landscape of artificial intelligence applications against COVID-19.”