UN, WHO & Mila Map the AI vs COVID-19 Battlefield

Despite attempts to contain COVID-19 spread, as of March 26 more than 530,000 people had been infected worldwide and the number of new cases continues to grow at an alarming rate.

AI tools have already joined the fight, guiding UAVs to automatically disinfect public areas, tracking disease spread vectors, diagnosing patients, etc. A new project from researchers with the UN Global Pulse Data Science Team, the World Health Organization and the Mila – Quebec AI Institute looks at current studies and programs that are using AI to tackle the COVID-19 crisis and suggests some promising future research directions. The team categorizes the AI applications in three areas; medical, which includes individual patient diagnosis and treatment; molecular, comprising drug discovery-related research; and societal.

From Diagnosis to Cure

Most clinical applications of AI during the COVID-19 pandemic response have been in medical imaging diagnosis, amid growing interest in using medical imaging for screening and diagnosis. “It has been found that COVID-19 has particular radiological signatures and image patterns which can be observed in CT scans,” the team points out. Identifying patterns is a time-consuming task even for experienced radiologists, and with patients flooding hospitals and overwhelming the medical system, a quick accurate diagnosis can be a lifesaver. Training data is key to boosting the speed and accuracy of machine learning CT lung scan based diagnosis. To achieve higher accuracy, some studies combined off-the-shelf software with customized ML approaches, while other studies employed human-in-the-loop methods using small manually-labelled batches of training data.

COVID-19 cases present a unique respiratory pattern, and some studies have exploited this to develop non-invasive prediction tools that can use for example Kinect depth camera streams to identify possible infections. Other research teams have trained Gated Recurrent Unit neural networks to classify abnormal respiratory patterns. Although the detection of an unusual respiratory pattern doesn’t necessarily lead to a COVID-19 diagnosis, it can be a useful non-invasive measurement when screening at large scale.

The biochemistry applications of AI meanwhile include predicting the structural proteins of the SARS-CoV-2 virus, improving viral DNA testing; explore existing drugs that could be re-purposed, and discovering new chemical compounds for treatment. DeepMind for example shared SARS-CoV-2 3D protein structure predictions generated using the company’s AlphaFold deep learning system.

Avoiding an “Infodemic”

As the number of cases continues rising, real-time short-term forecasting of infection rate is valuable information for both medical professionals and public policymakers. To cope with the unprecedented pandemic and evolving factors that can affect disease dynamics, information models need to evolve as well. ML techniques for epidemiological modelling tasks have provided essential insights and enabled better understanding of complex infection and spread factors.

With public information being regularly updated by local authorities and international organizations and mainstream and social media producing nonstop coverage, the COVID-19 pandemic risks becoming an “infodemic.” Leveraging AI models that can identify, track and analyze the spread of information can counter misinformation through strategies such as automated fact-checking and relevance analysis.

The authors underline the importance of open science and international cooperation at this time, and propose “the creation of extremely diverse, complementary teams and long-term partnerships” in the ongoing fight against COVID-19.

The paper Mapping the Landscape of Artificial Intelligence Applications Against COVID-19 is on arXiv.

---

Journalist: Fangyu Cai | Editor: Michael Sarazen