During the last couple of months, the COVID-19 outbreak more known as “coronavirus” or “Wuhan” was the most discussed topic online and on the empty streets. We had different opinions about it from conspiracy theories to comparisons with regular influenza, but today we are in front of a fact that COVID-19 is a pandemic and somehow we need to protect from it, stop its spreading and, finally, fight it.
As a person, who works in the AI field, in this blog, I’d like to share some directions where I and my colleagues can work to help public and private sectors all over the world. I hope that entrepreneurs and members of the government will see the opportunities not only for the temporary getting rich schemes or control of the citizens but as important parts of a strategical roadmap to improving public health and security.
Public Health Surveillance
The main danger of coronavirus is its spreading speed. That’s why governments put such measures as quarantines of the whole country because they can’t control the local outbreaks efficiently. One of the most straightforward steps towards the detection of sick people in the analysis of CCTV cameras that are already around us and finding people who show severe symptoms of the diseases and isolate them, who they contacted with and disinfect the corresponding surfaces. At Neurons Lab we have developed a demo solution, which you can see on the video above. The dataset to start research by yourself is available here.
Remote biosignal measurement
Some of the symptoms as temperature or pulse are very important to ignore them and rely only on the visual appearance that might be confusing. But of course, we can’t stop any person to measure their pulse, blood pressure or temperature. Maybe we could do it remotely somehow? Actually, there are some developments in computer vision that can estimate pulse and blood pressure based on facial skin analysis:
AI discovers the heartbeat in your face
A medium-depth dive into the technology that reads vital signs from video
Researchers find a way to measure blood pressure with a selfie video
A selfie video might be all you need to find out your blood pressure, per a study by researchers in Canada and China…
Concerning the body temperature, there are some developments as well, but in this case, for practical use, we would like to look into the direction of thermal imaging, not the regular cameras. At least for now.
Image processing based body temperature estimation using thermal video sequence — IEEE Conference…
Temperature is the most significant vital sign in human body regulation system if this increases at a certain level it may be…
IoT and wearables
Of course, measurements like pulse is much more comfortable and convenient to gather from the wearable devices like fitness trackers and smartwatches that almost everybody of us already has. But was has pulse to do with this? It can tell about physical, emotional state, maybe about arrhythmias, but influenza? Some researches state, that heart rate variability analysis and its deviations from the normal state can show early patterns of flu and, in our case, coronavirus.
How to Use HRV to Predict Illness
Nobody likes having to take time out from training after getting sick; especially during intensive training blocks or…
Chatbots and communication
Apart from “scanning” people in public, we might want to rely on their consciousness and self-assessment. If you can measure your temperature and pulse daily and record your coughs time-to-time, you also can add it to your diary. Then, either doctor remotely, or an algorithm can recommend you to stay home, take some other preventive action or request the visit of the doctor if the symptoms are too severe. One of such solutions you can check out here:
Chatbots screening for new coronavirus — and turning up the flu — STAT
AN FRANCISCO — As the new coronavirus spreads, health tech startups with medical chatbots are scrambling to update…
Social media and open data
Maybe you’re not the fan of having a health diary or sharing things online, but there are a lot of us that share it with the complete strangers in Instagram or Twitter. This data is helpful for more broad analytics of how far the disease went away. With the information about the users, we also can analyze the social networks graph and try to estimate what social groups are at risk of being contagious. Concerning the data, you all know where to get it and what public APIs are :)
How AI is tracking the coronavirus outbreak
With the coronavirus growing more deadly in China, artificial intelligence researchers are applying machine-learning…
Canadian startup BlueDot analyses much more than just social media data: for example, the worldwide movements of more than four billion travellers on commercial flights every year; human, animal and insect population data; climate data from satellites; and local information from journalists and healthcare workers, pouring through 100,000 online articles each day spanning 65 languages.
How Canadian AI start-up BlueDot spotted Coronavirus before anyone else had a clue
On December 30, 2019, BlueDot, a Toronto-based startup that uses a platform built around artificial intelligence…
Obviously, before coronavirus, we have encountered many other infections diseases and studied the rules of their life cycle. We have created mathematical models that can describe the evolution of the disease in the population and we can re-use them, again and again, to explain the novel outbreaks and even predict at which stage they are (aka “how bad everything is”). The good starting point for your own analysis of the coronavirus cases all over the world is here, alongside fitting the logistic curve in order to understand at which point exactly the situation is right now in Italy (alongside with a tutorial on getting the up-to-date data):
Covid-19 infection in Italy. Mathematical models and predictions
A comparison of logistic and exponential models applied to Covid-19 virus infection in Italy.
Also, it is important to model not only the virus evolution itself but also the countermeasures that are used. Instead of blindly closing the whole border, mathematical models can help us to understand how to allocate prevention resources more focused and optimal in the early stages of the outbreak. Check out mathematical optimization application in the paper below; on the image: comparison of no measures (A) vs other pure mathematical-based measures.
COVID-19 has opened to us another problem of today healthcare: it doesn’t scale when the number of patients grows quickly (actually tired doctors work only worse) and in general, the number of false-negative diagnoses is pretty high. Machine learning diagnostics doesn’t get tired and scales naturally just with the increase of the computational powers. We already see how deep learning-based lung image analysis works well in a current pandemic as well:
AI Enables Doctors to Diagnose COVID-19 Infection in Seconds
YITU “Coronavirus Chest CT Smart Evaluation System” can compress the diagnosis of suspected cases to 2–3 seconds.
Drug development research
Last but not least, apart from detecting and preventing the spreading of the diseases, we need to think about the creation of the vaccines on the scale. One of the important steps towards its creation is understanding the structure and nature of the virus we’re fighting against. DeepMind with their experience in protein folding analysis also made their step in predicting the protein structure of the virus and made it open-sourced.
Computational predictions of protein structures associated with COVID-19
The scientific community has galvanised in response to the recent COVID-19 outbreak, building on decades of basic…
The dataset on COVID protein structure and other flu for your own analysis you can find here:
GISAID - Global Initiative on Sharing All Influenza Data
More and more laboratories around the world are releasing through the GISAID Initiative the genome sequences of the…
Last but not least, AI can help in the analysis of already published academic works on viruses as SARS and current COVID. You can find a related dataset on Kaggle:
COVID-19 Open Research Dataset Challenge (CORD-19)
An AI challenge with AI2, CZI, MSR, Georgetown, NIH & The White House
Of course, most of the described approaches are very research-y and require major efforts to be really useful for most of the people. However, I find most of them indeed vital parts of the whole ecosystem for 360-degree defense:
- local outbreaks detection with video surveillance, remote biosignal measurement, and wearable data analysis;
- efficient treatment on a scale with the help of AI-based diagnostics of the samples;
- region-level analytics with social media analysis and chatbots;
- global development predictions with mathematical modeling based on the local and regional data updated regularly;
- meanwhile, drug development is boosted with AI-based simulations and molecule generations.
At Neurons Lab we are investing resources in several directions described above. You could see a demo of video surveillance analytics for symptoms detection and we’re working on mathematical models of infections spreading as well. We’re inviting for the collaboration interested parties, let us know!