Dr. Caroline Chibelushi, KTM for AI, presents her views on the capabilities of AI in helping us deter present and future threats due to Covid-19.

 

“So here’s a reality check: AI will not save us from the coronavirus, certainly not this time. But there’s every chance it will play a bigger role in future epidemics—if we make some big changes. Most won’t be easy. Some we won’t like. There are three main areas where AI could help: prediction, diagnosis, and treatment”

[MIT technology review]

Introduction

We are in an unprecedented time, a time when AI can speed up the process of finding answers to a lot of our questions. This article aims to briefly discuss various initiatives presented in different settings and in collaboration with the AI communities around the world. These communities are collectively aiming to find solutions to the problem that has touched every corner of the globe. In addition, this article presents the challenges faced by the AI communities and suggests the best way to go about them.

It started in Wuhan and landed in London before we could prepare ourselves. By week two of Covid-19 being in the UK, supermarket shelves were emptied, flights grounded, education systems transformed, railways barely operating, people were working from home and others even being paid not to work. So, what is the future for Covid-19 and how will it affect us? The answer to this is complicated at best, but a report published by Imperial College London predicts social distancing, self-isolation and a rolling lockdown could last until September 2021, alluding to the fact that we may continue with this way of life until a vaccine to treat Covid-19 is available. Unlike other countries in the world, the UK’s economy has also been uncertain since the 2016 Brexit vote. A combination of Brexit and Covid-19 challenges will massively affect the UK economy and the way people go about their daily activities, be it at home, work or even their social environments.

The Covid-19 threat is real and has found its way into all areas of our lives, ranging from the economy to social interaction and to the environment. These threats are enormous, may last for a long-time and some bigger, currently unknown challenges will emerge with time.

The role of technology from now onwards and for the foreseeable future is unquestionable. The only way the UK and the rest of the world will be able to cope and bring their economies back up to speed is by adopting different technologies – mainly those associated with or that embed Robotics and Artificial Intelligence (RAI) in their applications.

Already, every thread of the economy, society and environment in all parts of the world has changed. We are communicating more often and becoming more creative with technology than ever before. AI is currently being used to analyse large amounts of data related to this pandemic from all parts of the globe. However, until we get the drugs and vaccines to treat Covid-19, life will never go back to normal as we know it. Treatment is the world’s highest priority and the speed of achieving this will require contributions from various technologies, particularly AI.

 

The following are some of the initiatives currently being developed:

 

Prediction

Natural-language processing algorithms are used by BlueDot and Metabiota to monitor news outlets and official health-care reports in different languages around the world, flagging high-priority diseases, such as coronavirus, or more endemic ones, such as HIV or tuberculosis. Their predictive tools investigate air-travel data to assess the risk that transit hubs might see by infected people either arriving or departing.

There are a number of AI powered initiatives generated by Johns Hopkins such as the Covid-19 tracking map [1] based on its datasets [2], and also Susceptible-Exposed-Infectious-Recovered (SEIR) predictive modelling. CoronaTracker Community Research Group have developed an AI tool to predict and forecast Covid-19 cases, deaths, and recoveries through predictive modelling. Guy’s and St. Thomas’ NHS hospitals in London have released an AI Covid-19 symptoms tracker. Its aim is to help these hospitals advancing their research in Covid-19.

 

Diagnosis and monitoring

There are contrasting reports about the use of AI tools for diagnosis and monitoring. For example, Disease Control and Prevention (CDC) and the American College of Radiology do not recommend the use of X-rays and CT scans to diagnose Covid-19. The UK Royal College of Radiologists, however, says that “It is important to stress that these imaging appearances are generally non-specific and overlap with the appearances of other viral chest infections including influenza, MERS and SARS. The CT appearances alone will not obviate the need for viral testing and should not be viewed as equivalent to or replacing this.”

There are already hospitals which are deploying AI tools to detect Covid-19 on chest scans (deep learning algorithms are used to diagnose, triage, and monitor coronavirus cases from lung images).

The ambition is to further this technology to help predict which patients are most likely to need a ventilator or medication and who can be sent home. There are many AI tools (e.g. DarwinAI) which are emerging that use chest CT and X-ray scans to diagnose and monitor patients. They offer potential benefits in alleviating the growing number of patient scans for review by radiologists each day.

However, because this is such a novel disease, these initiatives are very limited, and developers do not have enough images to train existing AI tools. “The DarwinAI researchers say that the scarcity of Covid-19 radiography images in the public domain was the main motivation for their open-sourcing the COVID-Net project – as training data is key to boosting the speed and accuracy of AI-powered CT lung scan-based diagnosis” [MIT Technology Review].

Compared to a swab for a nucleic acid test which can be taken in a car or any isolated location, a chest scan requires an enclosed space with staff nearby. Other challenges include patients’ exposure to radiation doses and increased exposure to staff and facilities. UK and US radiologists and other international organisations do not recommend CT scanning as the first line of screening tests [3]. Preliminary reports indicate that chest radiographs may have diagnostic limitations in Covid-19 [4].

AI suppliers need to be aware that lung CT scan findings often include multiple mottling and ground-glass opacity (GGO) [4b] which may not detect Covid-19 accurately but highlight that there is a problem in the lungs which requires investigation. GGO can be a manifestation of a wide variety of clinical features, including malignancies and benign conditions, such as focal interstitial fibrosis, inflammation, and hemorrhage [4c]. For example, lesions with GGO that do not disappear are often believed to be lung cancer or its precursor lesions [4d].

 

Treatment

Covid-19 treatment may be available as early as in three months, far sooner than a vaccine which will take between 12 months to 3 years of development. The contribution of AI in enabling a combination of antiviral and anti-inflammatory treatments is invaluable. BenevolentAI used its AI drug discovery platform to explore approved drugs that could potentially stop the progression of Covid-19. The aim is to identify a drug that could affect the virus directly by inhibiting the cellular processes that the virus uses to infect human cells. BenevolentAI identified Baricitinib, a drug which was originally developed by Eli Lilly. Baricitinib has entered clinical trials fast, thus reflecting the urgency of the global pandemic and the significance of AI and advanced technologies in facilitating the discovery of treatments and their potential positive impact on patients. [5].

The partnership between an AI company Iktos and research centre SRI International aims to combine artificial intelligence and a novel automated discovery platform for accelerated development of new anti-viral therapies. They will use deep learning to generate many novel drug candidates that scientists can then assess for efficacy.

A study led by NYU Grossman, in collaboration with both the Wenzhou and Cagnan peoples hospitals, has reported their AI tool can accurately predict which patients that have been newly infected with the Covid-19 virus would go on to develop severe respiratory disease. When fully developed, physicians will be able to assess which patients need to be hospitalised, and who can safely go home. This AI tool found that the severity of this condition is based on three features; the levels of the liver enzyme alanine aminotransferase (ALT), reported myalgia and haemoglobin levels. Again, this tool is currently limited as it has been developed by a small data set [6].

 

Prevention and control

Embedded systems such as AI-driven robots have a big role to play in outbreaks such as Covid-19. Chinese state media has reported that drones and robots are being used by the government to cut the risk of person-to-person transmission of the disease. These drones and robots remotely disinfected hospitals, delivered food and enforced quarantine restrictions as part of the effort to fight Covid-19 [7]. A company from Denmark is also selling a robot that can disinfect indoor spaces. [8].

 

Data

“..Most won’t be easy. Some we won’t like” [MIT Technology Review]

Incidents such as this epidemic may need international agreements that will allow patients’ data to be collected and shared responsibly to AI communities. Dr. Eric Topol from the Scripps Research Translational Institute is running a study using an app that allows people with smartwatches to track their resting heart rate and collect the data in an anonymised format, to protect patient privacy. This type of data can help scientists spot viral outbreaks. There’s also a need to analyse patients’ data to identify those at risk and predict the likelihood of some people requiring intensive care treatment in the case of hospitalisation.

We are currently missing the most important aspect of the Covid-19 response, which is developing standardised data collection models and generating synthetic data resources out of it. There is still time to improve this by developing platforms that will enable the collection of rich data that might inform a fast and coordinated national and international response. At the moment, individual nations are picking their own models to predict when cases might peak, with different localities using different formulas and private groups coming up with their own projections. A bigger-picture study, experts say, would allow individual governments to connect the dots and deliver a proactive response, rather than waiting to see where the virus emerges next.

Sometimes it is unclear for AI developers to know what is going on in hospitals. Prediction tools could generate accurate and useful information if public health data wasn’t locked away within government agencies, as it is in many countries. At the moment AI is forced to lean more heavily on readily available data like online news. It takes a while for the media to pick up a potentially new medical condition, and when it is realised, it is normally too late. There is an opportunity to finalise and use differential privacy to build trust among data owners and at the same time allow scientist to use this data to support research and development. Differential privacy is a mathematical technique that makes this process rigorous by measuring how much privacy increases when noise is added. The method is already used by Apple and Facebook to collect aggregate data without identifying particular users.

 

Conclusion

Initiatives to reduce the threat of Covid-19 are currently in progress. The most immediate and pressing matter is the development of drugs and vaccines to treat this viral condition. Technology, specifically AI, can speed up the process of identifying candidate molecules for drugs and vaccines. Also, AI is able to support the process of investigating any drugs and vaccines that have the potential to be repurposed for Covid-19 treatment.

The accuracy and speed of achieving this will depend on the data available. This is a global pandemic – there will be a lot of data available from every country. However, the biggest challenge of all is to create data collection models, and develop trusted platforms where governments can share information. There are a few different options for managing these issues. One option, is to relax patient data ethics while we try to deal with the pandemic, identify a trusted body to host all the data, but also allow global access to this data, subject to sufficient security checks.

 

Other options include:

1. Strengthening AI research initiatives that aim to develop AI solutions based on small datasets.

“Small and limited datasets are strongly represented in the domains of artificial intelligence in healthcaresmart operations and predictive maintenance and autonomous vehicles.” [9]. The learnings from small and limited data sets allow both the AI suppliers and users to leverage the benefits of current Artificial Intelligence developments without needing unaffordable large annotated effort.

2. Synthetic data generation.

There are a lot of companies that are currently developing tools to generate synthetic data for different applications. More research and collaborative initiatives need to be encouraged to develop this initiative further.

3. Encourage the development of more AI tools.

Tools that are developed, based on small data sources, to be open source for people to experiment on and use them to generate data. This data will be used to further develop the next stage of these AI tools to achieve accuracy faster.

 

AI communities who are developing tools to reduce the threat of Covid-19 should be aware of the data limitations and subsequently avoid providing false hopes. Also, doctors and governments in the world are sceptical of the use of CT and X-ray imaging in the diagnosis of Covid-19. There are currently not enough images to learn from and this is not a simple problem that can be trained in ImageNet in the first instance. Like any other system, ImageNet has never been familiar with images related to Covid-19, and so, its reliability will be limited.

Symptoms of the disease may not show up in scans until sometime after infection, not useful as an early diagnostic. A chest scan requires an enclosed space with staff nearby which, unless what AI has to offer progresses, this may prove to be more expensive and time consuming than taking a swab.

Covid-19 broke out and no one had any knowledge about the symptoms and threats it carried. At the beginning of the epidemic we were told that cough and fever were the unequivocal signs that we could have contracted Covid-19. It is only now that we know the disease can cause a wide variety of symptoms and even no symptoms at all. Doctors are also using their intuition to support patients by providing them with oxygen, paracetamol or NSAIDs or putting them on ventilators if the situation becomes severe.

Emerging AI tools are intuitive, having being developed with larger data sets than what doctors can see. Understandably doctors are very busy. They have no time to reflect on the information they see or capture from each patient. Their immediate focus in the moment is to support patients to get better.

While this is the case at the moment, it is fair to say that AI tools should be trusted to offer varying support at this time when we do not have a drug or vaccine. These tools should assist in the diagnosis, prediction and help us find the treatment faster, and so their development should be encouraged and supported. The most important thing to stress here is that AI tools are purposely there to assist humans and not to make decisions on behalf of humans.

Also, AI tools will be our saviour in collecting large amounts of data at these confusing times. When this data is analysed, it will help the world to predict, prevent and also be able to deal with future pandemics.