Using AI to help contain the Coronavirus

As I am writing this article, the number of people killed in China by the new coronavirus (2019-nCov) has risen to 82, with almost 3,000 confirmed ill. The national New Year holiday in China has been extended by 3 days to Sunday. Wuhan is in lock down and several other cities have imposed travel bans. To add, at least 44 cases have been confirmed abroad, including in Germany, the US and Australia and with approximately over 100,000 flights occurring per day, the risk of the condition spreading internationally is severe. It is clear that we do not need to reiterate the severity of the potential economic impact this outbreak will have, although that is not the focus of this article. I aim to propose how we can leverage artificial intelligence (AI) to help contain the health risks of such an epidemic.

The role of AI in public health is one that is still in its early stages, but has already made a number of significant leaps in recent years. By analyzing vast amounts of data from a variety of sources, it is possible to use AI to create predictive models for identifying future disease outbreaks. A well-known case where AI has helped contain an epidemic was in 2018 when the ‘Dengue Outbreak Prediction Platform’ was developed – this was able to predict where outbreaks of Dengue would occur in Malaysia and Brazil 3 months ahead of time with 80% accuracy. We can therefore use a similar architecture to help predict where outbreaks of the new coronavirus could happen next so that governments can accurately predict the resources needed in specific locations. This approach although has its own limitations.

The big question is – how do you rapidly diagnose for the new coronavirus? In short, there is no distinct answer, but diagnosis should be suspected in patients with signs and symptoms of pneumonia. According to the WHO, any individual that experience such symptoms usually exhibits a respiratory rate (RR) of > 30 breaths/min and an oxygen saturation (SpO2) level of < 90%. It is possible to measure both RR & SpO2 remotely and non-invasively through what is called (remote photoplethysmography) rPPG technology, effectively monitoring these vitals from just a video feed of an individual’s face.

Remote photoplethysmography (rPPG) allows for the uninterrupted control of human heart activity by detecting pulse color changes in human skin using a multi-wave RGB camera. Several important rPPG methods for video pulse extraction have been proposed in recent years and a better understanding of how core rPPG works can benefit multiple systems/protocols for video health monitoring, such as heart rate monitoring, breathing, SpO2, blood pressure, neonatal monitoring, and measuring atrial fibrillation and mental stress.

Since the input required to monitor such vitals is just a video feed, there is no better place to start than China. The country has an estimated 170 million CCTV cameras – that is about 1 for every 12 people – and is rolling out some of the world’s most advanced surveillance software. With developments in rPPG technology, there’s no reason why we cannot augment the features of these cameras to monitor people’s vital signs. To add, customs departments at airports could also use rPPG to screen individuals for abnormal vital signs as they pass through passport control – another application to prevent the international spread of the disease.

In quarantined Wuhan, fever clinics are singling out anyone with a fever of 99.1 degrees Faranheit or above. In theory, this is not practical because early symptoms of fever and cough are indistinguishable from the usual winter suspects. This method may lead to many false positives which can lead to self-inflicted transmission (i.e if you quarantine 100,000 people because they have surpassed the threshold for 99.1 degrees Faranheit and only 1% of that population actually have the coronavirus, you are risking a self-inflicted transmission of the disease to the other 99% within the quarantined area if you are not able to rapidly screen for the disease).

By no means is the use of rPPG to measure abnormal RR & SpO2 comprehensive in itself to lead to a definitive diagnosis of the 2019-nCov, but it certainly is an effective and scalable method using already in-place infrastructure (cameras) in airports, hospitals and other institutions. In future, we hope R&D into rPPG continues to allow for the greater use of such an application.

Interested in implementing rPPG technology within your organization? We can us to find out more.