A portable surveillance device powered by AI and machine learning has been invented by researchers at the University of Massachusetts Amherst.
The device, aptly named FluSense, can detect coughing and crowd size in real-time. The data obtained can be used for the tracking of flu-like diseases and transmission trends. It is believed that coughing may help health authorities better understand the spread and transmission of respiratory outbreaks in large crowds.
At a time when IoT companies are teaming up to help national governments, municipalities, and NGOs around the globe to curb the spread of the coronavirus by creating packaged (hardware + networking + software) technology solutions, new technologies can literally be the difference between life and death.
The researchers at UMass Amherst claim that in addition to wider public spaces, their new edge computing technology could be used in healthcare environments such as hospitals and waiting rooms to further expand the number of health monitoring tools accessible to them. It could also be used to help predict seasonal flu and other respiratory outbreaks. This proves that emerging technology can greatly reduce the impact of a pandemic in a time when these discoveries can be bittersweet.
About the device
Before the device was built, the researchers had to develop a lab-based cough model. Being in the US, the guidelines were followed as per the Health Insurance Portability and Accountability Act (or HIPAA, for short). It’s important for any medical research team to be HIPAA compliant in order to protect patient privacy. For example, any data collected during research that’s stored electronically or on the cloud needs to be HIPAA compliant.
The researchers then trained a deep neural network classifier to draw boundary boxes on people representing the thermal images.
By using a microphone array, thermal camera, and neural processing system to passively track and identify speech and coughing sounds, the researchers then concentrated on four individual waiting rooms at the health services clinic of UMass Amherst, installing the FluSense system in them. Between December 2018 and July 2019 more than 350,000 thermal images and 21 million non-speech audio clips were obtained and analyzed by the researchers.
FluSense was also able to forecast disease rates with a high accuracy rate by making use of the data obtained during this period. The researchers noted that the data collected “strongly” correlated with the campus test results for influenza and other respiratory diseases.
This gives us a very real indication of how much IoT could aid us in future pandemics while proving the value to be found when combining AI with edge computing to allow data collection and analysis right at the source.
Designed to predict population-based outbreaks and transmission
It is important to note that FluSense was not designed to identify where individuals are or who may be ill. Rather, it was designed to study and predict population-based outbreaks and transmission models. In this way, it could be used to help public health authorities and city managers identify when and where an outbreak is on the horizon.
FluSense may also be used to help existing attempts to forecast influenza, such as the FluSight Network, a group of flu prediction teams.
In recent years, IoT technologies and devices have found themselves under scrutiny after becoming entangled in security-related controversies, as such FluSense was designed and created to be an autonomous, privacy-respecting system.
One of the creators of the device, assistant professor Tauhidur Rahman has said that:
“I thought if we could capture coughing or sneezing sounds from public spaces where a lot of people naturally congregate, we could utilize this information as a new source of data for predicting epidemiologic trends.”
The logical next step would be to test FluSense’s capabilities in other public areas and places to validate its effectiveness beyond the current hospital environment.
Can we trust AI to fight COVID-19?
AI has become a part of our everyday lives. It has become so ingrained with everything we do, that it has even become one of the primary resources to monitor and report cases during this pandemic. Deploying such technologies often meant balancing the need to conquer the virus with the need to protect individual privacy.
As the initial crisis gives way to long-term strategies and public health initiatives, policymakers need to build confidence in AI to ensure that potential safeguards can be implemented and sustained.
Many are optimistic about AI’s capabilities, although others remain skeptical. The latter also reflects on a perceived threat to their jobs. AI’s superpowers are currently being used to help crack viral transmission chains across the globe.
For example, Russia maintains COVID-19 quarantines by large-scale surveillance with CCTV cameras and facial recognition. China uses AI-powered drones and robots to detect population movement and social gatherings, and recognize those people with fever or those not wearing masks. Meanwhile, Israel uses AI-driven communication tracing algorithms to send people personalized text messages to quarantine them after being close to someone with a positive diagnosis.
Most of this life-saving AI is powered by personal data. In fact, South Korea’s high-octane combination of credit card payment details, mobile location, CCTV, facial scans, temperature sensors, and medical records was a crucial part of a larger strategy to track communications, check vigorously, and implement targeted lockdowns. Combining these effects helped flatten the country’s curve.
Because FluSense can differentiate coughing from other non-speech audio forms, correlating coughing with the size of a given crowd may provide a useful index of how many people are likely to experience flu-like symptoms. From a technological perspective, FluSense is especially interesting as all practical processing work is done locally, through the Intel neural computing engine and Raspberry Pi.
Symptom information is then sent wirelessly to the lab for collation, but the heavy lifting is performed right at the edge. FluSense has significant advantages over other health surveillance techniques, particularly those based on internet monitoring, such as the Google Flu Trend and Twitter.
Despite these advantages, we also need to take privacy seriously, cautiously and pragmatically. This is especially important as people may be more likely than ever to neglect their civil liberties. Even though 88% of people believe privacy is more important than convenience, data protection authorities will likely grudgingly accept that exceptional times will overshadow even the strongest privacy rights.
Where privacy is curtailed, it is important to consider all aspects of AI ethics to preserve public trust in medium to long-term use. If organizations wish to ensure sustained public engagement, they must ensure that the data voluntarily given in the sense of providing a social good is treated with the utmost care. Since a vaccine is at least eighteen months away, long-term approaches are needed to help track efforts while maintaining public trust and cooperation.
FluSense shows that machine learning can support and improve flu prediction. When adopted at a more conventional stage, epidemiologists and other health experts may access vital data and gain a deeper understanding of how a virus spreads and identify external causes, susceptible populations, rate of virality, and more.
The technology may also allow medical professionals, governments, and even larger industry players to plan for flu outbreaks more efficiently by helping determine whether travel bans, social distancing, or medical supplies are required.