Wearables Provide Speedy COVID Screening

There’s a new player in the effort to quickly and effectively screen populations for COVID-19: This week, a Princeton spin-off company launches a coronavirus-screening app for businesses that takes two minutes and uses data from commercial wearable devices.

In early clinical tests, the tool was 90% accurate in predicting if a person was positive or negative for the virus—even if they had no noticeable symptoms. Current rapid diagnostic tools, such as temperature checks, are not effective at detecting and preventing the spread of COVID, according to federal authorities.

The app, CovidDeep from Brooklyn-based NeuTigers, is powered by deep neural networks that identify patterns within sensor readings taken by wearable devices. Initially developed at an Italian hospital during some of the worst days of the pandemic, the tool is already in use in nursing homes in France and the US. This week, it launches for businesses who want to rapidly screen residents or employees, such as healthcare workers.

“We really want to contribute to mitigating this pandemic,” says Adel Laoui, CEO and co-founder of NeuTigers. The company will continue to gather more data to make the tool as accurate as possible for all people, he says.

In early March 2020, the NeuTigers team began adapting their StarDeep health platform (an AI system under development to detect diabetes) to screen for COVID-19. The platform relies on ultra-efficient AI algorithms that run directly on small devices, such as smartphones or watches.

This technique, called edge AI, allows a person’s private health data to remain on their device, rather than being sent to the cloud, a practice that some apps and AI applications have faced criticism for. In a recent IEEE report on the use of wearable sensors to combat COVID, data privacy and security were among the chief concerns highlighted by digital health experts. “Everything is happening on their device; nothing is going out,” says Laoui. “Users are in control of their health data.”

NeuTigers developed CovidDeep through clinical testing at San Matteo Hospital in Pavia, Italy. Led by IEEE fellow and NeuTigers co-founder Niraj Jha, an electrical engineer at Princeton University, the team collected health and sensor data from 87 individuals in three categories: healthy, COVID-positive asymptomatic and COVID-positive symptomatic. They then used this data to train deep neural networks to identify patterns of COVID infection.  Their grow-and-prune strategy, Laoui says, improved the algorithms’ accuracy while minimizing the energy required.

The team achieved an accuracy of up to 98.1% in testing with real and synthetic datasets, as outlined in a preprint paper published on arXiv and currently under peer review at a leading journal. Now, with data from over 500 people, the CovidDeep app takes 2 minutes to complete and is 90%+ accurate at predicting the presence of COVID-19 infection, according to a company press release this week.

To use CovidDeep, a user answers a short health history and symptom questionnaire on the app and inputs health sensor data. Currently, some of the data comes from an Empatica wristband: galvanic skin response, skin temperature, and inter-beat interval, the time between heartbeats. The rest is recorded by other off-the-shelf devices: blood pressure is measured with a blood pressure monitor, and blood oxygen saturation levels taken with a pulse oximeter.

The application is currently being sold to businesses and is not directly available to consumers. Laoui says the company will continue to adapt it for other health wearables, such as smart watches made by Fitbit and Apple. In the future, NeuTigers plans to apply the StarDeep platform to the detection of depression, Alzheimer’s disease and more.

Smart device-based tools have been touted as one of the best possible ways to quickly screen large populations for COVID and other infections. A recent popular avenue has been smartphone-based apps that detect COVID in the sound of a cough.

If a person receives a positive result on any such app, they should see a healthcare professional to follow up with diagnostic testing.

The original article can be found at: IEEE Spectrum