One of the greatest challenges in treating patients with coronavirus is an overall lack of information about how the disease progresses. In order for doctors and epidemiologists to accurately track and diagnose this information in large numbers, each patient would have to be monitored 24/7 through the entire infection cycle. Adding to this complexity are patients who appear to be asymptomatic, but in fact have the disease and are not being tracked at all.
Researchers at Northwestern University and the University of Illinois at Urbana-Champaign have developed a way to not only collect the data through novel wearable devices, but also organize and analyze the massive data sets required to provide health professionals with an accurate picture of disease progression. The project is being funded through the National Science Foundation’s RAPID program.
Former University of Illinois at Urbana-Champaign Professor John Rogers and his research lab at Northwestern developed a wearable device that is being used to track COVID-19 patients and healthcare providers at several Chicago-area hospitals. The device can monitor respiratory activity such as coughing and shortness of breath, two key symptoms of the virus.
“We’ve developed a novel wearable device that sits at the base of the throat and continuously monitors the wearer’s cough, heart and respiratory sounds,” said Rogers, the LouisWhile tracking already infected patients is important to measure the disease progression and response to treatment, it is just one mode of use. Frontline healthcare workers and high-risk communities such as nursing homes are also important to monitor, so that contagion can be identified early to reduce the risk of spread.
Rogers and his team were able to monitor a nurse before, during, and after she had contracted the virus. The data collected, in combination with the nurse’s detailed notes throughout her illness, allowed the team to pick up heart rate spikes and changes in coughing activity that would have gone undetected in standard non-ICU hospital care. This type of continuous, at-home tracking through a wearable device is unprecedented in the healthcare industry.
There are 50 devices on people in all three of the monitored populations. Currently, the devices can track metrics such as coughing rate and count, respiratory rate and sounds, heart rate and sounds, and body orientation and activity – as well as correlating the data to the time and date. Every few days, the wearer can take off the device and place it onto a wireless pad for charging and wireless data transfer to a tablet or smartphone. The data then passes automatically to the cloud where algorithms yield basic parameters that their physician can view and interpret it.
The sheer amount of data that is generated by the devices can be overwhelming for physicians or caretakers who are monitoring dozens of patients, however. When field tests of the device first began in mid-April, the original 25 patients generated more than a terabyte of data in the first week of clinic and home wear. That’s where Shanbhag’s team comes in.
“Naresh’s group will be able to unlock detailed, and sometimes subtle, parameters from the raw data, because ultimately, the physicians want important information, not raw data,” said Rogers. “They want to know the meaning of the data and, with Naresh’s algorithms, insights into the progression of the patient.”
Shanbhag’s group hopes to develop an algorithm that can not only determine a coughing event but also analyze the nature of that cough – including its intensity, whether it’s wet orExtracting such data out of a patch that is barely thicker than a band-aid is no easy task. Shanbhag’s group has two main goals when it comes to processing the data: analyze the ever-growing set of data in the cloud, and move some of the processing to the patch, to make the entire process more energy efficient.
“We would like to bring powerful data analytics based on machine learning to develop prediction methods,” said Shanbhag, a CSL researcher and the Jack S. Kilby Professor of Electrical and Computer Engineering. “Initially we’ll be running that analysis in the cloud, but we want to bring those algorithms closer to the patch. We are concerned about both the accuracy and energy cost of making such predictions. The current process is energy expensive so the goal is to bring as much of the data processing closer to the patient as possible without losing accuracy.”
The current process requires data to be transmitted from the patch to a local device, such as a smart phone or a tablet, and then to the cloud, where it is processed. The group would like all of the processing to be done on the patch and device, as opposed to in the cloud, to save energy consumption. This would also alleviate the security and privacy concerns associated with transmitting and storing sensitive health related data in the cloud, where it more vulnerable to hacking.
Perhaps the greatest research contribution, however, is the potential to predict a patient’s progress through the disease.
“We will acquire data for both ill and healthy patients, and learn the characteristics of the data by developing COVID-19 specific, low-complexity machine learning algorithms,” said Shanbhag, “We’ll then use the learned models for predicting whether a patient is ill or not and how the disease will progress over time for new patients or individuals.”
As algorithms are developed by Shanbag’s group, they will be deployed by Rogers’ and his team through the devices, which will then provide improved data back to Shanbhag. This feedback loop will improve the accuracy of algorithms and reduce the complexity to enhance the capabilities of the next generation of patches.