Electronic Nose Uses Solid-state Sensors and Neural Processor to “Sniff Out” COVID-19

March 03, 2021 by Tyler Charboneau

Combining volatile organic compound (VOC) sensors and an ML chip, the "DiaNose" can pinpoint cancer, kidney failure, and now COVID-19.

A recent partnership between BrainChip Holdings and NaNose Medical has yielded a new COVID-19 testing solution boasting higher accuracy than RT-PCR screening. The Nano Artificial Nose is said to actively analyze patient breath samples while remaining highly portable. How exactly does it work? 


Building Upon Existing Technology

Officially dubbed the DiaNose, NaNose’s technology has actually existed since 2017. Based on the Technion Israeli Institute of Technology’s artificial nose, the device has since screened numerous patients for Parkinson’s disease, cancer, kidney failure, and MS. COVID-19 is the latest disease to join that diagnostic list. 

Like many other "electronic noses," the DiaNose works by detecting volatile organic compounds (VOCs) present in the breath. VOC levels can fluctuate depending on one’s level of activity or breathing patterns—however, the concentration of these exhaled compounds notably increases when someone becomes ill. According to BrainChip, these particles are biological markers for specific diseases. 


The basic working principle of a multiplexed nanomaterial-based sensor array

The basic working principle of a multiplexed nanomaterial-based sensor array. Image used courtesy of ACS Nano

The breath of diabetic patients is often reported as smelling fruity or sweet. Additionally, ammonia-like odors are associated with kidney disease. The body expels ketones—a class of VOCs. Over 3,000 different VOCs can exist in human breath. Zeroing in on those most relevant to COVID is crucial. 


Introducing New Patient Data

How have BrainChip and NaNose Medical enabled COVID detection? The answer is artificial intelligence and plenty of training. AI algorithms spot patterns in data that help scientists find deeper meaning in each pool of patient data they oversee—minus the tedious process of manual analysis. 

Scientists repeatedly expose such algorithms to specialized data groupings called training sets. Knowing what’s important in each set ahead of time, researchers hope that their AI can arrive at the same "conclusion."


Flowchart for detecting disease volatolomics

Flowchart for detecting disease volatolomics. Image used courtesy of the Small Journal

While developing the COVID testing functionality, NaNose Medical collected 130 patient samples via nanometer sensors. This array gathers gaseous droplets and inspects them for the presence of VOCs. Patterns are uncovered in the structure of these compounds while ignoring potential environmental contaminants. Ionization determines what VOCs are present. 

NaNose Medical doesn’t explicitly state which VOCs signal COVID infection. However, a UK study linked the presence of ethanal, octanal, acetone, acetone-butanone, methanol, isoprene, propenal, heptanal, and propanol to coronavirus. Interestingly, exhaled methanol concentrations were actually lower in COVID patients. It’s possible that DiaNose searches for similar markers. 


Processing Technology Facilitates Diagnosis

After gathering initial sensor data, scientists sent that data off to BrainChip. This data became integral in training BrainChip’s Akida neural processor using machine learning (ML). The Akida SoC is designed to work like a supercharged human brain—containing the equivalent of over 1.2 million neurons and 10 billion synapses. 

AI and ML excel in this application because of the black-and-white nature of VOC detection. Context and common-sense understanding aren’t critical (think Amazon Alexa), so DiaNose isn’t tripped up. The Akida chip has native support for these operations. 


Schematic of nanomaterial-based sensors detecting VOCs to identify disease

Schematic of nanomaterial-based sensors detecting VOCs to identify disease. Image used courtesy of the Small Journal

Akida also doesn’t require many of the complimentary microelectronics (memory units, external CPU, etc.) needed to power other SoCs. It’s thus much easier to integrate into a compact, portable device like the DiaNose. It’s also much cheaper than other complex neural systems. 


The DiaNose Advantage

Nanometer sensor arrays hold immense promise in future testing. A study from August 2020 unveiled an accuracy of up to 94% when differentiating COVID patients from control groups. Additionally, such methods are up to 95% accurate at distinguishing COVID from other lung infections. 

Because DiaNose and Akida can be packaged together, medical professionals no longer need to send test kits off for remote analysis. This testing procedure may produce an accurate COVID screening in only a few minutes, allowing a patient to know his or her results before leaving the office. Instant results allow positive individuals to isolate accordingly, while negative patients may rest easier.

Respiratory droplets are vehicles for disease transmission, which makes accurate breath testing a powerful tool. The apparatus’s handheld nature makes it simple to use. Professionals will no longer need to collect and store hazardous, biological samples. 

DiaNose represents a possible diagnostic turning point—all while incorporating evolving software and hardware technologies relevant to engineers.