Frequency comb breathalyzer detects Covid with excellent accuracy | Hacker News Reader

Frequency comb breathalyzer detects Covid with excellent accuracy

nist.gov

95 pointsgeox8 comments

I remember in 2020 when we thought we'd soon have accurate instant tests at the door to venues so we could with high confidence have safe events. I still wish we had that (today's "rapid" tests don't really work for people who recently became contagious).

There were also dogs who could smell Covid, and a big name (MIT?) was going to release an app that could hear your cough and detect if it was Covid... I wonder about those too.

Having a pet dog that can smell it would be awesome, or maybe worrisome, because you might end up dodging people left and right on your dog walks.

This technology behind this is very cool - I hadn't heard of optical frequency combs before.

My amateur interpretation is that the device shines light at a breath sample, capturing the signatures of the molecules in the sample (kind of like mass spectrometry), and the researchers trained an ML algorithm to classify the signatures as "COVID" and "Not COVID." For the ML part, the training set was n=140 and the testing set was n=30

I think the point of this study was to see if this approach was even possible, i.e. a proof-of-concept. They have a lot of work ahead of them to see if this could become a robust enough system to be used widely.

Frequency combs are awesome, and exploit the relationship between temporal and spatial distributions of light.

To have a short (femtosecond) pulse requires it to have a large spectral bandwidth.

Therefore to create a frequency comb to span a large span of wavelengths requires starting with a femtosecond pulsed laser and all the associated optics to amplify, stretch and "comb" the output.

This is very expensive and bulky, there are some interesting approaches to produce chip-scale frequency combs which may make devices such as this more accessible.

Whilst n = 170 is not bad, this would still need bigger testing group. But seems promising?

The researchers measured an AUC of 0.849 for their COVID-19 predictions. An AUC of 0.8 or greater for medical diagnostic data is considered “excellent” accuracy.

The AUC of a rapid antigen test is over 97%, right?

That sounds way high, even when the antigen tests were new, if the 'true' condition is considered PCR-positivity.

Rapid antigens miss tons of PCR-positives, but that's not as bad as it sounds because plenty of those PCR positives are unlikely to be very infectious, if at all.

Also, I've seen reports that antigen test false-negative rates have been rising, modestly, with newer variants.

Yeah, that was my next comment --- I just had COVID and tested negative 3 days into symptoms. My question would be, if that's the performance you get with 95+% AUC...

Reply on news.ycombinator.com