Brochure (pdf) - https://p.widencdn.net/jdgicj/Cepheid-GeneXpert-Xpress-Syste...
Promo video - https://www.youtube.com/watch?v=lPTHRqDyWC4
About the size of a small laser printer.
I'm not clear how it returns results so quickly. I thought that the RT-PCR process required numerous heating and cooling cycles which is what makes it take as long as it does.
Edit: the keyword to look for might be "isothermal" amplification, which means same - temperature amplification. Again, no clue if that's what they're using, I just see that they have patents on using isothermal techniques. These allow for amplifying DNA without cycling the temperature, usually letting the reaction accelerate quickly until a signal is detected, and it can accelerate really quickly sometimes. You know those videos of supercooled liquid water where you bump the bottle and the whole thing freezes? Imagine that you have a situation like that where adding non-coronavirus dna bumps the bottle incorrectly and nothing happens, while adding coronavirus dna makes the whole thing freezes. It's a rapid signal that just gets generated, but often (like with lamp) you just end up with a polymerizated mess that you can't use for DNA engineering in synthetic biology which is why you might use your temp. cycling methods for those uses. Also, they're good for telling you if something is or isn't there, but not so good at telling you how much dna you started with. To figure that out you again use a cycling reaction with the amount of dna doubling each step until it levels out so you know how many doublings happened and can calculate backwards (quantitative rt-pcr, or qPCR).
https://patents.google.com/patent/US8900828
https://en.m.wikipedia.org/wiki/Rolling_circle_replication
https://en.m.wikipedia.org/wiki/Loop-mediated_isothermal_amp...
edit: as pointed out below - the largest of the machines has 80 slots and could probably do 2000/day taking around 45min-1hr/test all in
Contrast that with a modern real-time PCR machine which typically has capacity for 96 samples, with some machines able to test 384 or even 1536 at a time. The tradeoff is that those machines take 2-3 hours and the lab needs to do nucleic acid extraction separately.
The main advantage of this machine is that it does the nucleic acid extraction and PCR in one cartridge and in ~20 minutes. It's great for clinics and field hospitals but for an established laboratory running hundreds or thousands of tests per day, I'd wager that the standard methods are much cheaper and a bit faster overall.
It's pretty neat-- they have 1, 2, 4, or 16 slot things that you can slot samples into as you're ready, and they process in the background... or the big robot. It's also neat in that they have combined cartridges to do a couple of influenza detections + RSV in one rapid test (with a different color of fluorescence for each).
But you're spot on in that this has a costly consumable (the cartridge)
The nice thing is that it catches acute infections even earlier than IgM.
Where I think there's room for them to introduce false positive rates (that we would not know without having access to proprietary information) is in how they design something called a primer sequence. this is a little bit of DNA that will go and find to a different little bit of DNA that's part of a larger bit of DNA. Basically it lets you target it a little bit of DNA within a larger bit of DNA. you want to make this primer so that it only will bind to coronavirus DNA but will not bind to any other DNA. It's important to design it that way because in PCR the DNA that amplifies to give you the signal is whatever that primer was able to bind it to.
If you accidentally make a primer that can bind to some non coronavirus DNA (something we call off Target binding) then you would end up with false positives. There are bioinformatics tools to evaluate this and presumably the FDA and the company did some work into making sure they didn't make any mistakes when designing that, but that's what I think of first when it comes to false positives.
The fast turnaround time is from integrated sample prep and reagent cartridge. It just uses standard thermocycling PCR. PCR doesn't take long when volume is small and heating surface is large.