They say "Hospitals deploy" but where? how? who? I haven't heard any from colleagues in the US or China, even from people who do medical AI work. One major rol

They say "Hospitals deploy" but where? how? who? I haven't heard any from colleagues in the US or China, even from people who do medical AI work. One major role for AI that I see here is in prognostication. Are there features within radiographs or CT scans that could predict disease severity? Maybe ones that radiologists & clinicians can't discern yet? This would help triage care in capacity-constrained settings. But as the article says, that's "in the future". I haven't seen any robust research in this area yet, and a useful prognostic model would probably incorporate more than just imaging -- symptomology, vital signs, etc. For screening, on the other hand, there are some opportunities but also challenges. When PCR tests/materials/reagents are limited and caseload is high, CT scanning can be a powerful tool to screen and differentiate "sick" from "probably not sick", as was done in Wuhan and Northern Italy. But the AI doesn't offer a whole lot -- the findings are not particularly subtle. Amusingly, from the article: "Teams in China and the United States found that the lungs of patients with COVID-19 symptoms had certain visual hallmarks, such as ground-glass opacities... and areas of increased lung density called consolidation." This is repeated in every AI article on COVID imaging. Like, this is literally Radiology 101 -- any radiologist and many non-radiologist physicians would have been able to say this before COVID existed. Viral pneumonias (SARS, influenza, COVID) and other things (edema, atypical pneumonias, drug reactions) all cause ground-glass opacities. Consolidations happen when the lung gets socked in or super-infected (i.e. with bacteria). Another issue is that early lung infection does not appear on CT, which is common for almost all pneumonias. So CT misses early and especially early mild infections, which is not ideal for screening. And finally, interpretation capacity (what AI could help with) is not a bottleneck. There are TONS of radiologists who are underutilized right now as elective imaging and procedures are way down (e.g. not many screening mammograms or sports injury MRIs being done these days) and on top of that it would not take a radiologist very long to say "this is probably not COVID" or "this could be COVID" or "this is really bad lung disease". Even with lung scans being fast and if we used every CT scanner on them, AI wouldn't really help increase our throughput interpretation-wise. Of course, I write from the perspective of a highly-resourced health system. In the developing world, radiology access, like medical care in general, has always been an issue. I think Kenya as a country has like 100-200 radiologists for what, 40 million people? While a single mid-size hospital in the U.S. could have 200 radiologists. And Kenya's on the better side. So there AI could play an huge role in providing care (and certainly not just in this pandemic!).