[lijf]

You are of course right about the cost and accessibility, low field not necessitating specific rooms or ‘no go’ zones, and presumably easy operation (controlled via iPad app) being very interesting with regards to increased availability of MRI scans(in Sweden the rate limiting step at the moment is usually the availability of trained MRI technicians). In my initial comment I mostly thought about some clinical situations in my current practice where it would be useful. Sorry if the following becomes a bit rambling.

Reasonably there are a plethora of indications and given that these machines will be installed at the point of care (A&E, ICU, etc) and supposedly are safe to use on basically all patients, one can imagine that it will be used very frequently. Though in this case the scan-time (30 min + change) could become a limiting factor (but, given it’s price, I guess you could just buy more machines). The image quality is of course not the same as with a modern clinical system but it is most certainly good enough for a preliminary test, and may very well decrease the number of patients sent for a ‘normal’ diagnostic scan (though the opposite may also be true, if you see something unexpected in the images, or if there are artefacts which are difficult to interpret). As with the ‘fast MRI’ I referenced to earlier, my current idea would be that it can be used as an initial screening, but, as always, if the test is normal, but the clinical suspicion persists one would have to continue with other tests.

Regarding whether the image should be read by a specialist or not, I am of course biased towards the specialist :) As I see it the main advantage of having the clinician interpreting the image is that they have ‘direct access’ to the patient, and can ask them specific questions based on the findings of the test. Though, time-utility wise it’s faster for a specialist to read a normal test. During med school I did a rotation at a primary care facility in the north of Sweden, in a municipality with 6000 inhabitants, located almost 2 hrs away from the closest hospital, the GP:s there read chest x-rays themselves at that time. Now the images are transferred digitally to the main hospital of the region and read by a radiologist. One can of course wonder whether it’s cost effective to train a large number of radiologists to read all the new studies which will be produced. My hope here is that computer aided diagnostic tools will increase the productivity of radiology specialists.

My experience from working together with neurosurgeons and neurologists who are very good at reading images in general, is that they anyway prefer to consult us for the interpretation, (they would rather operate than become specialist radiologists). In my opinion the professionals at risk of being side-stepped are the general radiologists, and I think this is already happening as telemedicine makes it possible for smaller hospitals to pay for specialist readings on a per study basis.

edit: regarding time utility, one would of course have to include the time it takes for the referring physician to write a request, and for the radiologist to write a report. And another advantage of having the clinicians read the image is that they have more information about the patient than they put in the request. So it’s also very possible that radiologists will not be consulted (depending on medicolegal circumstances and reimbursement systems etc.)

[Amarok]

My experience as well is that there's a relationship of trust between radiologist and surgeon that a trained algorithm cannot replace. Same reason why AI isn't a threat to sales people for high priced items.

I'm actually doing what we call a taster in neuroradiology and it's been fascinating. Strongly considering applying for residency next year!

I did my elective at the PET center in Turku, great team. They even let me play with the rat MRI for Alzheimer's. They had some cool projects under Marco Bucci at the time, I believe he just started at Karolinska.

[radicalbyte]

Interesting, thanks for getting back :)

I worked for a couple of years in medical imaging (on a RIS) and our customers were very Radiologist focussed for everything except head scans (where the neurologists also read the scans).

I would hope to see AI-based assist technology hitting the market in the next 3-5 years for X-Ray/CT. Something in the form of the different rendering modes that you can use in a viewer. It's hard to tell though because the whole field is driven by hardware not software. Software seems to be an afterthought. The software seems more focussed on lock-in (integration, meh).

I had a vision for a future where the doctor in a hospital could push an image out to the "cloud" of teleworking radiologists, with the image being read in real time. It would be like load-balancing all of the work over the entire world. Technically it's tractable but the combination of privacy issues, the aforementioned hardware-driven market and the state of Hospital IT is what makes it impossible. Well that and the scaling problem (you need a critical mass of top-quality teleradiologists on your side to do it).

[exikyut]

Yeow, 30 minutes?

I'd say the couple of MRIs I've had done took around 10-15min or so. I think I shifted a few mm slightly at one point (possible since you're not quite bolted down, just held firmly). For 30 minutes I'm quite sure I'd probably shift around quite a lot, either involuntarily, because I got uncomfortable, or because of mild claustrophobia. :/