-
AI in your practice
Im curious how many rads are currently using some AI in their practice? CAD does not count, but everything else does. I assume it adds more time to read the studies because you have to review the AI interpretation/findings. After feedback and refininements from our current generation of radiologists, it will eventually get to the point where it will be able to do reads for normals on its own im sure. Up until that point though, we are increasing our time spent reading studies and helping to refine the algorithms. Should we not expect some extra compensation for doing so? Yet, everyone is doing this for free. How about asking for .1% of all future revenue that the software generates in perpetuity? .1% doesnt seem like a lot, but it will be because the software will be making billions.
-
14 Replies
-
They arent.
-
Until there is a reimbursement, no one will.
-
We use RAPID. Its been useless. The perfusion only works for mca occlusions of which no one needs help. Stroke resident was going to blow off a basilar artery occlusion because RAPID gave it the all clear. Had to explain skull base makes perfusion unreliable. Great example of a false sense of security from technology. By the same token there was concern for vasospasm on a CTA from neurology because of RAPIDs differential in arterial opacification. Had to explain RAPID was just counting a massive aneurysm on the contralateral side as vessel opacification – making the normal side look bad. All in all its been a work generator and not a help.
-
We have research with one of the major vendors for a pilot project to use their AI package to check for PTX to prioritize critical case review on the work list.
-
How is it doing at finding PTXs? The pulmonary nodule AI that i have seen performs horribly; so much so that i dont even look at it
-
Just read a neuroquant that called asymmetry in the hippocampi. I didnt appreciate it so I looked at the numbers generated by the neuroquant. It was calling one of the hippocampi as 20% of the cerebral hemisphere brain volume. Thats not only absurd, its dangerous. You dont get second chances at hippocampectomies.
-
We use Clearread for nodule detection. Mixed bag. Lots of misses for subpleural location. Agree 110% that it adds to interpretation time and not as time saving as company lit would have you believe.
-
Quote from brickydragon
You dont get second chances at hippocampectomies.
[:D][:D][:D]-
Our nodule AI is terrible as well. What I really hate that you really cant adjust the margins it thinks are part of the nodule. Its okay for round but non round nodules geez. Also its so inconsistent in size of nodules it measures. Example I measured these 2 modules but had to start all over cuz computers crashed. Pulled up the study and had it measure nodules and now they measurements are different again. Why is it different? Did the nodule grow on both the new and old studies when I rebooted? Sometimes it says nodule isnt new and it clearly isnt new. I can see it but on a different slice than the AI is looking at. Sometimes I have to explain in my report AI says its new but its not correct. Once they asked me if I want to talk about their product to prospective clients. I said only if they are okay with me saying the truth. They asked the admin to ask me. When she told them what I said they asked her to ask me what are the issues? Where should I begin? You have paper?
I began listing.Turns out they didnt need me after all.
Also said to me through the admin that its odd.. the issues I was reporting and that no one else seems have to an issue like mine. I replied to PACS admin
I guess its just me then. Just happens the AI software just wants to annoy me and it makes an effort to follow me no matter what workstation I use. Maybe I insulted its ancestor when I said I hate Bing and like google better. I hope if I buy a Tesla in the future it doesnt talk crap to my car. Yep its me ..not your program
The admin knows its terrible and she hates dealing with them for other issues related to the AI which cause her to do additional work.-
Anyone using AI software for non contrast CT scan head bleeds? I imagine the early AI software should be really good for that.
-
Im interested but Im betting it will be awful. Workflow interruption to evaluate choroid plexus 95% of the time.
-
Why would it be really good ? Its probably gonna overcall a ton of frontal artifact
-
One thing I have learned from dealing with vendors in healthcare:
Whatever problem youre experiencing you are the ONLY person (or site( having that problem and NO ONE has ever raised it with them before.
Oh, and no, they dont have a users forum.
-
The head bleed and rib fx AI seemed pretty good at the RSNA. Of course if it wasnt they just find better images until They got desire effect.
-
-
-
-
-
-
-
-
-
