-
Where AI device should be in the image data flow?
Looking for some perspective on potential issues…
We are considering putting in an AI solution for. The vendor is asking about our PACS capabilities and where the device should be placed in the image flow. We are dealing with multiple PACS systems of varying capabilities. The AI device does not work on all x-ray, some will be just passed through. We are targeting ED imaging, so throughput speed is a concern.
Thoughts appreciated… We are in the process of consolidating some of our PACS, data migration being the rate limiting step. We are looking at the rules we can put in place to avoid some of the issues. We are also trying to understand how many images are going to be affected to understand the manual workarounds that might be created depending on our implementation choices.
Other things we should consider? -
3 Replies
-
Consider running away from a vendor who can’t tell you where the best placement is relative to your needs.
Multiple PACS with varying capabilities all using AI. Nice. You planning on individual interfaces to each PACS? Lots of time and $$. How are you justifying the purchase? I have 100+ questions but my first one is what AI vendor are you using, have the different PACS vendors interfaced with the AI vendor yet, where, how ws it done…is it a seamless interface that launches on its own? So many questions….
-
The only interface is DICOM, there is no PACS client-side integration. The augmented images come in via DICOM. Similar to MG CAD.
This is a pilot. The vendor wants to put the device in parallel with the PACS to avoid their solution becoming the bottleneck. I prefer to put it in front of the PACS to avoid the race condition, but wary of the potential bottleneck.-
I suggest you point the vendor at IHE Radiology’s [link=https://www.ihe.net/uploadedFiles/Documents/Radiology/IHE_RAD_Suppl_AIW-I.pdf]AIW-I[/link] (AI workflow for imaging) profile and tell them you want a standards based-solution. It should save you and the vendor headache in the future.
Putting the AI between the modality and the PACS opens the possibility that an AI system failure will prevent any reading of those images, which is probably a worse failure-mode than a race condition on the PACS.
Also, is there one AI system per X-ray scanner, is there on per PACS, or is there one for the entire deployment? If the PACS receives the images from the AI system, will you still be able to distinguish the originating scanners (i.e. the AE title may identify the AI, not the modality).
Most commonly, and simply, I am seeing PACS configured for auto-forwarding of appropriate (e.g. based on procedure code, modality, body part) images to the AI. The “race” condition is either handled according to AIW-I using UPS to coordinate when a study is ready to read; managing study “read-ready” state using another mechanism; or just letting the workstation’s “additional images are available” functionality notify the radiologist.
-
-
