One of the interesting AI projects presented at ECR yesterday. The software engineers were very open about the difficulties the algorithms encountered when reading hyperactive stroke in the 1st 6 hours on CT, just like us radiologists!! They do not claim to replace radiologists but are able to send a notification about an abnormality detected by their software. It is CE certified for use it seems. Any PACS or modality vendor integrating these kind of AI decision support software with their products.
Hi Neelam. That is really interesting. I have not thought much about the correlation before. But realizations like this would probably be useful also in work allocation. I.e. not only abnormality notifications like this one to drive worklists but if we're also able to aggregate statistics from many AI applications like this one, since they all focus on quite narrow use cases. The radiologists' input in combination AI algorithms could probably more accurately rank the difficulty/points for specific exams more accurately when planning workloads. That is not something I've heard talked about before.
In any case, assuming you are at ECR now, head over to the Sectra booth if you are curious to see what a QureAI<-->Sectra PACS integration for workflow assist and prioritization support could look like in practice.
Another interesting AI application for Brain CT https://brainomix.com/e-aspects presented at ECR To support stroke management by the interventional neuroradiologists.
These kind of AI applications should enter into standard clinical practice. 2 areas to consider: 1. How do we incorporate it into the PACS reading workflow. Something we need to talk to PACS vendors. Yes, Filip it would be useful to chat about implementing AI decision support into a clinical workflow. 2. How do we get NHS to agree to fund this? There is no real cost savings from this--as you would still need a human report to be created. Are there any funding pots to tap into?
The Change Healthcare Workflow Intelligence tool fully supports this workflow. From helping to decide which cases should be sent to the AI product to then using the findings of the AI product to influence the urgency of the examination on the reporting list. To see how we do this, come and see us in the Radiation Safety Area at ECR
All PACS vendors are in a race to implement AI! Every PACS vendor has an AI corner at ECR-Sectra, Carestream, Agfa, Fujifilm were some vendors I visited today. Most are integrating with validated software already becoming available from various AI vendors. Vendor neutral integration is happening. Brain bleed, skull fracture, spine fracture, CXR shadows, CXR tubes, mammography abnormalities/calcifications etc are all being made available as decision support outputs: 1. However careful workflow adoption is required Images need to go from Modality to AI server then be visible on PACS-to avoid radiologist reporting them before AI output is a available! 2. Display by vendors are generally by a toggle on and off icon for superimposed AI markers and text outputs. 3. Some have a accompanying DICOMised PDF report from AI added as a separate series which is useful. 4. Something that is lacking in ALL PACS is the disclaimer and information about the sensitivity and specificity of the AI algorithm used. I think this information is important to ensure frontline doctors and radiologists do not consider the report from AI to be 100% accurate and remain aware of its limitations. Currently none of the vendors display this information which worries me. All the scientific papers by the AI developers are happy to accept the limitations of the software. The limitations need to explicit. 5. Worklist prioritisation is happening on PACS worklists from the AI information. For those of us with RIS driven workflow we need HL7 outputs from AI for worklist prioritisation in the following segments and fields a. Abnormal flag-OBX 8 b. Findings-haemorrhage, fracture etc etc-OBX 5 However, adoption of AI in PACS is certainly on the horizon. This is good news indeed.