-
Thought you would like to see some dialog on the article posted by someone who has been in the industry as long as I have: :
WHAT THEY SAID:
Mike – This is a great opinion piece – but that goes without saying.
Consider that AI has been here for a very long time. An animal by a different name – an algorithm. Many pre-AI buzz algorithms existed as best-fit models for filling scheduling slots, models for reconstructing or enhancing lower dose imaging, or algorithms converting signals to images. We have managed to build trust in all these tools because of validation.
However, the 500+ companies and AI applications beyond only imaging promise to be impactful with reporting, and data mining. Validation and ongoing monitoring of performance remains a challenge to build trust.
In the US, we continue to hear about cost recovery as a barrier to adopting clinical AI imaging models. Like the days of PACS, the debate is who pays for these productivity/clinical sensitivity tools – indeed to be a big topic at RSNA.
Perhaps we should consider if there is a parallel in how PACS companies differentiate their software similar to how modalities differentiate their offerings by “including” AI triage tools like modalities that use AI for reconstruction/detection in their systems. Is this a sustainable model for AI?
As always – thanks for sharing.
MY REPLY:
Always good to hear from an industry old-timer like yourself and get your perspective.
For the most part the “animals” you identified we have trust in because they do what they say they would do. The biggest challenge AI has in medical imaging has is defining what exactly AI is and is supposed to do. Many- but not all- of the 500+ AI vendors are offering pattern recognition- CAD- and pitching it as AI. That is not what AI is or is supposed to be.
AI uses information from select sources – typically images and reports- to complement image interpretation and propose a diagnosis based on disease-specific data . “True AI” uses information from multiple clinical sources to compliment image data interpretation and propose a disease-agnostic diagnosis. The number of vendors doing AI correctly are few and far between. Frankly, most of AI in medical imaging today is used as a confirmation tool- did I call it right and did I miss an incidental finding, nothing more, nothing less.
Cost recovery can be a challenge but if you know how to correctly promote AI that isn’t as big an issue as it’s made out to be including what I have said in the past. I would outline how a soild ROI in AI can be shown and sold to the C-suites but frankly I’ve had too many ideas borrowed from me and presented as their own so I am playing this close to the vest.
Differentiation? Most people developing the algorithms have only see a radiology department when their kid broke their arm and have no idea how radiology and radiologists work and the tools used and needed. That is up to marketing to bring forth a solid message. I can identify maybe a handful off good marketers in this marketplace if that. They know what they offer and the market needs and present a viable story. The rest of the vendors are just directing smoke into my nether regions…..
AI will be a part of PACS once the market trusts that it won’t slow reading time down. This is especially important recognizing that once AI gets included as a diagnostic tool it will initially increase payments…. until CMS reviews it and put the kibosh on it just like they did with CAD in years past.
So that’s my opinion. I don’t expect many to agree with all I said or even most. I at least want to go on record so that one when of the newly christened “experts” says something and the marketplace responds with hearty “Here, here!!” I can sit back and smile saying “I said that years ago”.
Mike Cannavo “PACSMan”
(407) 247-7345
