“One way we use it is to have AI scan ...”
Very informative post. Thanks.
In my mind AI is mostly pattern recognition which has been going on for years. What’s new is advances in image and sound processing, made possible by newer hardware and the army of software people slugging through the mud advancing the front.
The computer are just doing what they’re told as has been the case since Babbage.
I suspect is it pattern recognition put on steroids by AI/Machine Learning.
I visited a company probably 15-20 years ago, and they were trying to develop something like this without AI, and using a library of key images for comparison with a current exam and probably brute force computational power done over a wide area network that the Internet didn’t really have as well back then, but I don’t think they had the tools to make it work.
I suspect AI would change all that. It is interesting and useful, but I am in no way any authority on the workings of it, I just understand elements of integration into a workflow.
But AI can be useful. We have another vendor which uses AI to automatically generate impressions for Radiologist reports they dictate. One would think that wouldn’t be that hard in some respects. When a Radiologist dictates a report, they have different sections in the report such as the Description, Technique, Comparisons (to exams done before) Reason, Findings, and Impression.
So, they might dictate five paragraphs of Findings, then, when they get to the Impressions section, they have to re-dictate the key points from those five paragraphs of Findings and boil it down to a a few short key sentences that summarize those five paragraphs. Think of it as making five bullet points to describe a short story after reading it.
It is an area that takes time, because the Radiologist must comb through the five paragraphs and summarize only those in the Impression section. They have to be careful not to forget to include something, and to ensure they are accurate and don’t have a conflict with anything in those five paragraphs.
So, we provide the AI system a file with as many past reports as we can that were dictated by a given Radiologist, and it analyzes them to “learn how they talk”. Once it learns, when the Radiologist gets to the part they have to dictate the impression, it automatically generates an impression FOR them and...it does it in the grammar and syntax they normally use!
In other words, it looks nearly exactly the way they would dictate it themselves.
They initally found it kind of creepy, but...most of the time, they can just read it and click okay without doing much in the way of editing, so they came to like it. It cuts about a third of the time off the creation of the report, and minimizes the risk they might overlook some part of that five paragraphs they dictated and forget to include it in the report, which is a tedious part of the job they MUST perform correctly. AI just helps with that.
And...they STILL have to read it all the way through and approve it.