Invoking AI: don’t impact clinical practice

When developing imaging AI, it’s important not to lose sight of the end-user. All the special features and stunning visualizations in the world will be for

post image

When developing imaging AI, it’s important not to lose sight of the end-user. All the special features and stunning visualizations in the world will be for nothing if they detract from the work of the radiologist. Anything being introduced to try and improve efficiency, throughput or clinical value must not get in the way of clinical practice.

It’s also important that the radiologist is able to interpret the results of AI within their normal reading environment. As soon as you need to open a separate viewer or even bring up a different user interface within the PACS, the context switch delays radiologist productivity, and the support of that viewer component means that there's an additional IT burden to manage. Every AI requires some time to generate a result – in some cases it's seconds, in others it's minutes or hours depending on the complexity of the task. From a workflow perspective, the processing has to be as fast as possible and sometimes that requires the use of a GPU or cloud-based resources – whatever can be done to make the processing faster. However, sometimes there will be a delay, which then raises the question of how to present that information to the radiologist. An extreme decision would be to hold a study from a radiologist’s worklist until all the relevant information has been added. But if it’s an urgent case, or one that the radiologist feels they need to read immediately, then holding things up could delay patient care. A better route is to be able to allow a study that is still being processed to appear on the worklist, but to notify the radiologist that some results are still pending and that they shouldn't yet read that study unless they have to. Our partner Intelerad’s Clario SmartWorklist™ is a good example of how companies are working to address this challenge.