As artificial intelligence tools are increasingly deployed in clinical imaging pipelines, radiology practices will need new mechanisms for assessing, managing and utilizing this wealth of information.
These tools generate separate data streams from raw imaging data and radiologists’ reports, leading to what Massachusetts General Hospital experts call a “new AI data paradigm.”
“As these new data form an increasingly important part of the value generated by radiology examinations, radiology practices will need to develop policies and processes for governing these data flows,” Bernardo C. Bizzo, MD, PhD, with MGH-affiliated Harvard Medical School, and co-authors wrote Tuesday in a JACR opinion piece.
Below are top considerations for rad departments.
Capturing data from multiple sources
Artificial intelligence may create data from single imaging exams, modalities, radiology reports, decision support tools and other sources.
Such software typically generates structured data, the authors noted, increasingly using the Health Level Seven International’s Fast Healthcare Interoperability Resource Standard. DICOM Structured Reporting and data tags outlined by the ACR and RSNA can also help organize information.
“Data from all these sources must be integrated, evaluated, and routed for appropriate disposition, all under radiologist review,” the team wrote.
The 3 levels of access
All AI-generated imaging data should be stored in a “general-purpose” data repository known as the AI Core Archive, the authors explained.
This houses info detailing data sources, identifies different versions, the technologists and radiologists who created it, and logs rad interactions with the data. Such information would also be used to enhance AI tools and continuously monitor software.
The next level contains images and structured data for radiologists to view while reviewing a patient's imaging history. This is only accessible within radiology systems or PACS.
Finally, rad departments should have an archive level with images and structured data accessible to the patient and their care team. This info can be added to medical records or sent directly to the EMR.
How involved should radiologists be?
Radiologists will play a key role in reviewing AI-generated data streams, depending on the tool and situation, the authors explained.
At the most basic level, radiologists would look at information and where it’s going (EMR, imaging record, etc.), while deciding whether to include it in their report or not. A more intensive interaction would require radiologists to scrutinize whether data is accurate enough to be used in patient care.
These factors will all be key as healthcare continues to evolve.
“The standards underpinning this managed data flow and the infrastructure to enable it are in their infancy, but it is already clear that this new kind of operational work for radiology practices and radiologists will require additional investment by both practices and vendors,” the authors concluded.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.