Artificial intelligence will play an enormous role in the future of radiology, but in 2020, adoption is still limited and confined mostly to academia. Radiology Partners, however, is already harnessing AI in private practice and seeing promising early results.
RP executive Nina Kottler, MD, recently shared some of her company’s lessons learned from adopting these algorithms in regular clinical work. She believes AI in radiology will become increasingly more attractive to private providers, as algorithms bolster physicians’ efficiency and address a broader set of uses.
“Despite the possibilities of this transformative technology, AI is still immature, with technical and clinical barriers to overcome and few implementations in routine practice,” Kottler, RP’s associate chief medical officer of clinical AI, wrote Wednesday in JACR. “However, these barriers are not insurmountable, and lessons can be learned by the early adopters of this technology.”
She boiled the El Segundo, California-based practice’s experiences down to five take-home points. Here’s a quick look:
1) There is no such thing as a seamless, off-the-shelf AI algorithm in radiology. Rather, implementation requires numerous actions, including prepping the data and radiologists.
2) Practices must be able to accurately identify a study at both the procedure and series level. Some imaging vendors offer this option, while others don’t.
“Therefore, it is an important topic to discuss with a potential vendor and to validate this filtering accuracy, because this component … is often absent from receiver operating characteristic curves,” Kottler advised.
3) Programs that bolster radiologists’ efficiency will also increase acceptance. And appointing a physician champion at each practice will help drive physician engagement, “as the greatest influence occurs locally,” Kottler advised.
4) Benefits from AI adoption can stem beyond their intended use. She gave the example of one algorithm Radiology Partners is using to flag imaging exams with critical findings, which has also proven useful in spotting studies with missed findings.
5) Active collaboration is critical to a winning implementation. That includes radiologists working in concert with the AI system, and ongoing interaction between the practice and vendor.