As artificial intelligence (AI) continues to be developed and incorporated into healthcare, radiologists should consider how the technology's promise is being presented to the public -- and manage their expectations about when it will be standard of care, according to an opinion published June 17 in the Journal of the American College of Radiology.
The need to set this context for patients is driven in part by how the media communicates about the technology, wrote a team led by Edmund Weisburg of Johns Hopkins Medicine in Baltimore.
"As the medical community incorporates deep learning and AI into some specialties, particularly emergency radiology, it may be prudent to ponder if we are on the cusp of unrealistic public expectations regarding the use of AI in routine radiologic diagnosis," the researchers wrote.
So how close is the use of AI as standard practice in radiology? Weisburg's group cited a study that found no significant difference between radiologists and AI in identifying pleural effusions, concluding that "AI demonstrates great potential for assisting radiologists in evaluating supine radiograph results and decreasing the volume of missed findings."
The team also highlighted a 2020 survey of chest radiologists and computer scientists that found both groups to be bullish about AI's influence in radiology (although about 15% of the computer scientists thought that would also mean that radiologists will be obsolete in a few decades).
How do patients feel about AI and radiology? Weisburg and colleagues cited another study that found patients to be more comfortable with a radiologist first reader and an AI second reader, rather than a standalone AI interpretation. And results from research that consisted of a questionnaire given to people in a fracture clinic showed that 95.4% of patients preferred a clinician to resolve unclear reading results rather than AI.
In any case, using AI for diagnosis of disease is far from the standard of care, and patients may not understand that.
"[We] are likely on the verge of having patients routinely ask if a radiologist will be using AI to help interpret their study," the team concluded. "This understandable desire to ensure the optimal diagnostic reading should propel us to help accelerate the adoption of AI into standard practice as long as the results are scientifically sound, and not just the beat of the latest fad. For now, we can tell patients that AI is already being used in several stages of the patient experience, is in increasing use in the emergency setting of select institutions, and although it has great potential in diagnosis, it is not in general use at this time but perhaps in the near future."
