How do radiologists remain relevant in an era of machine learning?
February 1, 2017
CHICAGO – With the rise of enterprise viewers and high-powered networks, many doctors are reading diagnostic images on their own, and bypassing the radiologists. Moreover, machine learning is being applied to radiology, and threatens to automate much of what radiologists now do.
So the question many radiology leaders are asking is, How does radiology stay relevant and important as a medical specialty?
“Radiologists were instrumental in the past. How do we ensure this is the case in the future,” asked Dr. Richard Baron, president of the Radiological Society of North America during his address at the annual RSNA conference here in November.
Dr. Baron urged radiologists to have more personal interaction with other physicians, so they can provide insights to their colleagues on a formal and informal basis. They can also learn from their fellow physicians, in this way, too.
Something has been lost, he noted, with the tendency of radiologists to isolate themselves in reading rooms and send off reports via computer networks.
“Technology has removed the need for face-to-face encounters,” said Dr. Baron. “Today, radiologists can work in a vacuum.”
Ideally, radiologists should be working in teams with other doctors, and in close proximity, so there are easy interactions.
He pointed to hospitals at the University of Colorado, the University of Chicago and Johns Hopkins as forward thinkers. “They’ve moved their radiologists nearer to the consulting physicians, to be closer for collaboration.”
As well, radiologists should strive to add more value to the reports they send to referring physicians – helping with the understanding of the patient and the possible treatments. “We need to provide more value, and not just write descriptions of what we see.”
To ensure complete diagnoses and reports, he advised radiologists to make use of the RSNA’s template library.
But he also noted that to answer complex questions, radiologists will have to develop expertise in additional areas, such as genomics. To do this, they must become more proactive, making the effort to constantly learn about new diseases, classifications and therapies.
“Radiologists must again become renaissance physicians,” said Dr. Baron.
He called for continuing innovation, saying that innovation in radiology has slowed recently, and that the position of radiologists as leaders in medical innovation has been challenged by other specialties. He called for more research to be done, especially large-scale, multi-centre trials.
Overall, he said that everything should be considered in the context of patient-centred care. “All decisions should be made on the basis of what is best for the patient – not what is most convenient for the radiologist or what is most lucrative.”
When it comes to patient-centred care, Dr. Baron mentioned a recent Israeli study that found when a radiologist has a picture of the patient in front of him or her, the accuracy of the work being done rises significantly. That’s because the photo is a powerful reminder that the work is being done for a real human being. In a keynote address on machine learning in medicine, Dr. Keith Dreyer, vice chair of radiology at Massachusetts General Hospital, observed that computers are now able to identify photographic images, such as faces, better than people can.
At its booth at the RSNA conference, IBM demonstrated its progress in this area. It showed how Watson could evaluate images, and consider other clinical data such as the personal history of the patient, and lab and medication reports, to make inferences and provide possible diagnoses.
One such system, called Avicenna, provides the diagnosis and shows the clinician how it arrived at the finding. The clinician can then agree or disagree with the diagnosis. In this way, the system becomes a radiologist’s assistant.
Some are asking whether Deep Learning systems of this sort will ultimately replace radiologists. Most expert observers agree that this is unlikely to happen.
“We see the technology as empowering a radiologist to see deeper, see more when solving puzzles,” said Dr. Eldad Elnekave, a radiologist and chief medical officer with Zebra Medical Vision, a Tel Aviv-based company that has produced computer systems that can diagnose CT scans and other images.
Dr. Elnekave says the process of reading images, combining the information with other data and arriving at a diagnosis is more complex than it may appear. “I have been trained by radiologists who have put together seemingly unrelated information to arrive at a correct diagnosis,” he said. “I can’t fathom a computer doing this.”