Evolution of Machine Learning Will Strengthen Radiology
Monday, Nov. 28, 2016
The rapid advance of artificial intelligence (AI) will not render the radiologist obsolete; rather, it will augment the profession, providing radiology with an opportunity to lead the way in precision medicine, said a leading expert during one of yesterday's plenary lectures.
The pace of advances in AI has accelerated in recent years, said Keith J. Dreyer, DO, PhD, vice chairman of Radiology Computer and Information Sciences at Massachusetts General Hospital (MGH) in Boston. In his lecture, "When Machines Think: Radiology's Next Frontier," Dr. Dreyer attributed the recent advances to the development of artificial neural networks that help computers learn in a similar fashion to humans—a process known as deep learning. This development enabled a revolution in AI, exemplified by the ImageNet Large Scale Visual Recognition Challenge Competition, an annual event where computers once lagged behind humans in visual recognition accuracy. After developers switched to deep learning algorithms in 2013, computers quickly began outperforming humans.
With AI improving at a much faster rate than human intelligence, Dr. Dreyer recalled the words of Ray Kurzweil, a leading expert on AI who famously predicted that technological singularity, or the point when machine intelligence surpasses that of all humans combined, was on the horizon.
"This point of singularity could happen in about 2029, just as Kurzweil predicted," Dr. Dreyer said.
But even as he detailed the rapid evolution of machine learning, Dr. Dreyer offered reassurance for radiologists fearing obsolescence. He recalled how, when IBM's chess-playing computer Big Blue beat world champion Garry Kasparov in 1997, Kasparov noted that humans had made the machine that defeated him. After the defeat, Kasparov incorporated the computer's analytical, unemotional approach into his game—an approach he named after the centaur, a creature from Greek mythology with the head, arms and torso of a man and the body and legs of a horse. Dr. Dreyer likened Kasparov's inspiration to the future of radiology, where humans work closely with AI-powered machines to optimize patient care.
"Radiologists will be the centaur diagnosticians, allowing machines to make us smarter, help us do more and give us more value," he said.
Clinical Data Science Critical to Radiology's Evolution
The impact of computer learning will be most apparent at the crossroads of radiology and the emerging field of clinical data science, which encompasses the collection, transformation and analysis of clinical data. Earlier this year, Dr. Dreyer helped open the new MGH Clinical Data Science Center — part of a new approach to diagnosing and treating disease that uses cognitive computational algorithms such as ML and artificial neural networks to, in effect, call upon the shared expertise of hundreds of radiologists when reviewing a patient scan.
"There is a tremendous amount of applications for A.I. in radiology," he said. "The radiology field itself is going to be the foundation of precision healthcare."
For instance, once computers are trained to analyze solid lung nodules, images could be sent to a secure cloud and evaluated according to Lung Imaging Reporting and Data System (Lung-RADS) guidelines. With more than 9 million people eligible for lung cancer screening in the United States alone, pulmonary nodules represent an enormous potential application for AI—and that represents only a small fraction of radiological findings.
"Soon we will be able to create a precision radiology report for all body parts and all examinations," Dr. Dreyer said.
Dr. Dreyer advised radiologists to ask for A.I. that not only automates but also augments what they do. Vendors, he said, should provide AI that improves reality, making a single procedure deliver even more value.
"We should use AI to expand our diagnostic and clinical roles, serving as our patients' trusted advisor," he said.