Artificial Intelligence: Friend or Foe to Radiology?
What really is AI and how will it affect the field of radiology?
On Feb. 16, 2011, the concept of AI made history, as IBM's Watson beat "Jeopardy!" champion Ken Jennings in the first-ever man-versus-machine competition in the show's history. Before that moment, AI was the basis of science fiction and Hollywood cinema, not to be taken seriously by the general public. Since that historic upset, AI has steadily marched into the scientific and public consciousness, and now not a week goes by without a reference in a major medical journal or newsstand publication.
But beyond the obvious hype, what really is AI and how will it affect the field of radiology?
As we ponder AI's impact on radiology, it is worth recalling the words of Lee B. Lusted, MD, the founder of the Society for Medical Decision Making, who in 1959, is quoted as saying, "We look to invent an electronic scanner computer to look at chest photofluorograms, to separate the clearly normal chest films from the abnormal chest films." Today's technology aspires to the same goal, with a corresponding result of many predictions of a doomsday scenario for radiologists. Certainly, some experts in health care seem to paint a bleak picture for radiology. Bioethicist and Affordable Care Act architect Ezekiel J. Emanuel, MD, PhD, suggested, "Machine learning will displace much of the work of radiologists."
Recent advances in radiology have demonstrated proven use cases, such as the application of large data sets to train algorithms to correctly identify a child's actual bone age or identifying a diagnosis of tuberculosis with a net accuracy of 96 percent. The application of AI to the field of radiology clearly takes advantage of some of the strengths of deep learning, namely in the areas of pattern recognition, segmentation, quantification, and large-volume throughput, and AI tools are poised to significantly augment the value radiologists can provide for their patients and health systems. Although over time AI may change the way radiologists practice, given the current significant hurdles, it is unlikely that AI will completely displace radiologists or other physicians any time soon.
Despite the quantum leap of technological advances in the field of AI, the development of accurate AI algorithms still relies heavily upon the availability of large volumes of annotated data. The data sets required are often not widely accessible, and privacy and ownership concerns may constrain the utilization of the appropriate data. In addition, the regulatory pathway and associated legal frameworks have yet to be defined clearly. The FDA 510(k) approval process requires demonstrating equivalence to a device previously approved by the FDA, and pre-market approval standards are much stricter for devices that make a final diagnosis. Legally, potential liability will impress caution upon vendors, and health care providers may believe that they put themselves at risk with technology that has been accurately described as a "black box," in which algorithms that have been "learned" are opaque even to the original developer. Finally, clearly the patient and doctor experience will be impacted, and public acceptance is far from guaranteed. As the chief technology officer of Boeing has been quoted as wondering, "Will the public be comfortable flying in a plane with no pilot?"
However, without a doubt, AI and its application to radiology will inevitably march on. Throughout history, innovation has continued to drive radiology's value. From advancements in image visualization to imaging biomarkers and precision medicine, innovative technologies have benefited radiology by improving health care for our patients. AI will certainly be no different.
If nothing else, the private capital markets will continue to reward industry vendors who achieve first-mover advantage by leading the field forward with new technologies. The AI worldwide market is expected to grow from approximately $2-3 billion in 2016 to over $60 billion within a decade. During that same time, medical imaging in AI diagnostics is expected to represent a $19 billion market opportunity by 2025. Hundreds of startup companies are pursuing these opportunities, and dozens of these are focusing on imaging.
But the real financial power comes from pillar corporations such as IBM, Apple, Google, Microsoft, and Facebook, all of which are making large-scale investments in AI applications for imaging. IBM, in particular, has pivoted toward AI applications in health care with a strong interest in imaging. In 2015, IBM purchased Merge Healthcare for $1 billion, acquiring 30 billion images at a stroke. At RSNA 2016, IBM's Watson showcased the future of AI in radiology, demonstrating how a radiologist could review imaging of a patient's aortic dissection and come to a correct diagnosis using a combination of pattern recognition and natural language processing of the electronic health record.
Beyond advances within the private industry, academic and nonprofit institutions have followed suit. Massachusetts General Hospital and Brigham and Women's Hospital in Boston have teamed up to establish the Center for Clinical Data Science. This effort looks to combine large-volume data sets from the two academic medical centers to find opportunities throughout diagnostics, therapeutics, population health, and personal genetics. The goal is to build upon industry-academic partnerships with vendors such as NVIDIA and GE Healthcare to provide clinically useful solutions that have wide impact across society.
In support of such academic and private industry partnerships, the ACR has established a strong AI presence through the ACR Data Science Institute™ (DSI). With the goal of promoting industry standards and transparency, the DSI will focus on developing clinically relevant AI use cases, defining industry standards validation, monitoring algorithm performance, and participating in educational initiatives to guide adoption by the radiology community. A key function of the DSI will be to address regulatory, legal, and ethical issues that accompany the application of AI to medical imaging. By being a trusted partner for government regulators, industry, and patients, the DSI will serve as an honest broker to ensure technological interoperability and an efficient approval process by the FDA. These efforts demonstrate to the radiology community that AI's game-changing technology serves to complement and augment the mission of radiologists to provide value to health systems and patients alike.
In short, AI has made great strides in recent years. The impact on all industries, including health care, will be profound for decades to come. Radiology has the opportunity to undergo a seismic shift in workflow, productivity, and quality by taking advantage of AI's various strengths. Radiologists who evolve to see AI as a partner to their own expertise will prove that adoption of the technology will lead to the betterment of patient care everywhere.
Raymond Liu, MD, FSIR, is an associate radiologist in the Department of Radiology at Massachusetts General Hospital.