Big data is the primary driver of machine learning and artificial intelligence, and medical imaging has important roles to play in the short, medium, and long-term evolution of this technology for health care. In the short term, leveraging deep operational data to develop Al models for operations management can increase workflow efficiency and improve access to imaging care. In the medium term, capitalizing on the bit depth and volume of image data will enable further development of diagnostic assistants using machine learning and artificial intelligence. In the long term, continuous learning Al will become the standard by which Al models adapt to changing data environments, to increase their robustness and mitigate bias inherent in their training data.
Imaging operations follow a largely linear workflow highlighted by timestamps that demarcate specific transitions. These can be used to train models that predict operational events such as the time patients may expect to wait for their imaging exams. In addition, such models may inform how imaging affects broad measures of health care performance such as length of stay in the emergency or inpatient settings. Such models can inform process improvements intended to streamline healthcare for greater operational efficiency.
As a diagnostic assistant, Al can be used for both findings detection and decision support. Although radiologists train for many years to detect subtle findings in medical images, human observation is an inherently fallible process. Computer systems trained to detect subtle findings are welcomed by most radiologists and can reduce the risk of medical error. The depth of knowledge required to master all facets of medical care exceeds the capacity of any one individual, and computer-assisted reporting with decision support is one means of enabling radiologists to reduce the encyclopedic knowledge required to practice unassisted. Machine learning and artificial intelligence have important roles to play in both findings-detection and decision support for computer assisted reporting of medical images.
Major challenges with Al in healthcare include limited reproducibility and sustainability, particularly as Al algorithms age over time. Naturally occurring changes in data environments may occur owing to changes in both hardware and software associated with imaging devices as well as other demographic changes in patient cohorts. Such changes often lead brittle Al algorithms to break over time. The promise of continuous learning Al is envisioned to overcome these barriers by enabling algorithms to “learn-to-learn” from feature discovery and frequent retraining on updated data sets that reflect the current local data environment.
Informed by big data in healthcare, machine learning and artificial intelligence promise to revolutionize the way in which health care is delivered globally. Medical imaging has a critical role to play in leading Al for operations management, diagnostic assistance, and continually improving with continuous learning Al.
James A. Brink is Radiologist-in-Chief at the Massachusetts General Hospital (MGH) and the Juan M. Taveras Professor of Radiology at Harvard Medical School. He earned a BS degree in Electrical Engineering at Purdue University and an MD at Indiana University before completing his residency and fellowship at MGH in 1990. He joined the faculty at the Mallinckrodt Institute of Radiology at Washington University School of Medicine where he rose to the rank of Associate Professor prior to joining the faculty at Yale University in 1997. Dr. Brink served as Chair of the Yale Department of Diagnostic Radiology from 2006 to 2013 prior to returning to MGH as Radiologist-in-Chief. Dr. Brink is Fellow of the Society for Computed Body Tomography/Magnetic Resonance, Fellow of the American College of Radiology, Past-President of the American Roentgen Ray Society, and Past-President of the American College of Radiology. For the National Academies of Sciences, Engineering and Medicine, Dr. Brink serves as Vice-Chair of the Nuclear and Radiation Studies Board. For the International Society for Strategic Studies in Radiology, Dr. Brink serves as President-Elect. Dr. Brink is a recipient of the American Roentgen Ray Society’s Gold Medal and an honorary member of the European Society of Radiology, the Japanese Radiological Society, the Italian Society of Medical Radiology, the American Association of Physicists in Medicine, and the International Organization for Medical Physics. In 2019, Dr. Brink was awarded ‘Outstanding Electrical and Computer Engineer’ from Purdue University, and in 2020, he was elected ‘Distinguished Emeritus Member’ by the National Council on Radiation Protection and Measurements. Dr. Brink has broad experience in medical imaging, including the utilization and management of imaging resources, with specific interest and expertise in issues related to the monitoring and control of medical radiation exposure.
MIC Edward J Hoffman Medical Imaging Scientists Award & Lecture Ceremony