Eliot L. Siegel, MD, Professor of Diagnostic Radiology and Nuclear Medicine and Vice Chair of Information Systems at the University of Maryland Medical Center in Baltimore, who has an international reputation of implementing new technologies, discussed the hot topic of artificial intelligence (AI). Dr. Siegel was the first radiology department chairman with the vision, determination, and bravery to fully embrace digital imaging technology, implement the first 100% entirely filmless radiology department in the world.
In 1993, the newly built Baltimore VA (Veterans Affairs) Medical System had the first hospital-wide picture archive and communications system (PACS) in the world, and by doing so, became a cutting-edge imaging informatics laboratory of sorts for decades to come. Dr. Siegel, who still heads its radiology department as well as all other radiology departments of the hospitals in the Veterans Affairs Maryland Healthcare System, is currently working with software developer RadLogics on the development of an AI app that detects COVID-19 on thoracic CT studies.
Like the panel session moderator, Paul Nagy, PhD, the Deputy Director of Johns Hopkins Medicine Technology Innovation Center, Dr. Siegel is both a skeptic of the hyped impact of AI in radiology and an enthusiastic supporter of its future ability to aid radiologists. Dr. Nagy had noted that PACS adoption was expected to proliferate in the 1990’s, whereas this really did not happen for another 10 to 15 years. Dr. Nagy cited the many technology enablers – such as the proliferation of CT scanners, the need to efficiently manage the hundreds of images requiring review in advanced imaging exams, and advancements in digital storage – that stimulated the need for and acceptance of PACS in rank-and-file radiology departments worldwide.
Figuring out what’s wrong with an image is a much more difficult task for a computer vision algorithm than figuring out what it has been trained to look for in the imageEliot Siegel
Dr. Siegel dismisses the idea that AI technology will replace radiologists. Rather, he thinks that AI has the potential within 10 to 13 years to be able to interpret and generate reports for the majority of chest radiographs and mammograms, and identify the exams that need more attention and radiologist review.
He pointed out that diagnostic AI apps are created from extensive training on image databases, but questioned if the databases are representative of the global population and the myriad variations of diseases and abnormalities that radiologists encounter. “How does one verify that image databases are comprehensive enough? How can the accuracy of an AI app be validated to the extent that it is accurate in any situation? What is the gold standard for the performance of an AI app and who defines this? Also, the value of an AI diagnostic/interpretive app will inevitably be diminished by new advances in imaging technologists, such as the way that apps developed for conventional 2D mammography have become made obsolete by mammography tomosynthesis (3D mammography).”
Dr. Siegel also pointed out that AI apps have inherent limitations, no matter how good they are. “Figuring out what’s wrong with an image is a much more difficult task for a computer vision algorithm than figuring out what it has been trained to look for in the image,” he said. “A radiologist evaluates the entire image in a clinical context.”
“One of the biggest challenges with AI software is that they perform impressively in research manuscripts, demonstrate a transparent and logical and exemplary performance for FDA clearance, but then perform substantially less well in actual clinical practice,” Dr. Siegel says. “This is related to the fact that different radiology imaging equipment and different patient populations result in differences in the images from the initial training and validation sets that are not distracting for humans but can change diagnostic accuracy substantially. Radiologists ‘recalibrate’ in new situations, but AI software does not have this ability right now due to the FDA clearance limitations. If the FDA would allow local/regional modifications to help allow the AI software to fine tune itself over time, this would be a very significant advance in AI technology that could potentially accelerate adoption.”
The Artificial Intelligence (AI) landscape confronting the radiographer profession will be outlined in sessions at ECR 2020, with leading practitioners urging the need for an evidence-based approach in order to deliver a safe and effective service for patients. The session, under the broad heading of “Artificial intelligence and the radiographer profession”, aims to discuss AI within the…
Dr. Siegel predicts that the next AI “killer app” will be the use of Deep Learning by modality vendors for image acquisition improvements, such as to improve contrast and spatial resolution, or replace or enhance iterative reconstruction. This type of app could also offer the potential for reductions in overall image reconstruction time, radiation dose, and/or the amount of intravenous contrast required. Dr. Siegel believes AI apps will eventually significantly contribute to improvements in efficiency and productivity for radiologists. The development of AI clinical “suites” of applications that focus on a particular disease or area of the body could combine the finds of an AI diagnostic app with clinical databases that suggest specific diagnoses, recommend additional studies, and suggest different treatment possibilities.
Combinations of applications that work cooperatively on a single suite could also be of great benefit to radiologists. Dr. Siegel suggested, as a hypothetical example, a renal disease suite could combine AI applications that segment the kidneys, search for renal masses, evaluate dynamic contrast enhancement to evaluate function over time, measure renal size, and characterize renal cysts. Other AI apps would then correlate the findings with a patient’s clinical status such as hypertension, renal function, urinary protein levels, and medication history to create diagnostic and therapeutic recommendations.
Image sharing and interoperability across healthcare enterprises
It’s been a long-held belief that the impact and omnipresence of cloud technology will render the use of DICOM CD/DVDs obsolete as the primary means of radiology exam transfer among unrelated healthcare entities. Don Dennison, a consultant and specialist in image interoperability, disagrees. “I’m going to predict that the CD will die a death of 1,000 stabs and take a very long time before being phased out because it is so easy to export images onto this still universally used media.”
DICOM CD/DVDs offer the advantage of being inexpensive, portable, and having a standard format for viewing. They can be easily retained by patients and used throughout the world. Disadvantages are that password protection or lack thereof can be problematic and they may not contain the associated radiology reports. Cloud-based imaging data exchange within healthcare systems or by accessing a patient or organization administered image repository offers secure, reliable image transfer. This well-established technology is flourishing within individual healthcare systems as health information exchanges (HIE), and in countries with single-payer, publicly-funded health models.
“The regional Diagnostic Imaging Repositories (DIRs) in Canada are an excellent example of the latter,” said Mr. Dennison. “It is in the interest of all patients and taxpayers to reduce the waste and risk associated with lack of image access by funding a solution. However, automated sharing among most DIRs is still extremely limited.”
I predict that we are going to see an emergence of reliable methods to share imaging data across imaging service providersDon Dennison
Mr. Dennison explained that within the United States, viable business models to support image transfer interoperability between or among healthcare systems’ HIEs with respect to costs, funding, privacy and system security issues currently do not exist. Additionally, issues relating efficiently adding transferred exams and reports to recipient PACS and radiology information systems (RIS) are substantial. New accession numbers and some unique identifiers are needed, procedure type descriptions may be different and need to be reconciled, and DICOM attributes in images may contain values that the recipient does not want to include in their records and series descriptions may need to be modified to match the recipient’s norm.
Per-use commercial image transfer systems and patient-controlled image repositories have existed for nearly a decade in the U.S., but they haven’t proliferated as expected, attributable in great part due to funding issues. “Clinicians want immediate access to patients’ images. Patients agree. But only when healthcare reimbursement models give priority to utilizing existing and relevant image data from prior exams to improve healthcare and reduce costs, and agree to fund access fees could cloud-based image transfer exchanges proliferate. I predict that this will happen,” he said. “I also predict that we are going to see an emergence of reliable methods to share imaging data across imaging service providers such as independent imaging centers and regional HIEs.”
Mr. Dennison predicts AI tools could automate reconciliation of externally acquired image data, such as identifying the patient’s medical record number, creating an order with local accession number and study instance unique identifier, correlating and renaming the exam type, technique, and procedure, and cleaning up or purging unneeded metadata from the originating institution.
Professor Elizabeth A. Krupinski, Ph.D., Vice Chair for Research in the Department of Radiology and Imaging Sciences of Emory University in Atlanta, Georgia, is an expert on the performance, productivity, and efficiency issues relating to diagnostic workstations and their use by radiologists. Her research has been highly influential in improving workstation ergonomics and optimizing diagnostic image and data displays. Diagnostic workstations and workspaces have incrementally evolved over the past 30 years, and Dr. Krupinski doesn’t expect dramatic changes in the near term in some aspects. But she does believe that there will be continuing change to improve computer interfaces to make things easier with respect to manipulating image display and command functionality.
Some things are classic and have been around since the early days of computers with respect to form and functionality, and Dr. Krupinski believes that the computer mouse is one of these. “Radiologists are used to using the mouse, whether simple or complex in design. It is inexpensive and a very ergonomic tool, making interacting with computers possibly better than with some more high tech devices. I predict that the mouse will be here, perhaps forever, because it might be the best tool for computer interaction,” she said. “But it’s a habit for me,” she conceded, saying that she wasn’t motivated to change. She commented that new generations of radiologists may have a different opinion because they grew up from childhood interacting with other devices like smartphones and tablets using swiping and tapping commands, for example.
“Manufacturers will spearhead these changes, based on market demand and competitive factors, as they always have,” Dr. Krupinski said. “Either smartphones and tablet devices will need to evolve to the point where they are practical for routine daily image interpretation or workstation interfaces will need to evolve to be more like smartphones and tablets. “I don’t think that this will happen any time soon, because there does not seem to be the financial motivation to do so.”
In the film-based era, radiologists and physicians who ordered exams congregated in radiology departments to discuss them. Dr. Nagy believes that the impact of COVID-19 and the accelerated use of telemedicine presents an opportunity. Physicians using today’s teleconferencing-enabling technologies in both their personal and professional life may begin to value electronic face-to-face collaboration with radiologists, he predicted.