AI in Practice initiative shows potential of advanced computing
The project from RSNA is gaining vendor support and shows a path to supporting radiology practices through vendor collaboration and the use of standards.
Implementing artificial intelligence in healthcare has proved challenging, but an ongoing demonstration effort by the Radiological Society of North America is offering a glimpse at how the advanced computing technology could be applied to imaging.
Called Imaging AI in Practice, the initiative is bringing together vendors, subject matter experts within RSNA and standards organizations to set out practical ways in which AI could support healthcare, focusing on radiology.
The project provided demonstrations at the recent RSNA conference in Chicago, walking visitors through a variety of potential scenarios in which AI could play a role in assisting clinicians, whether in actual image analysis or at a more basic level, such as optimizing the routing of information or in organizing clinicians’ workdays.
Even though it’s just a project at this point, the RSNA initiative is gaining support from a variety of vendor participants, and it places importance on using a variety of standards that enable interoperability and easier information exchange. As such, it may provide a springboard to help AI gain traction and demonstrate its utility for imaging professionals. Expansion of this initiative could offer a path to gain acceptance of AI in use by clinicians, and a focal point to achieve industry consensus around needed interoperability and use of standards.
Developing a growth path for using AI in imaging is important because radiology has long been seen as an important entry point in the diagnostic process for the use of advanced computing. While hopes have been high for the last few years, actual adoption has been slow – as medical professionals need to be won over to the technology, and AI capabilities need to be integrated into workflows and show demonstrable, consistent benefits.
The RSNA initiative, however, aims to builds consensus around a variety of uses, interoperability between vendors support and demonstrate the utility of AI in imaging, said Ali Tejani, MD, a diagnostic radiology resident at the University of Texas Southwestern Medical Center in Dallas, and a clinical champion for the project.
The initiative is growing under the direction of RSNA’s radiology informatics committee, and it offered an expanded presence at the annual conference, featuring an interoperability demonstration to showcase new technologies and communication standards needed to integrate artificial intelligence (AI) into the diagnostic radiology workflow.
The demo featured contributions from 22 vendors, with integration among 32 products; many of those solution suppliers are well-known in the imaging space, and the use cases enabled them to demonstrate new tools and practice enhancements enabled by AI. The five use cases in the RSNA included areas such as employing AI to optimize efficiency in scheduling and other instances, or using AI in the emergency department for findings such as stroke, worrisome nodules or COVID-19.
The project also emphasizes the importance of standards in enabling AI integration into diagnostic imaging, Tejani said. For communications standards, the project builds off of DICOM, the Integrating the Healthcare Enterprise (IHE) standards, FHIR and FHIRcast. From a semantic framework, it features RSNA’s Radiology Lexicon (RadLex), RadElement common data elements and RadReport reporting templates.
“The focus of the project is that we have these innovative tools, but how do we actually bring this to patient care? How can we make them talk to each other. Integrate into workflow?” he said. “We are not actualizing the full potential (from AI tools) that we ought to. To do so, we need to focus on clinical interoperability.”
The demonstration initiative focused on five use cases to demonstrate how AI can assist radiologists, both in doing tasks that are aided by advanced computing and support increased use of evidence-based care, and in supportive uses, such as helping with workload forecasts, scheduling radiologists and optimizing their reads, so as to improve turnaround times for reports, and to be responsive to the complexities of cases and urgency for diagnosis.
Artificial intelligence can assist radiologists in a number of ways, supporting their ability to make well-based diagnostic decisions. For example, the technology can aid in the reconstruction of magnetic resonance (MR) images and it can enhance image resolution, which may enable clinicians to make decisions without exposing patients to excessive levels of radiation from higher-resolution studies.
AI orchestration also can help serve as a “traffic cop” for directing studies to the most appropriate AI tool, and it can serve as a worklist manager to improve efficiency, Tejani noted.
“Patients may have multiple AI algorithms involved in their care,” he said. “Do we have the standards in place to have these tools be able to talk to a radiology information system, an EMR or engage natural language processing?”
The RSNA initiative also aims to gain input from radiology professionals and increase their exposure to and confidence in artificial intelligence. “What is needed is true interaction with the radiologist,” Tejani concluded. “One of the lessons we’ve learned is we need to talk more about how these applications are deployed. It’s not intuitive, and there has to be more education to introduce the technology and increase trust.”
RSNA hopes to continue to build out the project, engage more vendors and offer replicable examples of how to successfully incorporate AI into medical care.