While the DSI AI use cases offer approachable examples of how AI could improve radiology practices, there is even more to them than that. Use cases are also an open invitation to those developing AI to create a new AI model — or improve on an existing one.
Use cases help AI developers
Because radiology experts are involved in all stages of use case development, AI developers can get a sense for popular AI projects among radiologists by looking for the clinical areas with the most use cases. Recently, Classifying Suspicious Breast Calcification, Acute Appendicitis, and Pneumothorax have topped the list for most searched use cases. Panel chairs are currently developing methods to prioritize high-value use cases to ensure they are developed first.
Use cases can also inspire vendors planning future developments — or share insights for tweaking current projects with additional clinical details and implementation requirement. This approach helps ensure that models nest seamlessly in the radiologist workflow and outputs contribute to clinical actions.
Since clarity and consistency are important, use cases are tagged with agreed upon standards – Common Data Elements (CDEs) – developed by ACR, RSNA, and subspecialty radiology organizations. These standards help promote crossover of algorithms among facilities, while supporting the developer path towards clinical implementation.
Use cases help radiologists
For the radiologist, DSI is continuing to expand the range of solutions where AI can offer support. There are thousands of potential use cases for narrow AI tasks in each domain of radiology.
Developing use cases strengthens the leadership role radiologists have begun playing in AI development, while ensuring that AI implementations fit well into the clinician workflow. Our use cases are also the foundation of ACR tools developed to help make AI more accessible to the radiologist community. For instance, use cases set standards for dataset annotation for ML projects in ACR AI-LAB™.
Non-interpretive use cases
New to the 2019 release, the DSI formed the non-interpretative use case panel, which was chaired by Alexander J. Towbin, MD, and Adam B. Prater, MD, MPH. The panel developed use cases to add value to the care stream beyond image interpretation. Since there are many areas from which to choose, panel experts first identify areas with the greatest value to radiology, then define specific ways for AI to solve problems within them. A model to predict patient no-shows and alert patients on expected wait times based on the availability of imaging equipment is one example of a new use case developed by this panel.
Some institutions have already begun developing models like these to increase efficiency and improve the patient experience. The DSI non-interpretive use cases set standards for inputs and outputs. They also have the potential to accelerate the use of AI at institutions that aren’t currently using it.
Collaborations with other societies
Several use cases born from collaborations across specialty societies are part of the latest release. The DSI partnered with the Society of Thoracic Radiology to release use cases on incidental pulmonary nodules and supported Kaggle challenges aligned to Define-AI use cases. DSI also supported an AI use case challenge with the Society of Abdominal Radiology to accelerate the pace of use case generation and demonstrate how individual radiology organizations can strike up projects aligned to their interests with the use case framework. We will expand our work with other specialty organizations, including the Society of Skeletal Radiology, to promote openness in AI development for radiology in coming months.
Adding to AI use cases
DSI will regularly publish new use cases after a two-month public comment period. This continual stream of new use cases will feed the AI ecosystem and help to sustain the growth of new solutions to improve medical care.
The DSI encourages the radiology community to get involved by commenting on use case proposals and submitting new use case ideas. Openings are also available on use case panels for ACR members who would like to join. Visit the DSI website to discover ways to participate.
Bibb Allen, Jr., MD, FACR | ACR DSI Chief Medical Officer | Diagnostic Radiologist, Grandview Medical Center, Birmingham, AL | April 04, 2019 (with thanks to Jordan Meyer of the ACR’s Data Science Institute)
As radiologists, we strive to deliver high-quality images for interpretation while maintaining patient safety, and to deliver accurate, concise reports that will inform patient care. We have improved image quality with advances in technology and attention to optimizing protocols. We have made a stronger commitment to patient safety, comfort, and satisfaction with research, communication, and education about contrast and radiation issues. But when it comes to radiology reports, little has changed over the past century.