By Lynn Antonopoulos
The technology, in use in a real-world clinical setting in four Norwegian hospitals, flags urgent cases, allowing radiologists to prioritize and review them more quickly, which can lead to faster diagnosis and treatment for patients in need.
“As the first hospital trust in Norway implementing an AI application in radiology in clinical use, there was a need to document the evidence of the outcomes,” said Line Tveiten, BMedSc, an implementation leader at Vestre Viken HT in Norway. “As the AI application was not implemented as a study, the effects were measured retrospectively.”
According to Tveiten, prior to the implementation, each hospital in the system had its own unique process for patient management.
“We focused on the variations in workflow across hospitals. This required significant dedication from professionals to identify all details in order to avoid additional workflow adjustments,” Tveiten said.
Once implemented, the AI system helped triaging cases by flagging results in the institution’s Radiology Information System. Patients could therefore be directed based on the AI result and the radiographer’s assessment. Data was collected on over 35,000 patient X-rays to track changes in workflow, patient wait time, reductions of consultations and prioritization.
With the AI tool, radiographers discharged more than 8,500 patients who had no fractures, reducing the total patient wait time by 250 days, Tveiten noted. Consultations dropped by more than 6,000, making it easier for health care workers to prioritize sick patients.
Tveiten highlighted the importance of change management and user adaptation in the implementation.
“Workflow changes have the possibility to promote standardization across hospitals giving patients more equal care,” she said.
While the AI tool helped reduce crowding in the emergency department as expected, the AI tool is not used autonomously at her institution and radiologists must still sign off on examinations as schedules allowed.
“We were initially expecting a better beneficial impact in the reading time for the radiologists,” Tveiten said. “But the experience shows no, or just small reductions in the total reading time.”
Their validation showed that the solution performed close to the radiologist in certain anatomic areas, though less effectively in others. “This demonstrates the potential for autonomous use of the solution in select examinations,” she said.
Tveiten emphasized that implementing an AI application in clinical care can have unforeseen impacts on other departments.
“It is especially important that referrers understand what the solution is used for and how AI impacts patient flow,” she said. “If injuries other than fractures are suspected, they must be managed with a consultation at the emergency department, in order to ensure patient safety.”
Two RSNA 2024 poster presentations covering the external validation and sequential adoption of the AI application used at Vestre Viken HT offer a detailed look at the implementation.
“Our experience shows that there is no one-size-fits -all solution for adopting AI, as each hospital operates with a unique workflow. It is essential to tailor the implementation approach to maximize the benefits within the specific context of each hospital,” Tveiten said.
Access the presentation, “The Beneficial Impact of Implementing and AI Application for Fracture Detection in Clinical Routine at a Norwegian Hospital Trust,” (M6-SSMK03-1) and posters, “Benchmarkable External Validation for Sequential Adoption of an AI Solution” (M5A-SPMK-4) and “Strategic Implementation of an AI Fracture Detection Algorithm Sequentially,” (M5B-QI-2) on demand at RSNA.org/MeetingCentral.
© 2024 RSNA.
The RSNA 2024 Daily Bulletin is the official publication of the 110th Scientific Assembly and Annual Meeting of the Radiological Society of North America. Published online Sunday, December 1 — Friday, December 6.
The RSNA 2024 Daily Bulletin is owned and published by the Radiological Society of North America, Inc., 820 Jorie Blvd., Suite 200, Oak Brook, IL 60523.
