Tuesday, November 28 | 10:10 a.m.-10:20 a.m. | T3-SSMK06-5 | Room E351
Deep learning-based computer-aided detection (CAD) software shows potential for helping to avoid missed fractures on radiographs in the emergency department (ED), according to this talk.
Seeking to add evidence that computer-assisted fracture detection is accurate in clinical situations, a research team led by Huid Ruitenbeek of Erasmus Medical Center in Rotterdam, the Netherlands, developed a multitask deep neural network (MDNN). The MDNN was trained on 450,000 x-rays and tested on 150,000 x-rays for the purpose of detecting x-rays with and without fractures.
The algorithm was then tested on 5,489 radiographs, comprised of 1,877 studies with fractures and 3,612 exams without fractures, along with their respective clinical radiology reports, acquired from January 2019 to November 2022. The MDNN yielded an overall area under the curve of 0.95.
"The study showed that using CAD as an assistance tool can benefit high-workload healthcare facilities, especially in busy ERs, where patients might have different orthopedic conditions," Ruitenbeek et al concluded. "The lack of trained radiologists in ER and the increasing number of fractures require reliable and accurate CAD for automated detection and reduce missed fractures."
He noted that the results demonstrate the generalizability of the CAD algorithm in real-world clinical settings.
What were their other findings? Attend Ruitenbeek's Tuesday morning session to get all of the details.