Sunday, December 1 | 2:30 p.m.-2:40 p.m. | S5-SSGI04-1 | E451B
Emergency radiologists may be interested in learning about a machine-learning algorithm developed to focus specifically on detecting life-threatening splenic injuries.
A team of researchers from Canada retrained an open-source machine-learning model that had been initially used for the RSNA 2023 Abdominal Trauma Detection AI Challenge. After modification and retraining, the algorithm was able to classify splenic injuries and detect those that are high-grade.
Next, they performed external validation on a dataset of 1,216 split-bolus protocol, trauma CT scans performed between January 1, 2005, and July 31, 2021, on adult patients at a level I trauma center. Ground-truth labels were established by five abdominal radiologists using the American Association for the Surgery of Trauma splenic injury scoring scale.
Describing the machine-learning model performance as "robust," the team reported that it demonstrated an area under the curve (AUC) of 0.959, sensitivity of 94.2%, specificity of 85.7%, positive predictive value of 45.9%, and negative predictive value of 99.1%.
The model might enhance patient care by prioritizing radiologist worklists and facilitating early trauma intervention. Overall, researchers said performance metrics either met or surpassed those reported in existing literature, in both the three-class splenic injury classification and detection of high-grade injuries. Don't miss out on the rest of the findings at this Sunday afternoon presentation.