Back to the list
Congress: ECR25
Poster Number: C-16080
Type: Poster: EPOS Radiologist (scientific)
Authorblock: B. J. Van Der Zwart1, H. C. Ruitenbeek1, F. J. Bruun2, A. Lenskjold2, M. Boesen2, K. Ziegeler3, K-G. A. Hermann3, E. Oei1, J. J. Visser1; 1Rotterdam/NL, 2Copenhagen/DK, 3Berlin/DE
Disclosures:
Bastiaan Johannes Van Der Zwart: Nothing to disclose
Huibert C Ruitenbeek: Nothing to disclose
Frederik J. Bruun: Nothing to disclose
Anders Lenskjold: Nothing to disclose
Mikael Boesen: Advisory Board: Radiobotics
Katharina Ziegeler: Nothing to disclose
Kay-Geert A. Hermann: Nothing to disclose
Edwin Oei: Nothing to disclose
Jacob Johannes Visser: Nothing to disclose
Keywords: Artificial Intelligence, Musculoskeletal bone, Conventional radiography, Comparative studies, Computer Applications-Detection, diagnosis, Trauma
Purpose Objective: This study aimed to evaluate the performance and generalizability of an AI tool for fracture detection across three European hospitals using consecutive clinical cases to simulate real-world conditions.
Read more Methods and materials Background: Artificial intelligence (AI) models for fracture detection have shown promise in improving diagnostic efficiency and accuracy in radiology workflows. However, the generalizability of these models across different healthcare systems remains unknown as this could differ due to regional variations in imaging protocols and clinical workflows.Data collectionData was collected retrospectively at the three university hospitals: Copenhagen University Hospital Bispebjerg and Frederiksberg, Denmark (BFH), Charité Universitätsmedizin in Berlin, Germany (CUB), and Erasmus Medical Center in Rotterdam, The Netherlands (EMC). Each institution...
Read more Results Baseline characteristics: Data was collected consecutively from April 30th 2018 through June 30th 2018 at BFH, from April 30th 2021 through June 25th 2021 at CUB and from April 30th, 2022 through July 4th, 2022 at EMC according to the eligibility criteria until a total of 500 cases was reached for each hospital. No cases were excluded based on the exclusion criteria.   Mean age was 43.3 ± 19.5 years and consisted of 821 male and 679 female patients. The demographics...
Read more Conclusion Key results:  The AI tool demonstrated consistent performance across three European centers, with AUC values of 0.90, 0.91, and 0.95, showing robust generalizability. Subgroup analysis revealed significant underperformance in wrist/hand/finger fracture detection in one center, highlighting the need for local validation and possibly adjustments in anatomy-specific regions. The AI tool shows potential as a triage tool to streamline radiology workflows, especially in high-demand environments like emergency departments Summary statement: The AI tool showed robust and generalizable performance across multiple centers, but anatomy-specific adjustments are...
Read more References   Delong, E.R., D.M. Delong, and D.I. Clarkepearson, Comparing the Areas under 2 or More Correlated Receiver Operating Characteristic Curves - a Nonparametric Approach. Biometrics, 1988. 44(3): p. 837-845.
Read more
GALLERY