Comparative analysis of AI support levels in clinical interpretation of traumatic pelvic radiographs

Abstract Plain pelvic radiographs (PXR) remain crucial for initial trauma assessment, yet interpretation challenges persist. While artificial intelligence (AI) shows promise, its practical impact across specialties remains unexplored. We conducted a retrospective image-based, multi-reader multi-case...

Full description

Saved in:
Bibliographic Details
Main Authors: Yu-San Tee, Jen-Fu Huang, Yu-Ting Huang, Chi-Po Hsu, Huan-Wu Chen, Chi-Hsun Hsieh, Chih-Yuan Fu, Chi-Tung Cheng, Chien-Hung Liao
Format: Article
Language:English
Published: Nature Portfolio 2025-08-01
Series:npj Digital Medicine
Online Access:https://doi.org/10.1038/s41746-025-01923-5
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Plain pelvic radiographs (PXR) remain crucial for initial trauma assessment, yet interpretation challenges persist. While artificial intelligence (AI) shows promise, its practical impact across specialties remains unexplored. We conducted a retrospective image-based, multi-reader multi-case (MRMC) study using a standardized, prospectively planned evaluation protocol. A total of 26 physicians (8 radiologists, 10 emergency physicians, 8 trauma surgeons) interpreted 150 PXRs in three sequential sessions: without AI, with AI-alert, and with AI-visual guidance. AI assistance improved overall diagnostic accuracy from 0.870 to 0.940 (p < 0.001) and reduced interpretation time from 22.70 to 9.58 s (p < 0.001). Non-radiologists showed substantial improvements, with emergency physicians demonstrating increases in specificity (26.2%, p = 0.006) and positive predictive value (41.5%, p = 0.006). Trauma surgeons with AI-visual guidance achieved comparable accuracy to unaided radiologists (0.940 vs. 0.920, p = 0.556). Tailored AI assistance effectively bridges the performance gap between radiologists and non-radiologists while reducing reading time. These findings suggest AI integration could enhance clinical workflow efficiency across specialties in trauma care settings.
ISSN:2398-6352